* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download 4. Two sample classes encoded: motion verbs and `know verbs`
Lithuanian grammar wikipedia , lookup
Malay grammar wikipedia , lookup
English clause syntax wikipedia , lookup
Proto-Indo-European verbs wikipedia , lookup
Modern Greek grammar wikipedia , lookup
Ojibwe grammar wikipedia , lookup
Udmurt grammar wikipedia , lookup
Chinese grammar wikipedia , lookup
Macedonian grammar wikipedia , lookup
Old Norse morphology wikipedia , lookup
French grammar wikipedia , lookup
Kannada grammar wikipedia , lookup
Scottish Gaelic grammar wikipedia , lookup
Ukrainian grammar wikipedia , lookup
Japanese grammar wikipedia , lookup
Old Irish grammar wikipedia , lookup
Portuguese grammar wikipedia , lookup
Navajo grammar wikipedia , lookup
Germanic weak verb wikipedia , lookup
Germanic strong verb wikipedia , lookup
Polish grammar wikipedia , lookup
Modern Hebrew grammar wikipedia , lookup
Swedish grammar wikipedia , lookup
Russian grammar wikipedia , lookup
Sotho verbs wikipedia , lookup
Turkish grammar wikipedia , lookup
Latin syntax wikipedia , lookup
Georgian grammar wikipedia , lookup
Spanish grammar wikipedia , lookup
Ancient Greek grammar wikipedia , lookup
Lexical semantics wikipedia , lookup
Old English grammar wikipedia , lookup
Hungarian verbs wikipedia , lookup
German verbs wikipedia , lookup
Yiddish grammar wikipedia , lookup
Kagoshima verb conjugations wikipedia , lookup
Definition of the links and subsets for verbs Final Version 6 November, 7, 1996 Antonietta Alonge* Istituto di Linguistica - Università di Perugia Istituto di Linguistica Computazionale del C.N.R., Pisa Deliverable D006, WP4.1, EuroWordNet, LE2-4003 * This document compiles work from: the AMS group, in relation to Dutch; the PSA group, in relation to Italian; the FUE group, in relation to Spanish. People from all the groups have somehow contributed to it, but, in particular, Piek Vossen, Nicoletta Calzolari, and Simonetta Montemagni have provided significant comments and contributions to its content. Identification number LE-4003-D-006 Type Document Title Definition of the links and subsets for verbs Status Final Deliverable D006 Work Package WP4 Task T4.1 Period covered June-August1996 Date 07/11/96 Version 6 Number of pages 188 Authors Antonietta Alonge WP/Task responsible Project contact point EC project officer Status PSA Piek Vossen Computer Centrum Letteren University of Amsterdam Spuistraat 134 1012 VB Amsterdam The Netherlands tel. +31 20 525 4624 fax. +31 20 525 4429 e-mail: [email protected] Jose Soler Public Actual distribution Project Consortium Suplementary notes n.a. Key words Semantic databases, Information Retrieval, Language Engineering, Semantic Relations, Semantic Test Sentences Abstract This deliverable reports on the work carried out within Task 4.1, assigned to WP4. The objectives of this task were i) the definition of the relations to be encoded for verbs; ii) the identification of criteria for verification of such relations for each language; iii) the definition of the subsets of verbs for which relations will be encoded in the database. The document also reports on the experiments carried out ons specific semantic domains in order to foresee the main issues and problems to be dealt with when building the whole EuroWordNet database for verbs. Complete Status of the abstract Received on Recipient’s catalogue number Executive Summary The goal of this deliverable is to report on the work carried out within Task 4.1, assigned to WP4. The objectives of this task were i) the definition of the links for verbs; ii) the identification of criteria for verification of such links for each language; iii) the definition of the subsets of verbs for which relations will be encoded in the database. Of course, the results obtained so far with respect to these three issues should be considered, within reasonable limits, to be 'preliminary', since further work on data might produce some (minor) changes either as far as the links, and therefore the criteria to establish them, or the verbs selected are concerned. The document is organised as follows. Section 1. provides a brief introduction to the general goals of the project and to the specific goals of this Task. In section 2. we provide discussion and definitions both of the basic relations, already encoded with respect to verbs in WN 1.5 and also used to develop our resource, and of the new relations identified in order to code all the semantic information which we may acquire from MRDs, but also from other sources of lexical data, and which can be useful for both IRSs or other LE applications. As far as the basic relations already used in WN 1.5 are concerned, we have provided the possibility to add labels to them indicating disjunction, conjunction, factitivity and negation. On the other hand, the new relations added, linking verbs to words in other lexical classes, allow the encoding of more subtle semantic distinctions, thus giving us the possibility both to encode the kind of data which are typically found in MRDs (which are our main source of data), and to meet the user-requirements. However, since these 'new' relations require more complicated procedures in order to be acquired, for the time being their values will be coded only in relation to those subsets of verbs for which we have enough data. Indeed, we aimed at designing a 'model' general enough to incorporate the different types of semantic relations that are both extractable from dictionaries (and other sources) and of some usage for LE applications. The definition of such a broad model does not imply that we fill the whole of it. In fact, we will encode all the relations already indicated as being fundamental for our database within the TA, while the others will be filled more as an exemplification of their statute, their extractability, etc. In section 2., clear criteria for the identification of each relation are also provided, both as general criteria which define a relation, and as language-specific criteria which allow the identification of the relations for each language. Although we briefly refer to extraction issues, the extraction procedures for each wordnet will be described in detail in the deliverables D021 and D025. In section 3. we discuss issues connected with the definition of the subsets of verbs which will be encoded in our database, explaining in particular the methodology followed in order to identify a first subset of common Base Concepts. In section 4. we discuss the first results obtained by trying to code data on the relations chosen with respect to two 'sample' classes of verbs: motion and know verbs. Finally, the Appendix provides: i) the specific tests elaborated for each language in order to identify relations; ii) detailed data related to the Base Concepts identified so far for each language; iii) detailed data on the results of the 'experiments' on coding information on relations for both motion and know verbs in each language. Table of contents 1 Introduction ...........................................................................................................................................7 2 Identifying the prominent semantic relations and correlated links .....................................................10 2.1 Verb-to-verb links and verb-to-x links .........................................................................................15 2.2 Synonymy and synsets .................................................................................................................17 2.2.1 Detecting Synonymy between verbs ......................................................................................19 2.2.2 NEAR_SYNONYMY ............................................................................................................19 2.2.3 XPOS_NEAR_SYNONYMY................................................................................................20 2.3 Hyperonymy and Hyponymy .......................................................................................................22 2.3.1 HAS_HYPERONYM and HAS_HYPONYM ......................................................................23 2.3.2 HAS_XPOS_HYPERONYM and HAS_XPOS_HYPONYM ..............................................24 2.4 Antonymy .....................................................................................................................................25 2.4.1 ANTONYMY ........................................................................................................................27 2.4.2 NEAR_ANTONYMY............................................................................................................28 2.4.3 XPOS_NEAR_ANTONYMY ...............................................................................................30 2.5 Involved ........................................................................................................................................31 2.6 ‘Causes’ and ‘Is_Caused_By’ ......................................................................................................36 2.7 Has_Subevent and Is_Subevent_Of .............................................................................................43 2.8 Be_In_State and State_Of ............................................................................................................45 3. The definition of the verb subsets ......................................................................................................46 3.1 On selecting ‘Base Concepts’ .......................................................................................................46 3.1.1 Selection of the Dutch subset .................................................................................................47 3.1.2 Selection of the Italian subset ...................................................................................................56 3.1.3 Selection of the Spanish subset ..............................................................................................58 3.1.4 First experiment on verifying intersection among the subsets ...............................................59 3.2 On extending the subset................................................................................................................59 4. Two sample classes encoded: motion verbs and ‘know verbs’ .........................................................61 4.1 On coding the relations for two sample classes of verbs ..............................................................61 4.2 First results of work on Dutch ......................................................................................................62 4.2.1. Motion Verbs...........................................................................................................................63 4.2.1.1 SYNONYMY.....................................................................................................................65 4.2.1.2 HYPONYMY ..................................................................................................................66 4.2.1.3 ANTONYMY ....................................................................................................................66 4.2.1.4 ASSOCIATIVE ...............................................................................................................67 4.2.1.5 REFERENCE...................................................................................................................68 4.2.1.6 CAUSATIVE .....................................................................................................................69 4.2.1.7 INCHOATIVE .................................................................................................................70 4.2.1.8 VERB .................................................................................................................................70 4.2.1.9 PREFERENCE...................................................................................................................71 4.2.1.10. Information in definitions ...............................................................................................72 4.2.2. Verbs of knowing ....................................................................................................................81 4.2.2.1 SYNONYMY.....................................................................................................................83 4.2.2.2 HYPERONYMY ...............................................................................................................83 4.2.2.3. HYPONYMY .................................................................................................................83 4.2.2.4 ANTONYMY ....................................................................................................................84 4.2.2.5 ASSOCIATIVE .................................................................................................................84 4.2.2.6 REFERENCE...................................................................................................................85 4.2.2.7 CAUSATIVE .....................................................................................................................85 4.2.2.8 INCHOATIVE .................................................................................................................85 4.2.2.9 VERB and PREFERENCE ................................................................................................85 4.2.2.10. Information in definitions ...............................................................................................85 4.2.3 Conclusion ................................................................................................................................93 4.3 First Results of work on Italian ....................................................................................................95 4.3.1 Motion Verbs............................................................................................................................95 4.2.2 Know verbs ..........................................................................................................................100 4.4 First results of work on Spanish .................................................................................................100 4.4.1 Subset selection of motion verbs and ‘know verbs’ ...............................................................101 4.4.2. Extraction of relations ...........................................................................................................101 5. Appendix ............................................................................................................................................104 5.1 List of tests for identifying the relations in each language .........................................................104 5.1.2 Tests for Dutch .....................................................................................................................104 5.1.2 Tests for Italian ....................................................................................................................128 5.1.3 Tests for Spanish ..................................................................................................................135 5.2 Data obtained by the analysis performed on Dutch....................................................................142 5.2.1 List of Base Concepts...........................................................................................................142 5.3 Data obtained by the analysis performed on Italian ...................................................................159 5.3.1 List of Base Concepts...........................................................................................................159 5.3.2 Motion verb synsets and taxonomy......................................................................................162 5.3.3 Know verb synsets and taxonomies .....................................................................................165 5.4 Data obtained by the analysis performed on Spanish .................................................................167 5.4.1 List of Base Concepts...........................................................................................................167 5.4.2 Motion verb synsets and taxonomies ...................................................................................170 5.4.3 Know verb synsets and taxonomies .....................................................................................183 References ..............................................................................................................................................184 EuroWordNet D006: Definition of the links and subsets for verbs 1 Introduction In recent years a trend has emerged, both in theoretical and computational linguistics, towards integrating syntactic analyses with studies on the semantic features of words. This trend evidences a radical change of perspective in linguistics, according to which the lexicon has been assigned a new, central role in the organization of language. Many researchers now agree with the claim that the syntactic properties of words are ‘determined’ by (or somehow directly connected with) their semantic features, and many studies are being devoted to the identification of classes of words sharing semantic/syntactic characteristics and, more in general, to lexical semantics. Such studies present a direct interest for computational applications, since identifying lexical classes can be of great help to the task of building lexicons for Natural Language Processing systems (cf., e.g., papers describing the work carried out within the Esprit-Acquilex project). If the research carried out so far has demonstrated the necessity of linking syntax to semantics and, therefore, of analysing the semantic features of words, further work in the field of computational linguistics is emphasizing even more clearly the importance of developing semantic databases. It is now clear that Language Engineering applications can gain great advantages from the possibility of using resources containing semantic data on words. The main goal of the EuroWordNet project, in fact, is building a (multilingual) semantic database to be used as a tool for Information Retrieval Systems (but also for other LE applications), given that access to information stored electronically is still limited by key word matching or fixed indexing and classification systems. To be able to retrieve information by using semantic data, instead of matching key words, may multiply the possibility of identifying the relevant electronic text resources; moreover, the availability of a multilingual tool, by means of which whole sets of synonyms (‘synsets’) in different languages are connected on the basis of their semantic properties, may allow easy accessing of information stored in various languages (cf. Deliverable D001). The goal of this deliverable is to report on the work carried out within Task 4.1, assigned to WP4. The objectives of this task were i) the definition of the links for verbs; ii) the identification of criteria for verification of such links for each language; iii) the definition of the subsets of verbs for which relations will be encoded in the database. Of course, the results obtained so far with respect to these three issues should be considered, within reasonable limits, to be ‘preliminary’, since further work on data might produce some (minor) changes either as far as the links, and therefore the criteria to establish them, or the verbs to be selected are concerned.1 Within WordNet 1.5 (henceforth WN 1.5) English nouns, verbs and adjectives are organized into synonym sets, each representing one underlying lexical concept. Each ‘synset’ is then linked to others by means of a number of relations: hyperonymy, hyponymy, antonymy, meronymy and holonymy (for nouns), coordinate words (i.e., words which have the same hypernym, indicated only for nouns), causation and entailment (for verbs) (cf. Miller et al. 1993). Such relations already provide useful semantic information on words, however they do not seem sufficient to describe all 1 The results obtained in the various stages of our research can be found at the WWW-site of the project (http: //www.let.uva.nl/CCL/EuroWordNet.html). Thus, at the moment, the data available are the following: 1) first list of base concepts; 2) first list of relations; 3) first definition of relations with tests. LE2-4003 EuroWordNet 7 EuroWordNet D006: Definition of the links and subsets for verbs the meaning nuances which could instead be useful for IRS and other LE applications, especially because within MRDs, which are our main sources of data, information is stored in such a way that data coded by means of these relations are only a part of all the available and necessary data. For this reason, in EuroWordNet we have distinguished additional relations, even if, since they require more complicated procedures in order to be extracted from existing repositories of data, for the time being their values will be coded only in relation to those subsets of words for which we already have enough data. Indeed, we aim at designing a ‘model’ general enough to incorporate the different types of semantic relations that are both extractable from dictionaries (and other sources) and of some usage for LE applications. The definition of such a broad model does not imply that we fill the whole of it. In fact, we will encode all the relations already indicated as being fundamental for our database within the TA, while the others will be filled more as an exemplification of their statute, their extractability, etc. In general, the definition of the relations had to meet the following requirements: • to be possible, given the resources used, to extract the relations identified; • to provide a common framework to achieve coherence during building across the sites; • to make it possible to implement consistency checking mechanisms in the multilingual database; • to make it possible for external reviewers to verify the quality of the results obtained. In order to explore the effectiveness of different approaches, methodologies of analysis and solutions for encoding data, but also for arriving both at a final definitions of the relations to be included in our database and at a definition of clear criteria for identifying them in each language, each site performed an ‘experiment’, by taking into account two small semantic fields: i) the taxonomy built starting from the hyperonym verb(s) corresponding to WN 1.5 {move, change position} synset;2 ii) ‘know’ verbs. The purpose of the experiment, whose results will be reported in the deliverable, was, therefore, twofold: • to determine the best approach and methodologies of analysis for the different resources used by each group; • to explicitly define the relations and the criteria to encode them for each language. As recalled above, another basic issue to be dealt with within Task 4.1 was the definition of the subsets of verbs for which the relations will be encoded in the database. Thus, first of all we have carried out a work aimed at identifying a set of common ‘Base Concepts’ for each language wordnet, by taking into consideration the defining vocabulary of each dictionary. The construction of such initial subset was important to set up a common framework to build the individual wordnets. The aim was to create a core subset for each language that contains the most relevant synsets (according to the number of relations per synset and their position in the hierarchy) and is highly overlapped with the core subsets of the other languages. Secondly, criteria have been clearly defined to extend this first subset in order to obtain the about 15,000 verb synsets to be encoded for each wordnet, as indicated within the TA. The methodology identified should guarantee that each language wordnet will be maximally balanced and compatible with the others, in order to allow a real multilingual usage of the database. The document is organised as follows. In section 2. we provide a brief discussion and definition of the relations already encoded with respect to verbs in WN 1.5 and also used to develop 2 Here and later, the curly brackets are used to indicate a synset. LE2-4003 EuroWordNet 8 EuroWordNet D006: Definition of the links and subsets for verbs our resource, and of the new relations added in order to code all the information which we may acquire from MRDs, but also from other sources of lexical data, and which can be useful for both IRSs and other LE applications. Clear criteria for the identification of each relation are also provided. Although we shall briefly refer to extraction issues, the extraction procedures for each wordnet will be described in detail in the deliverables D021 and D025. In section 3. we discuss issues connected with the definition of the subsets of verbs which will be encoded in our database, explaining in particular the methodology followed in order to identify a first subset of common Base Concepts. In section 4. we discuss the first results obtained by trying to code data on the relations chosen with respect to the two ‘sample’ classes of verbs: motion and know verbs. Finally, in the Appendix we provide: i) the specific tests elaborated for each language in order to identify relations; ii) detailed data related to the Base Concepts identified so far for each language; iii) detailed data on the results of the ‘experiments’ on coding information on relations for both motion and know verbs. LE2-4003 EuroWordNet 9 10 EuroWordNet D006: Definition of the links and subsets for verbs 2 Identifying the prominent semantic relations and correlated links Within this deliverable we are concerned with the ‘Language Internal Relations’ (cf. D007 on the “High-Level Architecture of the EuroWordNet Database”) and the kind of criteria defined to identify them. We shall, however, try to limit ourselves to considerations which are particularly relevant with respect to the verb relations, since a discussion of the relations concerning nouns is provided in D005 (on the “Definition of the links and subsets for nouns”), together with a more general introduction to the characteristics of the relations and the database. Notwithstanding this, for clearness of exposition, in some cases it will be necessary to repeat information somehow already provided in the Deliverable on nouns. The definition of the final list of relations required a quite long period of discussion and elaboration by the partners of the project. Here below, the complete list of the relations on which we have agreed upon is reported for a quick reference, with the following information for each relation: i) its name, ii) the parts of speech linked (with the indication of the ‘direction’ of the linking), iii) further labels which eventually apply to the relation itself, iv) the type of data linked (i.e., Word Meanings or synset Variants). Relation Type Parts of Speech Labels Data Types NEAR_SYNONYM XPOS_NEAR_SYNONYM N<>N, V<>V N<>V, N<>AdjAdv, V<>AdjAdv reversed reversed WM<>WM WM<>WM HAS_HYPERONYM HAS_HYPONYM HAS_XPOS_HYPERONYM N>N, V>V N>N, V>V N>V, N>AdjAdv, V>AdjAdv, V>N, AdjAdv>N, AdjAdv>V N>V, N>AdjAdv, V>AdjAdv, V>N, AdjAdv>N, AdjAdv>V Dis, Con, reversed Dis, reversed Dis, Con, reversed WM<>WM WM<>WM WM<>WM Dis, reversed WM<>WM N>N N>N N>N N>N N>N N>N N>N N>N N>N N>N Dis, Con, reversed, not Dis, Con, reversed, not Dis, Con, reversed, not Dis, Con, reversed, not Dis, Con, reversed, not Dis, Con, reversed, not Dis, Con, reversed, not Dis, Con, reversed, not Dis, Con, reversed, not Dis, Con, reversed, not WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM HAS_XPOS_HYPONYM HAS_HOLONYM HAS_HOLO_PART HAS_HOLO_MEMBER HAS_HOLO_PORTION HAS_HOLO_MADEOF HAS_HOLO_LOCATION HAS_MERONYM HAS_MERO_PART HAS_MERO_MEMBER HAS_MERO_MADEOF LE2-4003 EuroWordNet 11 EuroWordNet D006: Definition of the links and subsets for verbs HAS_MERO_LOCATION N>N Dis, Con, reversed, not WM<>WM ANTONYM NEAR_ANTONYM XPOS_NEAR_ANTONYM N<>N, V<>V N<>N, V<>V N<>V, N<>AdjAdv, V<>AdjAdv reversed reversed reversed VA<>VA WM<>WM WM<>WM CAUSES V>V, N>V, N>N, V>N, V>AdjAdv, Dis, Con, N>AdjAdv reversed, not V>V, N>V, N>N, V>N, AdjAdv>V, Dis, Con, AdjAdv>N reversed, not IS_CAUSED_BY Factive, WM<>WM Factive, WM<>WM HAS_SUBEVENT IS_SUBEVENT_OF V>V, N>V, N>N, V>N V>V, N>V, N>N, V>N Dis, Con, reversed, not Dis, Con, reversed, not WM<>WM WM<>WM ROLE ROLE_AGENT ROLE_INSTRUMENT ROLE_PATIENT ROLE_LOCATION ROLE_DIRECTION ROLE_SOURCE_ DIRECTION ROLE_TARGET_ DIRECTION INVOLVED INVOLVED_AGENT INVOLVED_PATIENT INVOLVED_INSTRUMENT INVOLVED_LOCATION INVOLVED_DIRECTION INVOLVED_SOURCE_DIRECTION INVOLVED_TARGET_DIRECTION N>V, N>N N>V, N>N N>V, N>N N>V, N>N N>V, N>N, AdjAdv>N, AdjAdv>V N>V, N>N, AdjAdv>N, AdjAdv>V N>V, N>N, AdjAdv>N, AdjAdv>V N>V, N>N, AdjAdv>N, AdjAdv>V V>N, N>N V>N, N>N V>N, N>N V>N, N>N V>N, N>N, N>AdjAdv, V>AdjAdv V>N, N>N, N>AdjAdv, V>AdjAdv V>N, N>N, N>AdjAdv, V>AdjAdv V>N, N>N, N>AdjAdv, V>AdjAdv Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed Dis, Con, reversed WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM WM<>WM BE_IN_STATE STATE_OF N>AdjAdv, V>AdjAdv AdjAdv>N, AdjAdv>V Dis, Con, reversed, not Dis, Con, reversed, not WM<>WM WM<>WM FUZZYNYM XPOS_FUZZYNYM N<>N, V<>V N<>V, V<>AdjAdv, N<>AdjAdv WM<>WM WM<>WM EQ_SYNONYM EQ_NEAR_SYNONYM N<>N, V<>V N<>N, V<>V, V<>AdjAdv, WM<>ILIR WM<>ILIR HAS_EQ_HYPERONYM N>N, N>V, N>AdjAdv, V>V, V>N, V>AdjAdv, AdjAdv>N, AdjAdv>V N>N, N>V, N>AdjAdv, V>V, V>N, V>AdjAdv, AdjAdv>N, AdjAdv>V N>N N>N HAS_EQ_HYPONYM HAS_EQ_HOLONYM HAS_EQ_MERONYM LE2-4003 N<>AdjAdv, WM<>ILIR WM<>ILIR WM<>ILIR WM<>ILIR WM<>ILIR EuroWordNet 12 EuroWordNet D006: Definition of the links and subsets for verbs EQ_INVOLVED EQ_ROLE N>N, V>N N>N, N>V EQ_CAUSES N>N, V>V, N>V, V>N, V>AdjAdv, N>AdjAdv N>N, V>V, N>V, V>N, AdjAdv>V, AdjAdvV>N EQ_IS_CAUSED_BY WM<>ILIR WM<>ILIR WM<>ILIR WM<>ILIR WM<>ILIR EQ_HAS_SUBEVENT EQ_IS_SUBEVENT_OF N>N, V>V, N>V, V>N N>N, V>V, N>V, V>N WM<>ILIR WM<>ILIR EQ_BE_IN_STATE EQ_IS_STATE_OF V>AdjAdv, N>AdjAdv AdjAdv>V, AdjAdv>N, WM<>ILIR WM<>ILIR HAS_INSTANCE BELONGS_TO_CLASS N>PN PN>N WM>I I>WM Parts of Speech: N = noun V = verb AdjAdv = Adjective or Adverb PN = pronoun or name Labels: factive/non-factive = restricted to Lyon’s distinction between factive causes and modally more complex causes. reversed = for relations which are not explicitly stated but generated by automatic reversal. Dis/Con = for explicitly coded disjunction/conjunction of relations. not = for cases in which the negative implication holds between word pairs. Data types: WM = word meaning or synset I = instance ILIR = ILI record VA = synset variant Thus, for instance, if we take into consideration the ‘CAUSES’ relation we see that it may link: i) a verb (synset)3 to another verb; ii) a noun to a verb; iii) a noun to a noun; iv) a verb to a noun. Then, it may be further labeled, under certain circumstances. The labels ‘Con’ and ‘Dis’ respectively indicate the possibility of the relation forming conjunctive or disjunctive sets (cf. D001): for instance, to give causes both to receive and to have, whose synsets will be conjunct, whereas to race may cause either to win or to lose and these verb synsets will then be disjunct. The label ‘Factive’ is instead used to indicate that the relation necessarily holds. In fact, following Lyons (1977), different types of causality can be distinguished, reflecting the factivity of the effect: • factive: event E1 implies the causation of E2: e.g. “to kill causes to die”; 3 From now on, if not explicitly otherwise indicated, the words verb or noun will be used to indicate also the synsets containing them. LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 13 • non-factive: E1 probably or likely causes event E2 or E1 is intended to cause some event E2: “to search may cause to find”. The label ‘reversed’, then, is used when a relation has been automatically stated by reversing its counter-part relation. Indeed, it is a requirement of the database that every relation has a reverse counter-part. However, there is a difference between relations which are explicitly coded as reverse relations and relations which are automatically reversed because of this requirement. For example, to chain causes to be chained and, on the other hand, to be chained is necessarily caused by to chain, i.e. in this case we have a conceptually bi-directional link. In the case of the pair kill/die, instead, the fact that to die will result as being ‘CAUSED_BY’ to kill is due to the reversing of the factive-causation relation linking to kill to to die: thus, by labelling the relation as ‘reversed’ we imply that to die is not necessarily caused by to kill. In other words, to be able to distinguish between the conceptually-dependent and the automatically reversed relations we shall use the label ‘reversed’. Finally, there is the possibility to have a negation label ‘not’ to explicitly express that a relation does not hold. Such a negative relation is not necessarily the same as antonymy, which expresses opposition. For instance, to ie can in no case cause to move. As said above, here we shall take into consideration only the LIRs, assuming that interlingual relations (between each language and WN 1.5, which is used as an interlingual connection) are equivalent. Thus, for instance, SYNONYM and EQ_SYNONYM can be defined in a similar way, even if the latter holds between words of different languages. In principle, the relations are incompatible: i.e., if two verbs are linked by a relation they cannot be linked by means of any other relation. Some exceptions to this general rule have, however, been found. For instance, sometimes it can be difficult to distinguish between hyponymy and cause relation: this happens for the Dutch verb dichttrekken (to pull in order to close), which could be described both as a hyponym and as a cause of the verb sluiten (to close). In this case, it is very difficult to keep the two relations apart because the events referred to by the verbs are not disjoint and, in fact, it is also difficult to state if these events are really different (in any case, the decision has been taken to encode this, and similar cases, as hyponymy relations). Moreover, also the HAS_SUBEVENT relation may in some cases display a kind of overlap with other relations. For instance, to buy has been described as connected to the verb to pay by this relation (i.e., pay IS_SUBEVENT_OF buy - non-factive - / buy HAS_SUBEVENT pay), but since this can be seen as a sort of ‘class-inclusion’ relation, we could also state that the two verbs are related by hyperonymy-hyponymy.4 In order to develop criteria for verifying the existence of the semantic relations we tried to identify substition tests or diagnostic frames based on normality judgements (cf. Cruse 1986). For every relation, thus, some general tests will be provided containing different frames; by inserting two words, different ‘normality’/‘abnormality’ judgements will be determined (generally indicated by means of a ‘yes/no’ score, but, in some cases, also other values will be allowed: probably, unclear, 4 Cf. D005 for a similar discussion with respect to compatibility of hyperonymy/hyponymy and set-membership of nouns. Indeed, the HAS_SUBEVENT relation functions as a kind of ‘meronymy relation’ for verbs. LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 14 unlikely), on the basis of which a relation can be detected. These tests are semantic and not intended to cover syntactic differences, differences in register, style or dialect nor pragmatic differences between words. Since there are cases in which it is difficult to assign clear scores to the tests, or the results are all negative, but the test X has some strong relation to Y still works, we have created a relation called ‘FUZZYNYM’. A fuzzynym relation can hold between same part-of-speech words or across parts of speech. In the following, while defining the specific relations being encoded for verbs and providing general formal criteria for identifying them, we shall discuss in more details, when necessary, issues connected with the status of each relation and with the labels which eventually apply to it. In the Appendix also more specific tests for each language will be provided. LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 15 2.1 Verb-to-verb links and verb-to-x links Within WN 1.5 all the relations encoded apply among words of the same lexical class. However, both theoretical and practical reasons led us to feel the necessity of adding to our database also semantic relations applying to elements of different lexical classes. Indeed, one of the issues discussed within D001 was the necessity of interlinking the noun and verb networks. Actually, the Princeton WordNet uses a rigid distinction between nouns and verbs, mainly because of their different syntactic role in English. Furthermore, it is suggested that the responses of people to interviews parallel the rigid noun/verb distinction as well: nouns are related to nouns and verbs are related to verbs. As shown in D001 (p. 34), this “separation however leads to a very undesirable situation that very similar synsets (which in some cases even have exactly the same names for the tops) are totally unrelated only because they differ in part of speech”. Thus, rather then semantically distinguishing nouns and verbs, the decision was taken to distinguish between first-order and higher-order entities, where nouns can refer to first-order entities (concrete, physical things) and both nouns and verbs can refer to higher-order entities (properties of things, relations, acts, activities, processes, states, etc. (cf. D001, p.34). Further reasons motivating this decision were also indicated (ibid.): • In other languages it is either very difficult to distinguish nouns and verbs or the distinction does not even exist (Lyons 1977). • From an information retrieval point of view the same information can be coded in an NP or in a sentence. By unifying higher-order nouns and verbs in the same ontology it will be possible to match expressions with very different syntactic structures but comparable content (see the Sift project, LRE 62030). • By merging verbs and abstract nouns we can more easily link mismatches across languages that involve a part-of-speech shift. Therefore, we have inserted in the architecture of our database synonymy, hyperonymy/hyponymy and antonymy relations applying across parts of speech. As a consequence of further developments of our research, besides allowing the linking between noun and verb hierarchies, these relations can now be used to link also adjectives and adverbs to the noun and verb networks. Of course, since in the database being developed in this project we are only building noun and verb networks, we shall not create wordnets for adjectives and adverbs as well, but we shall simply include the adjectives or adverbs referred to by means of the ‘XPOS’ relations in the lexicon, without creating any other link for them. Within the Deliverable D001 it was, then, pointed out that MRDs will be our main sources of data and that the information there is stored in such a way that the relations used in the Princeton wordnet are not adequate to code the kind of data that they contain and that are necessary for IRS (and other LE applications). In fact, the frequency of ‘shallow’ hierarchy structures in dictionaries “suggests that other relations are more important or useful for classifying verbs” (D001, p. 44) with respect to those available within WN 1.5. Consider, for instance, Italian dictionaries: there we find many (classes of) verbs defined by means of a phrase whose syntactic head (the hyperonym) is not very significant LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 16 whereas what follows it (be it a noun, a NP, a PP, etc.) is of fundamental importance in order to obtain necessary information on the semantics of the verb considered. The following verbs of motion and their definitions can be taken as examples: camminare = to walk cavalcare = to ride “andare a piedi” to go on foot “andare a cavallo” to go on a horse coricarsi = “andare a letto” to go to bed navigare = to navigate “andare (detto di navi)” to go (said of ships) uscire = to go out “andare, venire fuori” to go, to come out All such verbs are hyponyms of the verb andare (to go) which simply indicates motion along a path, however: the first verb is a manner-of-motion verb, the second one indicates motion by means of a vehicle, the third one motion to a specific place (incorporated within the meaning of the verb itself), the fourth one is used to refer to motion performed (only) by specific vehicles, and the last one indicates motion from and to partially-specified locations (from inside to outside). In order to be able to code more detailed information on these verb meaning components we should, in principle, take into consideration any element in their definitions. A somehow similar situation holds in the case of the diventare/rendere (to become/to make) hyponyms which cannot be simply connected with their superordinates if one does not want to lose some relevant information: in fact, verbs like arrossire (“diventare rosso” = to become red) or imbiancarsi (“diventare bianco” = to become white), intristire (“diventare triste” = to become sad) and ingrassarsi (“diventare grasso” = to become fat) are all inchoative, as we infer from the fact that they are all hyponyms of diventare, but they also display semantic differences which may be very important from the point of view of LE applications: indeed, whereas the former two verbs refer to a change in colour, the third one refers to a change in mood and the latter to a change in weight. Again, it seemed necessary to define a model providing a framework which is enriched with respect to the WN 1.5 inventory, so to have links which might connect each entry-verb with all the significant elements found in its dictionary definition and useful from the point of view of a EuroWordNet user. Some new relations have therefore been identified, which should cover a number of semantic relations among words from different parts of speech, in order to give us the possibility to code the data on the semantics of words (different from those coded by means of already existing relations) which we are able to acquire in one way or another and which are relevant to delineate the meaning of the LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 17 verb. Indeed, we have assumed a view of the meaning of a word similar to that proposed by Cruse (1986) (and before by Haas, 1964), according to which the meaning of a word is conceived of as a kind of ‘semantic field’, containing all possible (grammatical) sentential contexts of the word, and all possible (grammatical) substitutes within those contexts.5 This means to see the meaning of a word as being made up, at least in part, of (the meanings of) other words, which correspond, therefore, to ‘semantic traits’ of the first word. Following Cruse, such semantic traits should carry the lightest possible burden of theory, and in fact no claim is made that they are either primitive, functionally discrete, universal, or drawn from a finite inventory; furthermore, in Cruse’s view there is no claim that the meaning of any word can be exhaustively characterised by any finite set of such semantic traits. Such an ‘enlarged’, even if lacking in strong theoretical assumptions, view of the meaning of a word may allow us coding a large inventory of semantic data, i.e., different arguments ‘incorporated’ within the verb root, by means of the indication of the actual words generally used to define another word (mainly within dictionaries but also in other sources). By adopting this view, we can code information on ‘semantic components’, by relating a word with other words (e. g., a verb with the words which can be found together with its hyperonym in a context) rather than words with simple labels for concepts. In other words, we aim at encoding semantic relations among words as they are lexicalized in a language system. Although this is different from encoding world-knowledge data, the linguistic database could then be used as a basis for eventually deriving a world-knowledge database. 2.2 Synonymy and synsets That of synonymy is the fundamental relation on which WN 1.5 is based, since the first organization of words is realized by means of ‘synsets’, i.e. sets of synonymous words. The notion of synonymy adopted is not the strongest one, according to which two expressions are synonymous if the substitution of one for the other never changes the truth value of a sentence in which the substitution is made, but a weakened version of it, stating that “two expressions are synonymous in a linguistic context C if the substitution of one for the other in C does not alter the truth value.” (Miller et al. 1993) The notion of synonymy adopted in our project in order to build synsets is the same as that of WN 1.5. Synsets are built on the basis of the possibility of a word to be substituted by another in a specific context, and word (synset) senses are therefore given by all the possible uses of a word in contexts. The general test used to state that two verbs are synonymous is the following: He Vs 1 entails and is entailed by He Vs2 5 This view of the meaning of a word is similar to the ‘relational’ view, adopted, e.g., by Lyons (1968; 1977), according to whom the meaning of a lexical item can be defined as not only dependent, but also identical to the set of relations which hold between the item in question and the other lexical items in the system. LE2-4003 EuroWordNet 18 EuroWordNet D006: Definition of the links and subsets for verbs Which means that both the following sentences must be true, for two verbs to be synonymous: - He Vs1 therefore he Vs2 yes - He Vs2 therefore he Vs1 yes Clear yes-yes = synonymy However, such test may be performed only by considering the actual context in which two (or more) verbs can be found. For instance, if we take into consideration the Italian verb allontanarsi (= either to move/go away or to go far), we can state that when used in the context (1a) below, it has certain synonyms, whereas in context (1b) it has others, since the verbs allowed respectively in (1a) and (1b) are in a relationship of reciprocal entailment in those contexts: (1a) (1b) Giovanni non poteva allontanarsi/muoversi (*andare lontano) da casa.6 Giovanni could not move/go away (*go far) from home. Giovanni si allontanò troppo/andò troppo lontano (*si mosse troppo) dalla riva e non riuscii più a vederlo. Giovanni went too far away (*moved far) from the bank and I could not see him anymore. To put it in other words, we could state that: in any sentence S where Verb1 is the head of a VP which is used to identify a situation in discourse another verb Verb2, which is a synonym of Verb1, can be used as the head of the same VP without resulting in semantic anomaly and in any sentence S where Verb2 is the head of an VP which is used to identify a situation in discourse another verb Verb1, which is a synonym of Verb2, can be used as the head of the same VP without resulting in semantic anomaly. 6 As a matter of fact, andare lontano can occur in such a context but with a slightly different meaning with respect to the synset of allontanarsi/muoversi, i.e. it cannot substitute that synset without altering the meaning of the sentence. LE2-4003 EuroWordNet 19 EuroWordNet D006: Definition of the links and subsets for verbs 2.2.1 Detecting Synonymy between verbs In the following, a general formal criterion for the identification of synonymy between verbs is provided: Verb test (1) Comment: Score yes a yes b Conditions: Example: a b Effect: Synonymy between verbs Test sentence If something/someone/it Xs then something/someone/it Ys iI something/someone/it Ys then something/someone/it Xs - X is a verb in the third person singular form - Y is a verb in the third person singular form - there are no specifying PPs that apply to the X-phrase or the Y-phrase If something/someone/it begins then something/someone/it starts If something/someone/it starts then something/someone/it begins synset variants: {begin, start} In practice, the relation of synonymy can be acquired from different information sources, i.e. from verb entries of both mono- and bi-lingual dictionaries. For instance, as far as Italian is concerned, explicitly tagged synonyms are extracted from the synonym fields of a monolingual dictionary and from a synonym dictionary; further information is acquired from synonymical definitions of monoligual dictionaries and semantic indicators of a bilingual dictionary. The data automatically extracted from the two latter sources require manual revision and, possibly, correction. Indeed, in these two cases the synonymy relation is not established on the basis of semantic, cognitive or distributional criteria, but simply on the basis of ‘external’ criteria such as, for instance, the definition structure: e.g. whenever a given word is defined by means of a single word (i.e. the definition does not contain either arguments or modifiers), these two words are taken to be somehow synonymous. Thus, when the verb returned by the automatic procedure as synonym displays more than one sense, firstly we have to (manually) identify which is the synonym sense, and then we have to perform tests to verify the existence of synonymy in a context. 2.2.2 NEAR_SYNONYMY As we have seen, the basic organization of words within our database will be provided by synsets, built by using a weakened notion of synonymy. There are, however, cases in which it is very difficult to state if two words, although displaying a very ‘similar’ meaning, can be put in a same synset, i.e. are ‘true’ synonyms in the sense explained above. In fact, sometimes the synonym test between two verbs may yield a not clear score and, more important, the two verbs hyponyms cannot be interchanged or the hyperonyms are also incompatible, even if these verbs cannot be considered either connected by a hyponymy relation or by any other relation. The NEAR_SYNONYMY relation can in general be used for these verb pairs. A general criterion to identify near-synonyms is the following: LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 20 Verb Test (2) Comment: Score yes, probably a Near-Synonymy between verbs Test sentence If something/someone/it Xs then something/someone/it Ys and yes/yes i) Xh and Yh are different or ii) Hx and Hy are different yes, probably b If something/someone/it Ys then something/someone/it Xs and yes/yes i) Yh and Xh are different or ii) Hy and Hx are different Conditions: - X and Y are verbs in the third person singular - X and Y are not linked by any other relation - Xh is the set of the hyponyms (if there are any) of X - Yh is the set of the hyponyms (if there are any) of Y - Hx is the set of the hyperonyms (if there are any) of X - Hy is the set of the hyperonyms (if there are any) of Y Effect: X NEAR_SYNONYM Y Y NEAR_SYNONYM X Using the NEAR_SYNONYMY relation we can keep sets of hyponyms separate while we can still encode that two synsets are close in meaning. 2.2.3 XPOS_NEAR_SYNONYMY As explained above, and also indicated within the table listing all the relations, some links have been created to allow synonymy (called in this case ‘near-synonymy’, because of its characteristics) between different parts of speech. In the following, we provide general criteria for the definition of XPOS_NEAR_SYNONYMY. LE2-4003 EuroWordNet 21 EuroWordNet D006: Definition of the links and subsets for verbs Verb test (3) Comment: Score yes a yes b Conditions: Example (1): a b Effect: Example (2): a b Effect: Verb test (4) Comment: Score yes a yes b Conditions: Example: a b Effect: LE2-4003 Synonymy between verbs and nouns denoting events or processes Test sentence If something/someone/it Xs then a/an Y takes place If a/an Y takes place then something/someone/it Xs - X is a verb in the third person singular form - Y is a noun in the singular - there are no specifying PPs that apply to the X-phrase or the Y-phrase - preferably there is a morphological link between the noun and the verb If something/someone/it murdersthen a murder takes place If a murder takes place then something/someone/it murders to murder (X) XPOS_NEAR_SYNONYM murder (Y) murder (Y) XPOS_NEAR_SYNONYM to murder (X) If something/someone/it moves then a movement takes place If a movement takes place then something/someone/it moves to move (X) XPOS_NEAR_SYNONYM movement (Y) movement (Y) XPOS_NEAR_SYNONYM to move (X) Synonymy between verbs and nouns denoting states Test sentence If something/someone/it Xs then there is a state of Y If there is a state of Y then something/someone/it Xs - X is a verb in the third person singular form - Y is a noun in the singular - there are no specifying PPs that apply to the X-phrase or the Y-phrase - preferably there is a morphological link between the noun and the verb If something/someone/it exists then there is a state of existence If there is a state of existence then something/someone/it exists to exist (X) XPOS_NEAR_SYNONYM existence (Y) existence (Y) XPOS_NEAR_SYNONYM to exist (X) EuroWordNet 22 EuroWordNet D006: Definition of the links and subsets for verbs Verb test (5) Comment: Score yes a yes b Conditions: Example: a b Effect: Synonymy between state-denoting verbs and adjectives (or adverbs)7 Test sentence if something/someone/it Xs then something/someone/it is Y if something/someone/it is Y then something/someone/it Xs - X is a verb in the third person singular form - Y is an adjective - there are no specifying PPs that apply to the X-phrase or the Y-phrase - preferably there is a morphological link between the noun and the verb if someone/it lives then someone/it is alive if someone/it is alive then someone/it lives to live (X) XPOS_NEAR_SYNONYM alive (Y) alive (Y) XPOS_NEAR_SYNONYM to live (X) 2.3 Hyperonymy and Hyponymy The relations of hyperonymy and hyponymy have been largely studied and used in theoretical and applicative works, and they play a central role also in our tool, since they individuate taxonomies, i. e. classes of words. Thus, for instance, in order to identify a set of motion verbs we can look for all the hyponyms of a synset containing top-verbs simply referring to motion. If we want to narrow our search to verbs referring, for instance, to directed-motion we can look for the hyponyms of a top-verb synset indicating directed-motion. The advantages of organizing the lexicon into taxonomies are obvious both for the theoretical field and for applicative uses, the most important one being the possibility of making inferences. The general criterion used to state the existence of the relations for verbs is given in the following (adapted from Cruse 1986): A verb synset X is a hyponym of another verb synset Y (and, by the same token, Y a hyperonym of X) if He is X-ing entails but is not entailed by He is Y-ing. Which means that the following sentences should be respectively true and false, for two verbs be connected by means of a hyperonymy/hyponymy relation: - He Vs1 therefore he Vs2 yes - He Vs2 therefore he Vs1 no Clear yes-no = V1 is a hyponym of V2 (and V2 is a hyperonym of V1) 7 Since languages tend to differ considerably in what are adjectives and what are adverbs (e.g., in Dutch many adjectives can be used as adverbs, whereas this is impossible in English), we have decided to allow both categories here and in every other case of a relation between a verb (or a noun) and an adjective. LE2-4003 EuroWordNet 23 EuroWordNet D006: Definition of the links and subsets for verbs 2.3.1 HAS_HYPERONYM and HAS_HYPONYM The general test provided above seems, however, not sufficiently discriminating, because it does not distinguish between verbs connected by a hyponymy relation and verbs connected by a more general entailment relation. In fact, in this test, V1 could be, for instance, to snore and V2 could be to sleep (indeed, He is snoring entails but is not entailed by He is sleeping), which are not connected by a hyponymy relation. Thus, the test should be reformulated in order to avoid similar wrong results. Since each hyponym is equivalent to a paraphrase in which its hyperonym is syntagmatically modified, we can state the following formal criteria for the definition of hyperonymy/hyponymy: Verb test (6) Comment: Score yes a no b Conditions: Example: a b Effect: Hyperonymy/hyponymy between verb synsets Test sentence X is Y + AdvP/AdjP/AdjP/NP/PP Y is X + AdvP/AdjP/NP/PP - X is a verb in the infinitive form - Y is a verb in the infinitive form - there is at least one specifying AdvP, NP or PP that applies to the Y-phrase to run is to go fast * to go is to run fast {to run} (X) HAS_HYPERONYM {to go} (Y) {to go} (Y) HAS_HYPONYM {to run} (X) Hyperonymy/hyponymy relations are typically acquired from dictionary definitions formulated as complex verbal phrases consisting of a syntactic head and of arguments and/or modifiers. In definitions with this syntactic structure, the head verb represents the hyperonym of the verb being defined; conversely, the definiendum is a hyponym of the definition head.8 8 It seems necessary to clarify that although by means of automatic procedures it is ‘easy’ to identify these relations within dictionaries, such identification does not directly lead to the development of taxonomies, i.e. chains constituted by a top hyperonym and all its direct and indirect hyponyms. Indeed, as we have already remarked with respect to the identification of synonymy between word senses, each genus verb may display different senses, one of which has to be identified as the hyperonym sense. As far as Italian is concerned such sense disambiguation has to be performed ‘manually’. LE2-4003 EuroWordNet 24 EuroWordNet D006: Definition of the links and subsets for verbs 2.3.2 HAS_XPOS_HYPERONYM and HAS_XPOS_HYPONYM As already mentioned, a link has been created to allow hyperonymy/hyponymy across parts of speech. In the following, general criteria for its identification are provided. Verb test (7) Comment: Score yes a no b Conditions: Example: a b Effect: Verb test (8) Comment: Score yes a no b Conditions: Example: a b Effect: Hyperonymy/Hyponymy between verbs and nouns denoting events and processes9 Test sentence If a/an Y takes place then something/someone/it Xs + AdvP/AdjP/NP/PP If something/someone/it Xs then a certain Y takes place - X is a verb in the third person singular form - Y is a noun in the singular - there is at least one specifying AdvP, NP or PP that applies to the X-phrase - preferably there is no morphological link between the noun and the verb If an arrival takes place then someone/it goes to a place * If someone/it goes then a certain arrival takes place arrival (Y) HAS_XPOS_HYPERONYM to go (X) to go (X) HAS_XPOS_HYPONYM arrival (Y) Hyperonymy/Hyponymy between verbs and nouns denoting states Test sentence If there is a state of Y then something/someone/it Xs + AdvP/AdjP/NP/PP If something/someone/it Xs then there is a certain state of Y - X is a verb in the the third person singular form - Y is a noun in the singular - there is at least one specifying AdvP, NP or PP that applies to the X-phrase - preferably there is no morphological link between the noun and the verb If there is a state of paranoia then someone fears something intensively * If someone fears something then there is a certain state of paranoia paranoia (Y) HAS_XPOS_HYPERONYM to fear (X) to fear (X) HAS_XPOS_HYPONYM paranoia(Y) 9 Note that here and below we provide tests identifying a verb as hyperonym and a noun or adjective/adverb as hyponym. Of course, similar tests can be used in the reverse cases. LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs Verb test (9) Comment: Score yes a no b Conditions: Example: a b Effect: 25 Hyperonymy/Hyponymy between state-denoting verbs and adjectives (or adverbs) Test sentence If something/someone/it is Y then something/someone/it Xs + AdvP/AdjP/NP/PP If something/someone/it Xs then something/someone/it is in a certain state of being Y - X is a verb in the third person singular form - Y is an adjective - there is at least one specifying AdvP, NP or PP that applies to the X-phrase - preferably there is no morphological link between the adjective and the verb If someone is horrified then someone fears something intensively * If someone fears something then someone is in a certain state of being horrified horrified (Y) HAS_XPOS_HYPERONYM to fear (X) to fear (X) HAS_XPOS_HYPONYM horrified (Y) 2.4 Antonymy Although the antonymy (or opposition) relation is not always indicated in the sources which we are using, Fellbaum (1993:49 ff) provides evidence that it is a particularly psychologically salient relation for verbs. Its semantics is complex and actually there is no general agreement on the terminology used (e.g., Lyons 1968, 1977 distinguishes among three types of ‘meaning oppositions’ and he uses the term ‘antonymy’ to refer just to one of these three types, but other authors make different distinctions). However, the notion we are referring to in our work seems to correspond more to what Lyons calls in general ‘opposition’, that is to a more loosely defined notion. If synonymy has been defined as the equivalence of meaning in a context, antonymy can be defined as the opposition of meaning in a context. Much of the oppositions among verbs are revealed by morphological elements: tie/untie, apparire/sparire (appear/disappear), approvare/disapprovare (approve/disapprove), etc. However, in other cases, the antonymy rises from the opposition existing between adjectives incorporated within the meaning of verbs: abbellire/imbruttire (prettify/uglify), dimagrire/ingrassare (slim/fat), etc. Still other verbs are then opposed because there is opposition between different kinds of semantic components incorporated within their meanings: for instance, if we take into consideration the hyponyms of the Italian verb andare (to go), we can see that within the meaning of certain verbs the hyperonym is conflated with semantic components denoting opposed direction (e. g., entrare/uscire - to go in/to go out, salire/scendere - to go up/to go down). Other pairs, considered to be antonymous in WN 1.5, are not found within the same taxonomies, neither can be identified by means of morphological markers; however, they occur within the same semantic field and “refer to the same activity, but from the viewpoint of different participants” (Fellbaum 1993:51): lend/borrow, teach/learn, buy/sell, etc. LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 26 Given our notion of antonymy, the general test used to state that two verb synsets are antonymous is the following, ‘bidirectional’ test: X is the opposite of Y Which means that both the following sentences must be true, in order the two verb synsets be antonymous: - He Xs entails that he does not Ys yes - He Ys entails that he does not Xs yes Clear yes-yes = antonymy A test like this, however, needs to receive some further constraints which avoid obtaining as a result two verbs contrasting in some of their semantic features but not really definiable as ‘antonyms’. For instance, the test above might work with X= eat and Y= sleep, since the two verbs refer to processes which can never occur at the same time (since the former requires that the protagonist is actively involved, whereas the latter requires a ‘passive’ protagonist), even if they are not each other’s opposites. In fact, antonyms typically form contrasting categories within the same dimension. This means that a verb not only contrasts with an antonym-verb in one or more features (e.g. active protagonist/passive protagonist) but that they often have to share the same hyperonym: i.e. they have to be competitors within a reasonable denotational range. This latter criterion prevents us from contrasting irrelevant pairs such as eat and sleep. There is, however, also the possibility to have antonyms not occurring within the same taxonomy; in this case, either their hyperonyms are antonyms or the two verbs display a particular relation: they share the main participants, but these have opposite roles in the situations10 referred to by the verbs themselves. For instance, vendere (to sell) and comprare (to buy) can be said to be antonymous because they are respectively hyponyms of cedere (to give) and ricevere (to receive), which are antonymous. The latter two verbs, then, are found at the top of their taxonomies, thus their antonymy is revealed by the fact that the indirect object of cedere, i.e. the addressee involved in the event referred to by this verb, is the subject (protagonist) of ricevere. The main information source for the antonymy relation are dictionary definitions. It is often the case that the definiendum is described in terms of its verbal opposite(s). In Italian, e.g., the antonymy relation can be formally characterized by the presence of the adverbial non (‘not’) pre-modifying the verbal head of a definition: this adverbial is used to trigger the identification of such relation; the value of the relation, i.e. the antonym of the definiendum, is the definition head. However, this is not the only way in which pairs of antonyms can be acquired from dictionary verb definitions: other more complex strategies can be envisaged for acquiring this kind of relation, which imply the simultaneous 10 The word ‘situation’ is used throughout the document to generically refer to the states or events or processes referred to by verbs. In keeping with von Wright (1963) and others, we shall, moreover, use i) the term ‘state’ to refer to a state of affairs; ii) the term ‘event’ to refer to a change of state; iii) the term ‘process’ to refer to sequences of micro-events and states, felt, though, at the linguistic level, as homogeneous ongoing situations (e.g., the situation referred to by the verb to write). The term ‘actions’ will also be used, to refer both to agentive events and agentive processes. LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 27 consideration and comparison of different verb entries. For instance, within the Italian LDB the two change-of-state verbs rendere (to make) and diventare (to become) are the hyperonyms of many verbs, in whose definitions they occur together with adjectives indicating the result of the change (e.g., abbellire (to prettify) is defined as “rendere bello” (to make beautiful), and ingrassare (to get fat) is defined as “diventare grosso” (to become fat)). Since the (eventual) opposite adjective is usually indicated in the definitions of adjectives, a procedure could be developed which retrieves all the oppositions between pairs of adjectives and searches, then, for the same oppositions in the definitions of the verbs within the taxonomies of the two genus verbs under analysis. Such a procedure would return, e.g., a verb like imbruttire (to uglify), defined as “rendere brutto” (to make ugly), as antonym of the above mentioned abbellire, and dimagrire (to slim), defined as “diventare magro” (to become slim), as antonym of ingrassare. Similar methodologies of automatic analysis could be used for retrieving verbs whose antonymy is caused by the opposition of incorporated adverbs or prepositions (e.g., entrare = andare dentro (to go in) / uscire = andare fuori (to go out)). Since it has been noted that sometimes there can be opposition only between two members of different synsets rather than between the whole synsets (this has been noted in particular for nouns, cf. D005), within the database also antonymy between synset members will be allowed, besides antonymy between synsets, called, then, ‘near-antonymy’. 2.4.1 ANTONYMY In the following general formal criteria for the definition of antonymy between single members of synsets are given. Verb test (10) Comment: Score yes a yes b Conditions: Effect: LE2-4003 Antonymy between single variants in verb synsets (a) Test sentence If something/someone/it Xs then something/someone/it does not Y If something/someone/it Ys then something/someone/it does not X - X is a synset variant in the third person singular form - Y is a synset variant in the third person singular form - X and Y are members of co-hyponym synsets X ANTONYM Y EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs Verb test (11) Comment: Score yes a yes b Conditions: Effect: Verb test (12) Comment: Score yes a yes b Conditions: Effect: 28 Antonymy between single variants in verb synsets (b) Test sentence If something/someone/it Xs then something/someone/it does not Y If something/someone/it Ys then something/someone/it does not X - X is a synset variant in the third person singular form - Y is a synset variant in the third person singular form - there is a hyperonym of X which is opposite to a hyperonym of Y X ANTONYM Y Antonymy between single variants in verb synsets (c) Test sentence If something/someone/it Xs then something/someone/it does not Y If something/someone/it Ys then something/someone/it does not X - X is a synset variant in the third person singular form - Y is a synset variant in the third person singular form - the situation referred to by X has an addressee and the addressee is the protagonist of the situation referred to by Y X ANTONYM Y 2.4.2 NEAR_ANTONYMY In the following general formal criteria for the definition of antonymy (‘near-antonymy’) between verb synsets are given. Verb test (13) Comment: Score yes a yes b Conditions: Example: a b Effect: LE2-4003 Antonymy between verb synsets (a) Test sentence If something/someone/it Xs then something/someone/it does not Y If something/someone/it Ys then something/someone/it does not X - X is a verb in the third person singular form - Y is a verb in the third person singular form - X and Y are co-hyponym synsets If he gets fat then he does not get thin If he gets thin then he does not get fat {to get fat, to put on weight} NEAR_ANTONYM {to get thin, to lose weight} EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs Verb test (14) Comment: Score yes a yes b Conditions: Example: a b Effect: Verb test (15) Comment: Score yes a yes b Conditions: Example: a b Effect: LE2-4003 29 Antonymy between verb synsets (b) Test sentence If something/someone/it Xs then something/someone/it does not Y If something/someone/it Ys then something/someone/it does not X - X is a verb in the third person singular form - Y is a verb in the third person singular form - there is a hyperonym of X which is opposite to a hyperonym of Y If he sells then he does not buy If he buys then he does not sell {to sell, to exchange for money} NEAR_ANTONYM {to buy, to purchase, to take} Antonymy between verb synsets (c) Test sentence If something/someone/it Xs then something/someone/it does not Y If something/someone/it Ys then something/someone/it does not X - X is a verb in the third person singular form - Y is a verb in the third person singular form - X has an addressee and the addressee is the protagonist of Y If he gives then he does not take If he takes then he does not give {to give} NEAR_ANTONYM {to take, to take away} EuroWordNet 30 EuroWordNet D006: Definition of the links and subsets for verbs 2.4.3 XPOS_NEAR_ANTONYMY As said above, also antonymy between different POS is allowed, as in the cases of synonymy and hyperonymy/hyponymy. However, this antonymy relation is only allowed between synsets (and not variants). Here below, some of the criteria for its identification are provided, i.e., those taking into consideration co-hyponym antonyms; for the other cases, cf. the tests given for near-antonymy. Verb test (16) Comment: Score yes a yes b Conditions: Example: a b Effect: Verb test (17) Comment: Score yes a yes b Conditions: Example: a b Effect: LE2-4003 Antonymy between verbs and nouns denoting events or processes Test sentence If something/someone/it Xs then a/an Y does not take place If a/an Y takes place then something/someone/it does not X - X is a verb in the third person singular form - Y is a noun in the singular - X and Y are (XPOS) co-hyponyms If someone falls asleep then awakening does not take place If awakening takes place then someone does not fall asleep {to fall asleep} (X) XPOS_NEAR_ANTONYM {awakening} (Y) Antonymy between verbs and nouns denoting states Test sentence If something/someone/it Xs then something/someone/it is not in a state of Y If something/someone/it is in a state of Y then something/someone/it does not X - X is a verb in the third person singular form - Y is a noun in the singular - X and Y are (XPOS) co-hyponyms If someone loves someone then someone is not in a state of hate If someone is in a state of hate then someone is not loving {to love} (X) XPOS_NEAR_ANTONYM {hate} (Y) EuroWordNet 31 EuroWordNet D006: Definition of the links and subsets for verbs Verb test (18) Comment: Score yes a yes b Conditions: Example: a b Effect: Antonymy between verbs and adjectives (or adverbs) Test sentence If something/someone/it Xs then something/someone/it is not Y If something/someone/it is Y then something/someone/it does not X - X is a verb in the third person singular form - Y is an adjective - X and Y are (XPOS) co-hyponyms If someone sleeps then someone is not awake If someone is awake then someone does not sleep {to sleep} (X) XPOS_NEAR_ANTONYM {awake} (Y) 2.5 Involved As mentioned above (cf. § 2.1), to avoid ‘shallow hierarchy structures’, additional relations, linking verbs to other parts of speech, were defined in order to encode information which can be found in the differentiae of dictionary definitions, especially with respect to verbs, and which can be very useful from the viewpoint of the EuroWordNet system user. One of these relations has been named ‘Involved’ (and its counterpart ‘Role’) and indicates a link between a verb and a noun (or, in some cases, an adjective or an adverb - cf. footnote 7) whose meaning is ‘incorporated’ in, or connected with, the meaning of the verb itself. The subtypes of the relation are reported again in the following: INVOLVED involved-agent involved-patient involved-instrument involved-location involved-direction involved-source-direction involved-target-direction The importance of coding information of this kind can be inferred also by an analysis of the most recent theoretical developments in lexical semantics. In fact, research in this field has demonstrated that there is often a direct connection among the arguments lexicalised within a verb root and its syntactic properties (cf., e.g., Levin 1993). Furthermore, it has been shown that, since (groups of) languages display different ‘preferences’ for patterns of lexicalisation of semantic components in verb roots (Talmy 1985), in order to develop LE systems these should be taken into due consideration (cf. Sanfilippo et al. 1992 for a description of the use of such data for MT). The specific subrelations chosen, then, have not ‘arbitrarily’ been chosen, but are the most prominent ones to describe the different semantic references involved in the meaning of a word. Indeed, besides providing the LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 32 possibility of encoding data on involved agents and patients, which are clearly relevant with respect to every class of verbs, these links both allow coding data on every kind of means used to perform an action (in fact ‘instrument’ is used here in a wide sense), and assume Gruber (1976) view according to which the semantics of motion and collocation can be seen as providing an interpretation for many semantic classes of verbs (besides the motion ones), e.g. verbs of bringing, saying, giving. Since the subtypes of the relation will be used to encode data on arguments strongly ‘involved’ in the meaning of a verb, and generally indicated within dictionary definitions, the labels used, although already ‘familiar’ to linguists because often associated with ‘thematic roles’, should not be understood as referring to the arguments which a verb can take, i.e. with the arguments co-occurring with it in a sentence. Indeed, the whole database is being built to encode data on the semantic content of words, described by means of a ‘relational’ approach, and our links will encode (only) data on the semantic features of arguments incorporated in the meaning of a verb, which certainly determine also the kind of arguments which a verb allows as its complements, but which do not exactly coincide with them. For instance, whereas in the meaning of a verb like to move, allowing both agent and patient arguments, there is no inherent reference either to a particular ‘involved-agent’ or to an ‘involved-patient’ (because, in fact, many kind of ‘objects’ can move), which therefore should not be encoded, within the meaning of a verb like sgambettare (an Italian verb meaning to kick one’s legs about and only used to refer to a movement performed by babies) there is an incorporated ‘agent-protagonist’ of the event that will be encoded by means of the relation INVOLVED_AGENT --> babies. Thus, according to our proposal, the various subtypes of the main relation should receive a value only in case there is a clear reference to a particular link within the meaning of a verb. As a matter of facts, as it happens for all the relations, also this link has a correspondent reverse relation which connects nouns to verbs and is called ‘ROLE’ relation (cf. Deliverable D005). Thus, although only the connection between a verb and an argument strongly involved in its meaning should be encoded by means of the INVOLVED relation, whenever a noun will be linked to a verb by means of the ROLE relation also the reverse relation will automatically be derived (but, in this case, a ‘reversed’ label will be added). For instance, the verb to hammer inherently involves only the noun hammer, which will be linked to it by means of the INVOLVED_INSTRUMENT relation. Though, the noun carpenter can be connected with the verb to hammer by means of the ROLE_AGENT relation, so that the correspondent link from the verb to the noun (i.e., to hammer --> INVOLVED_AGENT --> carpenter) could be automatically derived. In this case, as also in the case of all the ‘default’ reversals of relations, we shall automatically insert the label ‘reversed’ and allow disjunction of agents, in order to ensure the possibility of having multiple (possible) agents connected with a verb. In the following, we provide general criteria for the identification both of the main relation and of its subtypes. The test for the general relation (which should also apply to any subtype) is the following: Verb test (19) Comment: Score LE2-4003 Involvement relation in general Test sentence EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs yes a Conditions: Example: Effect: a 33 (A/an) X is the one/that who/which is involved in Y - X is a noun - Y is a verb in the gerundive form An hammer is that which is involved in hammering {to hammer} (Y) INVOLVED {hammer} (X) The next tests can then be used to elicit more specific involvements: Verb test (20) Comment: Score yes a Conditions: Example: Effect: a Verb test (21) Comment: Score yes a Conditions: Example: Effect: a Agent Involvement Test sentence (A/an) X is the one/that who/which does the Y - X is a noun - Y is a verb in the gerundive form A teacher is the one who does the teaching {to teach} (Y) INVOLVED_AGENT {teacher} (X) Patient Involvement Test sentence (A/an) X is the one/that who/which undergoes the Y - X is a noun - Y is a verb in the gerundive form A learner is the one who undergoes the learning {to learn} (Y) INVOLVED_PATIENT {learner} (X)11 11 It should be clear that the fact that we encode either an INVOLVED_PATIENT or an INVOLVED_AGENT relation for a pair verb/noun does not indicate that the one encoded is the only possible relation between the two, but only that it is the more directly involved one. In fact, even if a learner undergoes the learning, and is thus the INVOLVED_PATIENT of the verb, he can do it actively, i.e. teaching himself, and be also an agent in this sense. However, the former relation is that strongly implied by the meaning of the verb. In any case, the two roles AGENT and PATIENT could also be encoded as a conjunction of the two relations whenever this is necessary (i.e., directly involved in the meaning of a verb). For instance, in the case of verbs referring to motion by the protagonist of the event indicated, the protagonist is at the same time actively and passively involved in the movement referred to by the verb: e.g., a runner is the agent of the process indicated by the verb to run, but he is also the ‘patient’ of the same process. LE2-4003 EuroWordNet 34 EuroWordNet D006: Definition of the links and subsets for verbs Verb test (22) Comment: Score yes a Conditions: Example (1): Effect: Example (2): Effect: Verb test (23) Comment: Score yes a Conditions: Example: Effect: LE2-4003 a Instrument Involvement (also valid for nouns indicating things, used for doing something, which cannot be really called ‘instruments’) Test sentence (A/an) X is either i) the instrument that or ii) what is used to Y (with) - X is a noun - Y is a verb in the infinitive form An hammer is the instrument that is used to hammer {to hammer} (Y) INVOLVED_INSTRUMENT {hammer} (X) A sailing boat is what is used to sail with {to sail} (Y) INVOLVED_INSTRUMENT {sailing boat} (X) Location Involvement Test sentence (A/an) X is the place where the Y happens - X is a noun - Y is a verb in the gerundive form A school is the place where the teaching happens {to teach} (Y) INVOLVED_LOCATION {school} (X) EuroWordNet 35 EuroWordNet D006: Definition of the links and subsets for verbs Verb test (24) Comment: Score yes a Conditions: Example: a Effect: Verb Test (25) Comment: Score yes a Conditions: Example: Effect: a Direction Involvement12 Test sentence It is possible to Y from/to a place - Y is a verb in the infinitive form It is possible to run from/to a place {to run} (Y) INVOLVED_DIRECTION {place} (X) Source-Direction Involvement Test sentence (A/an/the) X is the place from where Ying happens / one Ys - X is a noun - Y is a verb The start is the place from where the racing happens {to race} (Y) INVOLVED_SOURCE {the start} (X) 12 In the same way as there is the possibility to identify a generic INVOLVED relation for all those cases of undecidable kind of involvement, there is the possibility of indicating the existence of a generic involved direction in the meaning of a verb. Indeed, some verbs (at least in Italian) inherently refer to change-of-position (i.e., motion from/to a non-specified place), without, actually, being inherently explicit with respect to the exact direction of motion. E.g., correre (to run) displays two main senses (indicated by different auxiliary selection): one of undirected motion (2a below) and one of directed motion (2b). The specific direction of the running is however non-specified within the meaning of the verb and can thus be variously specified by its complements: 2a. Ho corso per un'ora / * a casa ('I HAVE run for one hour / * (to) home') 2b. Sono corso a casa / via da casa ('I BE run away from home / home') In cases like these we can state an INVOLVED_DIRECTION relation with the noun luogo (place), without indicating a relation with a more specific noun. The INVOLVED_DIRECTION relation is thus useful in order to distinguish both among different kinds of incorporation in a language (e.g., the Italian verb nuotare (to swim) has no INVOLVED_DIRECTION) and among differences of lexicalisation across languages (e.g., to swim has a generic INVOLVED_DIRECTION), thus satisfying the need for the identification of differences in lexicalisation patterns emphasized within the TA, where it is stated that [this resource] “will enhance the fundamental understanding of lexicalisation patterns across languages which will be crucial for machine translation and language learning systems” (p. 17). LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs Verb Test (26) Comment: Score yes a Conditions: Example: Effect: a 36 Target-Direction Involvement13 Test sentence (a/an/the) X is the place to which Ying happens / one Ys - X is a noun - Y is a verb The ground is the place to which one collapses/falls heavily {to collapse, to fall heavily} (Y) INVOLVED_TARGET_DIRECTION {ground} (X) Mono- and bi-lingual dictionaries both contain relevant information connected with the INVOLVED relation; although the data related with typical arguments conflated within the meaning of a verb are generally found in definitions, also example sentences and semantic indicators may be used to acquire them. Furthermore, useful information is not only provided by verb entries: indeed, it is sometimes possible to use information found within the entries of nouns related to verbs. For instance, in the definitions of instruments there is generally the indication of the action that they are used for. Since it is not always possible to identify patterns uniquely connected with one subtype of the relation, the acquisition procedure cannot be fully automatic: in fact, heuristics have to be developed which return an ‘intermediate’ value for a relation-subtype, and which always require a manual verification (see Montemagni 1995 on the problems of the automatic identification of the thematic role associated with verb arguments, and Alonge 1992a, 1992b for proposals related to the development of heuristics for extracting such data). By adding in the EWN model such a relation we give the possibility to code much information (generally found in dictionaries) which is useful to outline the semantics of a verb. Probably these relations will be only partially given values in our database (for the time being), however we think that it is necessary to have the possibility to add information of this kind (for instance, probably we are going to code some data of this kind for those taxonomies for which we had extracted them in Acquilex, but we do not think we shall be able to extract these relations for all the verbs which will be taken into consideration). 2.6 ‘Causes’ and ‘Is_Caused_By’ 13 As already claimed, all the INVOLVED relations may help coding data on the similarities/differences in lexicalisation patterns across languages. As far as the latter two are concerned, for instance, we can see that the Italian verb sbarcare displays an INVOLVED_SOURCE_DIRECTION relation with barca and has a direct corresponding verb in English: to disembark related to the INVOLVED_SOURCE_DIRECTION boat. On the other hand, the Italian verb rincasare (to go back home), connected with the noun casa by means of the relation INVOLVED_TARGET_DIRECTION, has no direct correspondent in English. LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 37 The causal relation is used in WN 1.5 to indicate the relation between two verbs one of which referring to an event causing a resulting event, process or state referred to by the second verb (like in the case of show/see, fell/fall, give/have). The relation can be identified by means of the following general test: He Xs causes He Ys. Which means that the following sentences must be respectively true and false before two verbs can be connected by means of a causal relation: - He Vs1 causes Vs2 yes - He Vs2 causes Vs1 no Clear yes, no = causal relation. Whereas in WN 1.5 the causal relation only holds between verbs and, furthermore, only should hold (according to what stated in Fellbaum 1993: 54) between verbs referring to temporally disjoint situations, within our database this relation will be used also to link verbs and higher-order nouns denoting events or processes (henceforth ‘dynamic situations’ or dS) to verbs, higher-order nouns, adjectives or adverbs. In addition, we have distinguished among three possible cases of temporal relationship between the (dynamic/non-dynamic) situations indicated by the verbs (/other POSs) linked: • • • a cause relation between two situations which are temporally disjoint: there is no time point when dS1 takes place and also S2 (which is caused by dS1) and vice versa (e.g., in the case of to shoot/to hit); a cause relation between two situations which are temporally overlapping: there is at least one time point when both dS1 and S2 take place, and there is at least one time point when dS1 takes place and S2 (which is caused by dS1) does not yet take place (e.g., in the case of to teach/to learn); a cause relation between two situations which are temporally co-extensive: whenever dS1 takes place also S2 (which is caused by dS1) takes place and there is no time point when dS1 takes place and S2 does not take place, and vice versa (e.g., in the case of to feed/to eat). It is interesting to note that in the last case the causal relation may be sometimes hardly distinguished by the hyponymy relation (cf. the above reported Dutch verbs dichttrekken/sluiten). Thus, it becomes important to have tests which allow to distinguish among the different kinds of temporal relationships between the verbs being related: indeed, in case two words refer to situations which are co-extensive, the two words should also be tested for hyponymy and if this relation applies, we decided that the cause relation should not be expressed. As we have already recalled, then, different types of causality can also be distinguished with respect to the factivity of the effect (following Lyons, 1977): • factive: situation dS1 implies the causation of S2, e.g. “to kill causes to die”; LE2-4003 EuroWordNet 38 EuroWordNet D006: Definition of the links and subsets for verbs • non-factive: situation dS1 probably or likely causes S2 or S1 is intended to cause some situation S2 “to search may cause to find”. Thus, while the differences in the temporal relationships between the situations referred to by the words being linked will only be taken into account by means of different tests formulated to identify the existence of the causal relation, the factivity/non-factivity of the link will be encoded by means of labels applied to the relation itself (cf. above). In the following general formal criteria for the definition of causation relation are provided. Verb test (27) Comment: Score yes a no b Conditions: Example: a b Effect: Factive causation relation between verbs/nouns (general) Test sentence (A/an) X causes (a/an) Y (A/an) X has (a/an) Y as a consequence (A/an) X leads to (a/an) Y the converse of (a) - X is a verb in the infinitive form or X is a noun in the singular - Y is a verb in the infinitive form or Y is a noun in the singular to kill (/a murder) causes to die (/ death) to kill (/a murder) has to die (/ death) as a consequence to kill (/a murder) leads (someone) to die (/ death) *to die / (a) death causes to kill *to die / (a) death has to kill as a consequence *to die / (a) death leads (someone) to kill {to kill} (X) CAUSES {to die} (Y) factive {to die} (Y) IS_CAUSED_BY {to kill} (X) reversed {to kill} CAUSES {death} factive {death} IS_CAUSED_BY {to kill} reversed {murder} CAUSES {to die} factive {to die} IS_CAUSED_BY {murder} reversed {murder} CAUSES {death} factive {death} IS_CAUSED_BY {murder} reversed Obviously, the event of ‘dying’ is not necessarily caused by ‘killing’. This may either follow from the fact that the verb kill refers to one of the possible disjunct causes for die, or it may be expressed by making explicitly labelling the reverse relation “dying IS_CAUSED_BY killing” as reversed (as is done here). The following test is for detecting factive causation relation between verbs (but also holding for nouns referring to non-dynamic or dynamic situations) and adjectives (/adverbs) is provided: Verb test (28) LE2-4003 EuroWordNet 39 EuroWordNet D006: Definition of the links and subsets for verbs Comment: Score yes a no b Conditions: Example: a b Effect: Factive causation relation between verbs and adjectives (or adverbs) Test sentence X causes to be Y X has being Y as a consequence X leads to be(ing) Y the converse of (a) - X is a verb in the infinitive form - Y is and adjective to kill causes to be dead to kill has being dead as a consequence to kill leads someone to be dead *to be dead causes to kill *to be dead has to kill as a consequence *to be dead leads (someone) to kill {to kill} (X) CAUSES {dead} (Y) factive {dead} (Y) IS_CAUSED_BY {to kill} (X) reversed Non-factivity is more evident in the following tests: LE2-4003 EuroWordNet 40 EuroWordNet D006: Definition of the links and subsets for verbs Verb test (29) Comment: Score yes a no b Conditions: Example: a b Effect: Non-factive causation relation between verbs (/nouns) using a modal auxiliary14 Test sentence (A/an) X may cause (a/an) Y (A/an) X may have (a/an) Y as a consequence (A/an) X may lead to (a/an) Y the converse of (a) - X is a verb in the infinitive form or X is a noun in the singular - Y is a verb in the infinitive form or Y is a noun in the singular to search may cause to find to search may have to find as a consequence to search may lead (someone) to find ?to find may cause to search ?to find may have to search as a consequence ?to find may lead (someone) to search {to search} (X) CAUSES {to find} (Y) (non-factive) Dis {to find} (X IS_CAUSED_BY {to search} (Y) (non-factive) Dis This test is exactly the same as the previous one except for the inserted modal verb may which expresses non-factivity of the cause. Non-factivity here can be indicated by adding the label ‘Dis’, indicating that either the process indicated by to search may lead to the event indicated by to find or to some other contrary result. In the case a pair of verbs does not pass the first test on factive causality it may still pass a non-factive test. Some other modal constructions are used in the tests provided in the Appendix for each language. The tests provided so far are general tests to identify causal relation. However, we have claimed above that some more specific tests are needed, which emphasize the eventually different temporal reference of the words related by the causal link, in order to distinguish among cases of ‘genuine’ causal relation and cases in which we should not express the causal relation, but rather an hyponymy relation. The following test elicits ‘genuine’ cause relation: 14 A similar test can be used for the relation to adjectives (or adverbs), and this also holds for the tests provided below. LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs Verb test (30) Comment: Score yes a no b Conditions: Example: a b Effect: 41 Causation relation between verbs (/nouns) referring to temporally disjoint situations Test sentence If (a/an) X takes place it causes/may cause (a/an) Y to take place afterwards/later on the converse of (a) - X is a verb in the gerundive form or X is a noun in the singular - Y is a verb in the gerundive form or Y is a noun in the singular If sending takes place it causes receiving to take place later on * If receiving takes place it causes sending to take place later on {to send} (X) CAUSES {to receive} (Y) factive {to receive} (Y) IS_CAUSED_BY {to send} (X) reversed In case two words do not pass this test, but pass the following temporally-non-disjoint test, it is necessary to test them for the hyponymy relation: if the latter applies, then we shall not express the cause relation. Verb test (31) Comment: Score yes a no b Conditions: Example: a b Effect: Causation relation between verbs (/nouns) referring to temporally non-disjoint situations Test sentence If (a/an) X takes place it causes/may cause (a/an) Y to take place at the same time the converse of (a) - X is a verb in the gerundive form or X is a noun in the singular - Y is a verb in the gerundive form or Y is a noun in the singular - X and Y are not connected by means of the hyponymy relation If pulling takes place it may cause opening to take place at the same time ? If opening takes place it may cause pulling to take place at the same time {to pull} (X) CAUSES {to close} (Y) (non-factive) {to close} (Y) IS_CAUSED_BY {to pull} (X) (non-factive) A further issue is worth noting here. We have stated that dynamic situations may cause both other dynamic situations or non-dynamic ones. Whether the caused result is a dynamic event/process or a (non-dynamic) state can be inferred from the hyponymy relation of the result with dynamic/non-dynamic hyperonyms (e.g., state or change). This can be clarified by the following example: LE2-4003 EuroWordNet 42 EuroWordNet D006: Definition of the links and subsets for verbs i) ii) fall asleep V fall asleep V fall asleep V sleep V sleep N asleep A CAUSES CAUSES HAS_HYPERONYM HAS_HYPERONYM XPOS_NEAR_SYNONYM XPOS_NEAR_SYNONYM addormentare (make sleep) V addormentarsi (fall asleep) V addormentare (make sleep) V sleep V sleep N, change V be V sleep V sleep V asleep A CAUSES addormentarsi (fall asleep) V HAS_HYPERONYM cambiare (change) V HAS_HYPERONYM fare (make, cause) V In i) we see that the situation referred to by the verb to sleep is non-dynamic, as is expressed by its hyponymy relation with the verb to be. We also see that the noun sleep and the adjective asleep have near-synonymy relations with the verb sleep and are, therefore, non-dynamic. In ii) we see an example in which the Italian verb addormentarsi (to fall asleep) is caused by addormentare (to make sleep). The fact that we are dealing with two dynamic situations is again expressed by the hyponymy relation. From the hyponymy relation may be also inferred that addormentarsi is a ‘non-controlled’ process and addormentare is a ‘controlled’ action, since their hyperonyms display those characteristics. Using the combination of CAUSE and HYPONYMY relations, we can thus encode many different kinds of information, while we have a maximum of flexibility to relate words across the different parts-of-speech. The latter is very important because we will come across situations in which words from different parts-of-speech are CAUSES or which are results-of-CAUSES and also have synonymy relations or cross-linguistic equivalence relations (are linked to the same ILI-record) and therefore should also have the same CAUSE-relations. Dictionary definitions represent the main information source in the case of the causal relation. Different patterns are used in dictionaries to convey this kind of relation, all hinging on the use of specific constructions containing verbs indicating ‘cause’. In Italian dictionaries, like in Spanish, for instance, the most commonly used pattern is represented by the periphrastic causative construction with far(e) - in Spanish, hacer - (lit. ‘make’, ‘cause’) followed by a verb in its infinitival form: e.g., the verb accagliare (to curdle) in its transitive reading is defined as far coagulare (to cause to coagulate). This construction indicates that the verb being defined is the causative counterpart of the infinitive verb which follows far(e); in the example at hand, accagliare is related via causation to the action referred to by coagulare ‘coagulate’. Other more complex patterns involve the simultaneous consideration of different definitions of the same verbal headword; this is the case of verbs undergoing both causative and inchoative uses such as break. LE2-4003 EuroWordNet 43 EuroWordNet D006: Definition of the links and subsets for verbs 2.7 Has_Subevent and Is_Subevent_Of According to Fellbaum (193: 45), “the different relations that organize the verbs can be cast in terms of one overarching principle, lexical entailment”. Two basic kinds of lexical entailment can be distinguished: one involves ‘temporal inclusion’ (the two situations referred to by the verbs in the relation partially or totally overlap); the other involves ‘temporal exclusion’ (the two situations are variously temporally disjoint). These temporal relationships between verbs are taken as the basis for a further distinction of four kinds of entailment: a. + Temporal Inclusion a.1 co-extensiveness (e. g., to limp/to walk) a.2 proper inclusion (e.e., to snore/to sleep) b. - Temporal Exclusion b.1 backward presupposition (e.g., to succeed/to try) b.2 cause (e.g., to give/to have) Thus, within WN 1.5 class (a1) verbs are linked by means of the hyponymy (called - for verbs ‘troponymy’) relation; (b2) verbs are linked by means of a causal relation; and both (a2) and (a3) verbs are linked by means of a generic ‘Entailment’ relation. That is, the latter relation is applied to those cases that do not fall within the more specific classes of verbs linked by means of hyponymy or cause relations. Within our project we decided to encode data related to the WN 1.5 entailment relation in a different way. In fact, it seems to us that the ‘backward presupposition’- entailment relation can be expressed by using the causal relation in conjunction with the factivity label. For instance, as far as to succeed and to try are concerned, we may state the following relation: {to succeed} IS_CAUSED_BY {to try} CAUSES {to try} Factive {to succeed} Non-factive On the other hand, as far as verbs related by means of ‘proper inclusion’- entailment are concerned, we can use a link indicating more clearly the existence of this temporal relationship between the words involved, instead of the more general relation of ‘entailment’, which, as we have seen, is ambiguous since every relation is a kind of entailment relation. In fact, in these cases we may speak of a kind of verb-meronymy relation (cf. D005 for noun-meronymy) and describe it by means of the link HAS_SUBEVENT/IS_SUBEVENT_OF. Thus, we shall have: {to snore} {to sleep} {to buy} {to pay} LE2-4003 IS_SUBEVENT_OF HAS_SUBEVENT HAS_SUBEVENT IS_SUBEVENT_OF {to sleep} {to snore} {to pay} {to buy} reversed reversed EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 44 In the following, general criteria for the definition of the HAS_SUBEVENT relation between verbs (/nouns referring to events or processes) are provided: Verb test (32) Comment: Score yes a no b Conditions: Example: a b Effect: Verb test (33) Comment: Score yes a no b Conditions: Example: a b Effect: Has_Subevent/Is_Subevent_of relation between verbs (/nouns) (a) Test sentence Y takes place during or as a part of X, and whenever Y takes place, X takes place the converse of a) - X is a verb in the gerundive form - Y is a verb in the gerundive form Snoring takes place during or as part of sleeping, and whenever snoring takes place, sleeping takes place *Sleeping takes place during or as part of snoring *Whenever sleeping takes place, snoring takes place {to snore} (X) IS_SUBEVENT_OF {to sleep} (Y) {to sleep} (Y) HAS_SUBEVENT {to snore} (X) reversed Has_Subevent/Is_Subevent_Of relation between verbs (/nouns) (b) Test sentence X consists of Y and other events or processes the converse of a) - Y is a verb in the gerundive form - X is a verb in the gerundive form buying consists of paying and other events or processes *paying consists of buying and other processes {to buy} (Y) HAS_SUBEVENT {to pay} (X) {to pay} (X) IS_SUBEVENT_OF {to buy} (Y) reversed Just as with meronymy for nouns (see D005) there are subevents that can only occur as being embedded within one type of a larger situation and there are subevents that may occur in different embedding situations. Furthermore, there may be embedding situations for which the subevent is a necessary step or phase and there may be embedding situations for which the subevent is optional. Both these possibilities have an effect on the implication relation between the events. Since “sleep” does not necessarily imply “snore” but “snore” always implies the embedding event “sleep” the implication direction is from “snore” to “sleep” (and not the other way around from the embedding event to the subevent)). However, since “buy” necessarily implies “pay” but “pay” can also be a subevent of another embedding event, the implication relation is here from the embedding event to the subevent (from “buy” to “pay”). This difference has however nothing to do with the subevent relation as such but more the dependentness of the events. In the SUBEVENT relation it may be expressed by the label reversed or by disjunction of the relation with another subevent or another embedding event. LE2-4003 EuroWordNet EuroWordNet D006: Definition of the links and subsets for verbs 45 Nevertheless, one has to be aware that these implicational effects may interfer with the clarity of the intuitions aroused in the above tests. Our main information sources, i.e. the machine readable dictionaries, do not always provide relevant information with respect to this relation: e.g., Italian sources do it only sometimes and often in an implicit form, while Dutch ones do provide useful data. 2.8 Be_In_State and State_Of This relation has been created mainly to encode links between nouns that refer to anything being in a particular state expressed by an adjective(/adverb). These nouns often have an open denotation: i.e. they can refer to any entity to which the state applies, e.g. “the poor” refers to all entities which are in a “poor” state. Note that these nouns are not equivalent to the states: the entities which have the property “poor” are not states but normal first-order entities. Although we have foreseen the possibility of encoding this relation also between verbs and adjectives/adverbs, it seems a more important relation with respect to (first-order) nouns and therefore is described more carefully in D005. LE2-4003 EuroWordNet 46 EuroWordNet D006: Definition of the links and subsets for verbs 3. The definition of the verb subsets According to the TA, a second basic issue to be dealt with within Task 4.1, assigned to WP4, was the definition of the subsets of verbs for which relations will be encoded in the database. Thus, first of all we have aimed at identifying a set of common ‘Basic Concepts’ for each language wordnet, by taking into consideration the defining vocabulary of each dictionary. Secondly, criteria have been clearly defined to extend this first subset in order to obtain the about 15,000 verb synsets to be encoded for each wordnet, as indicated within the TA. The final vocabulary covered in EuroWordNet should have the following characteristics:15 1) include all the concepts needed to relate other concepts (i.e., no semantic domain should be isolated); 2) display high degree of overlapping across the languages; 3) include most frequent words in every language use (identified by means of researches on textual corpora); 4) contain language-specific lexicalization patterns; 5) contain some specific terminology, to be used to test the ability of the EWN database to be expanded in depth; 6) include all the Parole words, in order to facilitate the parallel usage of both databases. The general methodology identified to achieve this coverage has been to start with the identification of ‘base concepts’ (for a discussion of general issues connected with this identification cf. also D005). This should enable us to focus manual effort on base concepts, and use automatic techniques for the extensions. In order to reach maximal consensus, we are using the test and communication for the manual work on the base concepts, and we are exchanging and coordinating the definition patterns and extraction heuristics. After the base concepts will have been completely identified, the selections will be extended by each group making sure that the criteria 1) through 6) are met. The exact procedure for extending the base concept selection may vary from site to site. 3.1 On selecting ‘Base Concepts’ In order to identify a set of ‘Basic Concepts’, we have taken into consideration not simply words at the top of hierarchies extracted from individual sources, but words more generally felt to refer to ‘central’ concepts. Their ‘centrality’ is however reflected both in a high position in the taxonomies extracted from each language sources, and in a high number of relations displayed. Because of their nature, though, base concepts are generally highly polysemous, with low-quality definitions in 15 Since the general issues dealt with in this section are exhaustively discussed in D005, here we are going to provide only some necessary information on motivations and techniques for the identification of base concepts. LE2-4003 EuroWordNet 47 EuroWordNet D006: Definition of the links and subsets for verbs dictionaries; on the other hand, they should be (ideally) ultimately linked to any synset in the database. This will determine to focus manual effort on them in order to establish accurate relations, while in a second stage of our work, we should be able to encode relations for the other subset chosen by using (semi-)automatic techniques. Aiming at reaching a common nucleus of Basic Concepts, we have decided to choose, for each language, those top-verbs displaying a number of frequencies as genus verbs allowing to cover at least the 15% of the entire source used by each group. To the first set of top-verbs we have then added all their hyperonyms which were not already part of the selection. Our aim is obtaining high-quality first subsets with a maximal overlapping between them, both performing manual verifications and interchanging subsets between sites. Once this goal is achieved, it is possible to add other (more specific) areas of the vocabulary, where automatic extraction methods are more reliable. 3.1.1 Selection of the Dutch subset The Van Dale database Vlis is different from the traditional MRDs in that particular semantic relations between senses are already explicitly coded. We therefore can make use of some other criteria than the other sites. The following relations are in the Vlis database: Semantic Relations in the Vlis database ABBREVIATION ANTONYM ASSOCIATIVE CAUSATIVE HYPERONYM INCHOATIVE PARTITIVE REFERENCE SYNONYM WOMAN PREFERENCE FORM_VARIANT SYMBOL Verb Senses 0 148 557 5 8231 11 0 93 5049 0 108 0 0 The database does not form a closed wordnet in which all relations are unified in a single tree or a small set of tops. In fact, the Vlis database contains 298 ends for the verbs. The reason for making a sense a top or end in the hierarchy are often not well-motivated (the database is still in development). In many cases a sense is an end because the meaning was too complex (e.g. many higher-order-nouns referring to states, conditions, events) or information was of an encyclopaedic nature (names of places, LE2-4003 EuroWordNet 48 EuroWordNet D006: Definition of the links and subsets for verbs people, etc..). We therefore did not just take the Vlis tops as a starting point for selecting the base concept but we considered the following criteria to be important in order of appearance: • • • • number of relations position in hierarchy Vlis top senses frequency In total 284 verb senses have been selected. In the next sections we explain how this selection was achieved. • First criterion: number of relations Inspection of the Vlis database leads to the conclusion that the major criterion for being a base concept seems to be the number of relations a sense has.16 To get a sensible and manageable first selection in a consistent way, we first selected the senses with the most-frequent-relations making up around 15% of the total amount of relations. For this purpose we constructed a table with the cumulative-percentage of all relations, ranking the senses per number of relations. This is shown in the next table (Table 1). 16 At this stage all relations have been taken into account to constitute the importance of a concept. LE2-4003 EuroWordNet 49 EuroWordNet D006: Definition of the links and subsets for verbs Table 1 Number Relations 0 1 2 3 4 5 6 7 8 10 11 12 13 14 15 16 17 18 19 ...... 514 28605 1to9 >9 1to10 >12 1to15 >15 of Number of Senses Number of Links 0 9709 1998 812 477 281 209 155 107 82 59 31 28 20 35 25 15 13 11 ...... 0 14143 13748 395 13920 237 14003 140 0 9709 3996 2436 1908 1405 1254 1085 856 820 649 372 364 280 525 400 255 234 209 ...... 0 28889 Percentage of total Cumulative number of links percentage of total number of links 0,00% 33,61% 13,83% 8,43% 6,60% 4,86% 4,34% 3,76% 2,96% 2,84% 2,25% 1,29% 1,26% 0,97% 1,82% 1,38% 0,88% 0,81% 0,72% ...... 0,00% 100,00% 0,00% 33,61% 47,44% 55,87% 62,48% 67,34% 71,68% 75,44% 78,40% 81,24% 83,49% 84,77% 86,03% 87,00% 88,82% 90,20% 91,09% 91,90% 92,62% ...... 100,00% 0,00% 2,79% 1,68% 0,99% All verb senses with more than 12 relations resulted in 237 senses (1,68% of all senses) whose relations cover circa 15 % of all verb senses. We realised that senses of this first selection may have hyperonyms higher up in the hierarchy that do not have that many relations, but still can be base concepts. Therefore we generated all the hyperonym chains up from the senses of this first selection and extended this first selection with these senses. So our second selection contained all verbs with more than 12 relations united with all the hyperonyms of these senses up in the taxonomy. The results are in table 2: LE2-4003 EuroWordNet 50 EuroWordNet D006: Definition of the links and subsets for verbs Verb senses > 12 relations 237 hyperonyms 120 union 285 table 2 • Second criterion: position in hierarchy This last selection still contained many senses that we would not consider as base concepts. Another important criterion seemed to be at which level in the hierarchy a sense is positioned. The higher a sense is located in the hierarchy the more likely it is that it is a base concept . In order to determine the level of a sense we generated a top down taxonomy starting from the top senses for verbs. As downward relations, relations that imply a different degree of specificity, we used the hyponym and verb relations (the verb-relation is used to link compound verbs to their base form, which can be seen as a form of hyponymy) for the verb taxonomy. In general, relations that do not imply a different degree of specificity like synonymy, antonymy and causative could be disregarded for the levels of the hierarchy. However in the Vlis database we had to consider these as well because most of these relations, except synonymy, were allowed to have downward relations themselves. Therefore we extended each level with these sideward relations, i.e. for verbs: abbreviation, antonymy, associative, causative, inchoative, preference, form-variant and symbol. We disregarded the reference relation, because it functions as a morphological link between parts of speech. The number of senses at each level downwards and side-wards are in table 3, whereas table 4 contains the number of senses per level, where the levels are united: Total number of Vlis verb senses = 14.143 Number of Vlis verb tops = 298 Level 0 1 2 3 table 3 LE2-4003 number of downwards 298 2360 3740 3653 senses number of sidewards 268 367 480 463 senses number of senses 566 2661 4111 4029 EuroWordNet 51 EuroWordNet D006: Definition of the links and subsets for verbs Union of verb levels levels 0 and 1 0, 1 and 2 0,1, 2 and 3 table 4 number of senses 3146 6539 8977 percentage of all senses 22,2% 46,2% 63,5% These levels were needed to filter out the non base concepts of our second selection. Again we used the percentage of the total amount of senses to draw a line. The number of verb senses down to level 1 is 22 % of the total number of verb senses. We decided to take the intersection of these levels with our first selection of most-frequently related senses. • Intersections Verbs with more than 12 relations and their hyperonyms = 285 intersected with the verb top down taxonomy levels level 0 down to level 1 down to level 2 down to level 3 table 5 intersection (number of senses) 94 209 268 276 percentage of selection (1165) 32,9% 73,3% 94,0% 96,8% the first The cut-off point was determined by inspecting the lists of included and excluded senses per level for too specific senses. • Relevant Vlis top senses So far we generated base concepts first in a bottom up manner by taking the most-frequently related concepts, extended with their hyperonyms. This selection includes only a fragment of all the ends/tops in the Vlis database. This is shown in the previous table 5, where the intersection of the most-frequently-related senses ad the levels is given. Level 0 here represents the level of the Vlis ends or tops. The following table shows the overlap of the Vlis ends with the selection so far: Not selected Vlis ends total of Vlis ends verbs 298 table 6 LE2-4003 Intersection with selection so far 94 Not selected Vlis ends 204 EuroWordNet 52 EuroWordNet D006: Definition of the links and subsets for verbs Because the Vlis taxonomy is still in development many of these ends are not real top senses, and therefore irrelevant for our subset. To filter out the relevant ones we used the top down hierarchy and counted all the children down from a so-called top sense (at any level). We draw the line at 10 for both nouns and verbs. This resulted in 50 verbal Vlis ends with 10 or more senses related to it (any depth). From these 50 senses, 14 are new verbs which have not been included in the previous selection. The union of these results in the subtotal selection of 223 verb senses. • Finally: addition of frequency At AMS we only have frequency information at the word level. This appeared to be less reliable as a criterion for a basic semantic concept. Therefore, we only included frequent words when they also satisfy one of the other criteria: many-relations or being a Vlis end. This resulted in the addition of another 61 verb senses. The final subset of base concepts then consists of the union of this list with the previous result: 284 verb-senses. • Translation of the Dutch base concepts The above base concepts have been translated to equivalents in WordNet1.5. For each senses one or more translation have been given. The following problematic situations have been encountered when translating the base concepts: 1) a VlisSense is broader and corresponds with several WN1.5 senses. In that case multiple senses are selected. This can either mean that the VlisSense is a hyperonym or that the WN sense-distinctions are too fine-grained. Criteria for deciding that the senses are synonymous are: • if the selected senses have the same hyperonyms or other relations • if the selected senses have the same conceptualization (especially in those cases where the denotation is less important or can vary). Examples: keuze:2 = choice:1 0091731-n {the making of a choice} choice:2 3929489-n {what is choosen} 2) multiple VlisSenses can be linked to a single WN sense (the opposite situation). In this case each VlisSense is linked to the same WN sense. This either suggests that the senses are synonymous (should belong to the same synset) or the WN sense is a hyperonym. Later on, when the wordnets are compared it will turn out that non-synonymous senses are linked to the same ILI-record and a similar choice has to be made as in the previous case. LE2-4003 EuroWordNet 53 EuroWordNet D006: Definition of the links and subsets for verbs Examples: hout: 1 {the substance trees consists of} = 09057553-n wood 4 {the substance and any kind of these substances} houtsoort 1 { the kind of substance } = 09057553-n wood 4 {the substance and any kind of these substances} versiering: 1 {what is used to decorate} = 02029323-n decoration 2 versiersel: 1 {what is used to decorate} = 02029323-n decoration 2 {what is used to decorate} In the case of “wood” we see that the meaning type of wood is lexicalized in Dutch by a separate word. Since any word can refer to some entity or the type of this entity this is not expressed in the English meaning of “wood”. In the case of “decoration” we see that both words in Dutch have exactly the same meaning and thus have to be joined in the same synset as full equivalents. 3) a VlisSense represents an intermediate level between a WN sense and its hyperonym. The WN sense is too specific and the hyperonym is too general. Example: toon: 1 (too narrow) 04471153-n tone (to broad) 04731716-n stem: 2 (too narrow) 03477899-n (too narrow) 04599795-n (to broad) 04731716-n 7 sound voice voice sound 5 1 5 5 In this case both the more specific and the more general WN synset have been linked to the Dutch sense. 4) a VlisSense is equivalent or has a strong relation to a synset with another POS: arme: 1 Noun BE_IN_STATE 01550906-a poor 5 Adjective (poor people) verval: 1Noun XPOS_NEAR_SYNONYM 00122638-v decline 5 Verb vastzitten: 1 Verb LE2-4003 XPOS_NEAR_SYNONYM 00131922-a stuck 1 Adjective EuroWordNet 54 EuroWordNet D006: Definition of the links and subsets for verbs 5) A good clue for equivalence are the sets of hyponyms, meronyms, etc., linked to a VlisSense-WNSense pair. However, there can be two problems here: • the downward relations match but the definition and/or upward relations (the meanings) do not. • the meanings and upward match very well but the hyponyms, meronyms etc.. differ considerably. Here the definitions and upward relations (the true meaning of the sense and not its classifying role) should be taken as a starting point. The difference in classification can be due to other factors. Hyponyms can be classified in different ways (e.g. by function or constitution) or at different levels. The latter is very likely because WN uses many technical intermediate levels. These differences in classifying will follow from the comparison of the wordnets. In some cases the suggested translations seems inappropriate. In that case it helps to browse up and down the taxonomy to find alternatives. More specifically it helps to seen whether the translation of a hyponym is linked to a hyperonym in WN which is more appropriate. Translation-equivalence can thus also be generated in a bottom-up fashion. Another general observation is that the meanings in WordNet1.5 are often differentiated whereas only a single or less meanings are given in the Dutch source. For example, the Vlis only gives one sense for “schoonmaken” (clean) which corresponds with about 4 senses of “clean” in WordNet1.5 which have more or less the same hyperonyms and are only differentiated for the object of cleaning: schoonmaken 1 00023287-v 00109110-v 676 52 clean 1 clean 4 Level01 00106393-v 00881979-v clean 3 clean 6 9 senses of the verb clean Sense 1 clean, make clean -- (make clean by removing dirt, filth, or unwanted substances from; "Clean the stove!"; "The dentist cleaned my teeth") => change, alter -- (cause to change; make different; cause a transformation; "The advent of the automobile may have altered the growth pattern of the city"; "The discussion has changed my thinking about the issue") Sense 2 clean, pick -- (remove unwanted substances from, such as feathers or pits, as of chickens or fruit; "Clean the turkey") => remove, take, take away -- (remove something concrete, as by lifting, pushing, taking off, etc.; or remove something abstract; "remove a threat"; "remove a wrapper"; "Remove the dirty dishes from the table"; "take the gun from your pocket") Sense 3 houseclean, house-clean, clean house, clean -- ("She housecleans every week") => tidy, tidy up, clean up, neaten, straighten, straighten out, square away -- ("Tidy up your room!") => order, bring order to, bring order into LE2-4003 EuroWordNet 55 EuroWordNet D006: Definition of the links and subsets for verbs => arrange, set up => put, set, place, pose, position, lay -- (put into a certain place: "Put your things here"; also with abstract objects and locations: "Place emphasis on a certain point") => move, displace, make move -- (cause to move) Sense 4 cleanse, clean -- (clean one's body or parts thereof, as by washing; "clean up before you see your grandparents"; "clean your fingernails before dinner") => groom, neaten -- (care for the external appearance) => beautify, embellish, prettify -- (make more beautiful) => better, improve, amend, ameliorate, meliorate, make better => change, alter -- (cause to change; make different; cause a transformation; "The advent of the automobile may have altered the growth pattern of the city"; "The discussion has changed my thinking about the issue") Sense 5 clean -- (be cleanable; "This stove cleans easily") => be, have the quality of being -- (copula, used with an adjective or a predicate noun: "John is rich"; "This is not a good answer") Sense 6 clean, strip -- (remove all contents or possession from, or empty completely; "The boys cleaned the sandwich platters"; "The trees were cleaned of apples by the storm"; deprive wholly of money in a gambling game, robbery, etc.; "The other players cleaned him completely") => remove, take, take away -- (remove something concrete, as by lifting, pushing, taking off, etc.; or remove something abstract; "remove a threat"; "remove a wrapper"; "Remove the dirty dishes from the table"; "take the gun from your pocket") Sense 7 clean -- (remove in making clean; "Clean the spots off the rug") => remove, take, take away -- (remove something concrete, as by lifting, pushing, taking off, etc.; or remove something abstract; "remove a threat"; "remove a wrapper"; "Remove the dirty dishes from the table"; "take the gun from your pocket") Sense 8 scavenge, clean, remove unwanted substances from -- (as in chemistry) => remove, take, take away -- (remove something concrete, as by lifting, pushing, taking off, etc.; or remove something abstract; "remove a threat"; "remove a wrapper"; "Remove the dirty dishes from the table"; "take the gun from your pocket") Sense 9 clean -- (remove shells or husks from; "clean grain before milling it") => remove, take, take away -- (remove something concrete, as by lifting, pushing, taking off, etc.; or remove something abstract; "remove a threat"; "remove a wrapper"; "Remove the dirty dishes from the table"; "take the gun from your pocket") Likewise, we see that all the hyponyms of the Dutch “schoonmaken” are scattered over the different senses of “clean” in WordNet. This shows that the possibility of combining the resources using the equivalence relations gives a lot of input for improving relations and sense-distinctions. LE2-4003 EuroWordNet 56 EuroWordNet D006: Definition of the links and subsets for verbs In those cases that there was a complex-equivalence relation we have now simply listed the relevant WN synsets. The exact relation will come out in the next phase where the extracted relations of the Dutch wordnet will be compared with WN1.5. As suggested above some of these relations can be extracted automatically when comparing the wordnets. The file with the Dutch verbal base concepts consist of the following information separated by tabs: • • • • • • • • • • • • the dutch word be\i”nvloeden the dutch sense 1 frequency in corpus 2629 the number of relations 24 the position in the Vlis Hierarchy VlisNoEnd the translation in the bilingual influence,_affect the wordnet file position followed by the POS 00435835-v the wordnet word selected influence the sense number in the AMS database 7 the wordnet file position followed by the POS 01436060-v the wordnet word selected influence the sense number in the AMS database 8 3.1.2 Selection of the Italian subset As far as Italian is concerned, we have begun our search by taking into consideration data encoded on synonymy and hyperonymy/hyponymy within our LDB. 17 In order to identify candidate verbs, we have looked for items i) displaying a high frequency as hyperonyms, ii) displaying a high position in taxonomic hierarchies. The whole process, however, had to be performed together with a manual sense disambiguation of the hyperonyms themselves. Indeed, within our LDB we can only find indications of synonymical or taxonomic relations between a verb entry and a verb homograph, i.e. without any indications of the (verb) sense which is hyperonym (or synonym) of a verb entry; therefore, for every occurrence of a verb as genus, we have to verify the sense in which it is used in the definition under analysis. For instance, the most frequent genus found in our LDB is the verb fare (to do, to make), occurring 374 times as a hyperonym: by analysing randomly selected definitions containing it, we identified at least three fundamental different senses of the verb (mapping, in fact, to three different verb synsets in WN 1.5); in any case, our results, at the present stage, need to be refined by means of a more careful considerations of all the definitions containing fare. In a similar way, after identifying a first set of verbs with a number of frequencies >20, since the whole subset of verb definitions that we had to take into consideration was too great, we only performed a sense disambiguation of a part of the hyperonym verb occurrences, so that the whole data we have obtained 17 Actually, in Pisa a LDB is being used which consists of a set of mono- and bi-lingual machine readable dictionaries integrated with results of various analyses (e.g. taxonomical information) and procedures for merging the information between dictionaries. LE2-4003 EuroWordNet 57 EuroWordNet D006: Definition of the links and subsets for verbs so far should be considered in a preliminary form, and, more important, the available data on frequency refer to the senses of a verb altogether. In any case, this first manual disambiguation performed on the most frequent hyperonym senses has led us to choose some candidate verb senses among them, and i) to connect them with synonyms, thus building synsets; ii) to identify their positions within taxonomic hierarchies which were manually built while performing the sense disambiguation. Actually, the manual work to do in order to carry out these tasks is not trivial, because of the general ‘imprecision’ of our source (and of dictionaries in general). In fact, when disambiguating or building synsets/taxonomies, one has i) to verify the definitions given within the same source for the hyperonym homographs under analysis: this may lead to add, or correct, ‘senses’ in some way; ii) to eliminate problems of circularity, incoherence, inconsistency, etc. Thus, for the time being, among all the hyperonyms whose frequency as genus verbs is >20, we have preliminarily identified 75 undisambiguated verbs, which, after a first manual disambiguation of their senses, seem to correspond to 101 verb senses (as alread remarked, a more careful sense disambiguation is still needed). This resulted in a total set of 4,480 hyponymy relations (number of frequencies of the verbs as hyperonyms) within the LDB, which correspond to about the 30% of all the hyponymy relations which are encoded in our monolingual LDB. Perhaps not all the verb senses identified should be considered as ‘basic concepts’, even if most of them seem to play a central role in the language. However, until a more careful sense disambiguation is not obtained, a definitive list of ‘base concepts’ cannot be developed either. In any case, the final list will certainly contain a higher percentage with respect to the 15% requested as the minimum percentage of relations to be covered with the base concepts. (In the Appendix the complete list of base concepts identified so far is provided.) In a further stage of our work, we tried to map our base concept candidates to WN 1.5, encountering however some different kind of problems: • • • sometimes within WN two/more senses are distinguished which seem to map to one Italian verb sense and viceversa: e.g., the synset constituted by {avere, provare, sentire} seems to match both to sense 1 and to sense 3 of to feel; on the other hand, both senses of diminuire match to sense 1 of to lessen. In these cases, either the one sense in a language, corresponding to multiple senses in the other, can actually be considered a hyperonym sense of them (and a gap remains, due to differences in lexicalisation patterns), or the distinctions in one source are simply too fine-grained; sometimes no corresponding sense was found (e.g., sense 1 of dare); the levels of matching verbs within the two different languages hierarchies are sometimes very different (e.g., while esprimere is top-1 in our source, i.e. it has one level above, the (apparently) corresponding sense 1 of to express in WN is top-4). As it is clear, at the present stage our work can only be considered as ‘work in progress’ and needs to be refined when both the sense disambiguation work and the development of taxonomic hierarchies will be completed. As far as the problem of the mapping with WN is concerned, specific LE2-4003 EuroWordNet 58 EuroWordNet D006: Definition of the links and subsets for verbs choices have to be taken; therefore, the correspondences which we indicate in the section on Italian Base Concepts within the Appendix should be considered also preliminary. 3.1.3 Selection of the Spanish subset For the definition of the first subset of major concepts we have followed a metodology based on both frequency and comparison to WordNet 1.5. For ‘major concepts’ we should understand those which are both more frequently occurring in language and more relevant from a taxonomic point of view. The first requirement is covered by looking at frequencies. The second one by looking at Spanish equivalence of the uppermost nodes in a pre-existing conceptual taxonomy -namely WN1.5. With respect to frequencies we have decided to consider two complementary sources: unrestricted corpus and dictionary used as a corpus. In the first case because obviously high frequencies in corpora accounts for a large use of a word in the language. In the second case, high frequency of occurences in dictionary accounts for a high rate of relevance of a word in the organisation of lexical knowledge. Therefore, the procedure detailed below has been followed: 1.- The most frequent verbs occurring in the corpus of definitions of our available MRD have been extracted. 2.- The most frequent verbs occuring in LEXESP (a tagged 3.000.000 word corpus covering balanced areas and registers of the language) have also been extracted. 3.- Subsets obtained by means of (1) and (2) have been mapped to WN1.5 4.- Those Spanish verbs which mapped to a top-WN1.5 synset (synsets without hypernyms) have been selected -being rejected the rest 5.- Verbs from this resulting subset (4) which do not occur as genuses in the MRD definitions have been also removed 6.- Finally, Spanish synsets have been built by collapsing in them those verbs in (5) which mapped to th the same WN1.5 synset. Summing up, the process of selection of major concepts for Spanish is the following: [(a) + (b)] - [c] = subset [(TOP_WN1.5 & CORPUS_freq_1000+) + (TOPWN1.5 +& MRD_freq_500+)] - [nongenus] Where: (a) is the intersection of: (a.i) the verbs that have a frequency of 1000 ocurrences or over in Lexesp and (a.ii) the verbs that have translation or mapping to some top verb synset of WN1.5. (b) is the intersection of: (b.i) the verbs that have a frequency of 500 ocurrences or over in Vox definitions, and (b.ii) the verbs that have translation or mapping to some top verb synset of WN1.5. (c) are the verbs which not occur as genus in Vox LE2-4003 EuroWordNet 59 EuroWordNet D006: Definition of the links and subsets for verbs The basic set of major concepts described above must be enlarged in the next phase using at least the following criteria: • Adding new verbs appearing as genus terms in definitions of senses in (c) not included in the subset. • Adding grammatical verbs (modal, aspectual and copulative verbs). 3.1.4 First experiment on verifying intersection among the subsets For the time being only a first attempt has been carried out to verify intersection among the subsets. In fact, the final work will be done when the selections are refined and completed. Thus, we have verified the intersection between the Italian subset on one hand and the Dutch or Spanish on the other, by searching the same correspondences with WN 1.5 synsets. Italian has been found to share 37 synsets with Dutch (corresponding to 56 verbs/variants for Italian and 58 verbs/variants for Dutch), and 19 with Spanish (corresponding to 27 Italian variants and 25 Spanish variants). Further research is thus needed in order to reach a higher overlapping, at least between the Italian set and the other two. This work is still in progress and can only be completed after more manual work will have been performed on the base concepts (in particular for Italian). The whole research will be refined within Task 4.2, when also the semantic links for the first subset of verbs will have to be built for each language. 3.2 On extending the subset The identification of a subset of base concepts is only the first step of the definition of the verb subsets to be encoded within the database. Extension criteria have been identified within D001 and have been specified in more details by each site. As far as Italian is concerned, since the wole set of verb definitions within the monolingual LDB amounts to 14,556 and within the TA it is stated that about 15,000 verbs should be encoded in the database, all the LDB verb senses should be encoded. However, perhaps the available data will have to be revised, and, in case it is necessary, somehow enriched, in order to i) improve overlap with other sites, ii) add most-frequent corpus words not already encoded in the LDB. In general, the procedure to be followed to extend the subset is the following: 1) 2) 3) 4) 5) 6) finalize the base concepts extend the base concepts top-down extend the base concepts with most-frequent corpus words which are not part of selection yet extend the base concepts to improve overlap with other sites add typical Dutch, Italian, and Spanish lexicalizations add missing Parole words (in order to facilitate the parallel usage of both databases) LE2-4003 EuroWordNet 60 EuroWordNet D006: Definition of the links and subsets for verbs 7) add terminology (for the addition of terminology we shall consider the terminology that will be made available in the Interval project and the terminology that is included in the text corpora that will be used by Novell to verify the EuroWordNet database). LE2-4003 EuroWordNet 61 EuroWordNet D006: Definition of the links and subsets for verbs 4. Two sample classes encoded: motion verbs and ‘know verbs’ 4.1 On coding the relations for two sample classes of verbs In order to arrive at a final definitions of the relations to be included in our database and to explore the effectiveness of different approaches, methodologies of analysis and solutions for encoding them, we performed an experiment, by taking into account two small semantic fields: i) the taxonomy built starting from the hypernym verbs corresponding to WN 1.5 {move, change position} synset; ii) ‘know’ verbs. The purpose of the experiment was therefore twofold: • • to determine the best approach and methodologies for the different resources to explicitly define the relations and the coverage of the database. With regards to the first point, we needed to verify it since the sources that we are going to use have partially different characteristics and need different methodologies of research. The definition of the relations, then, needed a preliminary analysis of the kind of data, problems, etc., found both by working with the available resources and by trying to match each one’s results. It was then crucial that we agreed at an early stage on the criteria that define the relations and that we anticipated problems and decisions. (Cf. Section 6. in D005 for a more detailed general discussion of issues connected with the extraction process and the encoding of relations.) LE2-4003 EuroWordNet 62 EuroWordNet D006: Definition of the links and subsets for verbs 4.2 First results of work on Dutch The following verb synsets from WN1.5 were chosen for the experiment: WN1.5 synsets verbs MOVE <move, change position> (intransitive) <move, make move, displace> (transitive) <move, locomote, travel, go> (reflexive) Dutch equivalents in Vlis bewegen_1 bewegen_2 bewegen_6 (intr.) (trans.) (refl.) KNOW <know, cognize> (be cognizant or aware of a fact or a specific piece of information; "I know that the President lied to the people"; "I want to know who is winning the game!"; "I know it's time") weten_2 The Van Dale database is different from a traditional Machine Readable Dictionary in that some semantic relations between word senses are already explicitly coded. The Dutch experiment therefore first of all consists of a critical assessment of the data stored in the Van Dale database Vlis with respect to the data requirements of EuroWordNet and an investigation of the possibilities to improve this information using the information extractable from the definitions and the bilingual dictionaries. The following table shows how the different VLIS relations are distributed for the Dutch equivalents of “move” and “know” at their next level (for an explanation of the relations see below): bewegen1 bewegen2 bewegen6 weten2 LE2-4003 SYN 1 1 - HYPO 21 17 12 2 ANTON - ASSOC 1 1 REF 1 1 - bewegen1 bewegen2 bewegen6 weten2 HYPER 1 1 1 - CAUSE - INCHOA - VERB - PREF 1 EuroWordNet 63 EuroWordNet D006: Definition of the links and subsets for verbs If we traverse all relations in downward direction (including all levels, but not considering the reference relation across parts-of-speech) we get the following figures for the full fields dominated by these words in VLIS: bewegen_1 bewegen_2 bewegen_3 weten_2 Related senses at all levels 145 senses 565 senses 54 senses 24 senses In the following sections we will discuss these relations and the possible EuroWordNet results for BEWEGEN (move) and WETEN (know) respectively. 4.2.1. Motion Verbs The equivalent Dutch senses for the 3 chosen synsets of the verb MOVE are respectively the senses 1, 2 and 6 of the verb BEWEGEN. In the case of BEWEGEN we see that the sense distinction is due to the different syntactic patterns: Sense 1 denotes the intransitive verb: to change place or position (literally translated); Sense 2 is the transitive verb meaning: to make place or position change (literally translated); Sense 6 is the reflexive verb having almost the same definition as sense 1 except for the addition "of people and animals" (literally translated); It is not clear that they also correspond with relevant semantic differences, e.g. many intransitive movement verbs can both refer to causative and non-causative movements. The hyperonyms of these senses also only differ because of the syntactic differences. Both sense 1 and 6 are related to the intransitive verb VERANDEREN (non-causative change) and sense 2 is linked to the transitive verb VERANDEREN (causative change): sense 1 has HYPERONYM veranderen_1 (intransitive) (to become different) sense 2 has HYPERONYM veranderen_4 (transitive) (to make different) sense 6 has HYPERONYM veranderen_1 (intransitive) (to become different) For the transitive and intransitive sense of BEWEGEN these hyperonyms appear to be correct since the head words of the definitions are the verbs 'veranderen' (to change) and 'doen veranderen' (to make change). In the case of sense 6, this cannot be correct because a reflexive movement verb mostly denotes a causative event. LE2-4003 EuroWordNet 64 EuroWordNet D006: Definition of the links and subsets for verbs Examination of the hyponyms further illustrates that sense distinction based on syntactic different patterns is not a clear criterion. The hyponyms of the transitive BEWEGEN are not all transitives. For example the verb kwispelen_1 (to wag) is a an intransitive verb and it is also encoded as such in Vlis,. Furthermore there are hyponyms encoded as both transitive and intransitive. Instead of sense differentiation, these verbs are here only linked to the transitive hyperonym. One of the hyponyms of the intransitive BEWEGEN is a reflexive verb, namely omdraaien_6 (turn around). Why it is not a hyponym of the reflexive sense 6 is unclear. The hyponyms of sense 6 represent 10 intransitives verbs, 1 reflexive and 1 transitive/ intransitive verb. Looking further down at the taxonomy it appears that the correlation between sense distinction and syntactic differences is encoded even in a less inconsistent way. If we abandon the syntactic difference we see that three different semantic aspects seem to play a role for the concept of “movement”: • • • whether or not something else is being moved is something moved from A to B or is the position or form of the body or object changed on the same location causality of the event, is it controlled by an agent or not. It is not very clear what the first two aspects really mean. If something is not moving from A to B then part of the body or object is still being moved from A to B. In many cases it is not so clear what the difference is. When somebody “jumps” he may do this on the same location or going from A to B, but is this really relevant for defining “jump”? Something similar can be said for moving oneself and moving something else. A verb such as “drive” can be used in “He will drive to LA” and in “He will drive you to LA”, but to what extent do we refer to another type of event? Another aspect relevant for distinguishing senses is the way these senses would be linked, or what differences in the relations correspond with these senses. For the first two aspects, we see that they do not result in a different semantic configuration (i.e. motivate a different semantic relation with some other verb). This is different for the causality aspect of movement. Since there is a semantic relation reflecting the difference between a non-causative change and a controlled causative change, we may consider the differentiation of the concept into two correlating senses as well. The above syntactically-based senses can then be linked to these semantic senses as follows: • causative BEWEGEN, linked to causative VERANDEREN_4 (change) which is transitive. The next ‘Vlis’ senses are captured by this sense: intransitive BEWEGEN_1 transitive BEWEGEN_2 reflexive BEWEGEN_6 • ex. of hyponym: kwispelen (to wag) ex. of hyponym: wrikken (to wrench) ex. of synonym: verroeren (to move oneself) non-causative BEWEGEN, linked to non-causative VERANDEREN_1 (change) which is intransitive. The next Vlis sense correlates to this sense: LE2-4003 EuroWordNet 65 EuroWordNet D006: Definition of the links and subsets for verbs intransitive BEWEGEN_1 ex of hyponym: deinen (to heave, surge) Next all hyponyms of BEWEGEN (sense 1, 2 and 6) have to be reconsidered in terms of causative and non-causative movement, where the syntactic categories can be used as a clue. In those cases that a single sense is labeled as both transitive and intransitive we thus have to decide whether this alternation correlates with a difference in causation. If so, we have to split the sense in a causative and a non-causative meaning. Alternatively we can also link it to both causative and non-causative BEWEGEN using the disjunction operation. In all the cases of such causative alternation we can create a CAUSE relation from each causative meaning to the non-causative meaning. From this we can conclude that we have to be aware of over-differentiation of senses due to syntactic differences and when necessary these differentiations have to be adapted by merging or splitting senses. Syntactic patterns can nevertheless be used as clues for finding semantic correlations. 4.2.1.1 SYNONYMY If we look at the explicit synonymy relations we see that the transitive verb BEWEGEN (move) has a synonym relation with the transitive verb VERROEREN (stir). The reflexive verb BEWEGEN (move) as well has a synonym relation, namely with the reflexive verb VERROEREN (stir). We have the following comments on this. First of all the verb “verroeren” is merely a reflexive verb; it is only used as a transitive in the fixed expression(s): 'geen vin(ger) verroeren' (not move an inch). So we would suggest not to have this sense at all and therefore there cannot be a synonym relation with it. Secondly we do believe that the reflexive VERROEREN has a definition that strongly suggests synonymy with the reflexive BEWEGEN, namely: "zich bewegen" (to move oneself). However the verb has a strong negative polarity as the example sentences in the bilingual dictionary show: Verroer je niet (don’t move) Je kunt je hier nauwelijks verroeren (you can hardly move hear) Geen blaadje verroert zich (not a leaf is stirring) This is probably due to the fact that it is difficult to substitute BEWEGEN by VERROEREN in many non-negative contexts, e.g.: Je kunt je hier goed bewegen/ ?verroeren (One can move oneself well here) Hij beweegt/?verroert zich voortdurend in zijn slaap (He moves constantly in his sleep) The question is whether such polarity really affects synonymy. If we decide to join the reflexive sense of BEWEGEN with the transitive sense, representing a single causative sense, we are faced with the fact that VERROEREN (which then is synonymous) can only be used as reflexive verb and is mostly used in negative contexts. LE2-4003 EuroWordNet 66 EuroWordNet D006: Definition of the links and subsets for verbs 4.2.1.2 HYPONYMY We already discussed the hyponymy relations for BEWEGEN above. 4.2.1.3 ANTONYMY In total 148 items for verbs have an antonymy relation. Directly at the level of BEWEGEN we do not find any examples either but, indirectly, we do find an example “stilhouden” (to keep still): stilhouden 1 {to keep still} SYNONYM stoppen 1 {to stop moving} ANTONYM voortgaan 1 (to continue) HYPERONYM gaan 3 (to go) ASSOCIATIVE komen 1 (to come) ANTONYM uitblijven 1 LEAF (to not happen) Here we see that “stilhouden” (to keep still) is a synonym of “stoppen” (to stop) which has an antonymy relation with “voortgaan” (to continue). In its turn “voortgaan” is linked to “gaan” (go) which should have been linked as a hyponym of BEWEGEN but for some reason is not (see below). Further down we see that “komen” (to come) also has an antonym “uitblijven” (not to happen). However, this is beyond the level of BEWEGEN. At more specific levels (as far as we can collect them now) there are only 8 senses with an antonym link (out of 764 senses linked to the three senses of BEWEGEN at any level): centraliseren 1 (centralize) =antonym= decentraliserem 1 (decentralize) leegschenken 1 =antonym= (to empty by pouring) volschenken 1 (to fill by pouring) neergooien 1 (to throw down) opgooien 1 (to throw upwards) =antonym= verkrampen 1 =antonym= (to seize with cramp) ontspannen 7 (to relax) verspreiden 1 (to spread out) verzamelen 2 (to collect) LE2-4003 =antonym= EuroWordNet 67 EuroWordNet D006: Definition of the links and subsets for verbs We expect that many more movements reflect some opposition but these are either not related as such or are linked to hyperonyms which are not related to BEWEGEN (for some reason). In the case of “dalen” (descent) and “stijgen” (rise) we see that there is an odd hyperonym toenemen (increase) for “stijgen” which is not further related: dalen 1 (to descent) ANTONYM=> stijgen 3 (to rise) HYPERONYM=> toenemen 1 (increase) Upward LEAF ASSOCIATIVE => vallen 2 (fall) Upward LEAF In the case of “weggaan” (leave) and “arriveren” (arrive) we see that they are both linked to an alternative hyperonym “gaan” (to go). However, in this case no antonymy relation is encode in Vlis: weggaan 1 (to leave) SYNONYM => vertrekken 1 (to leave) HYPERONYM=> gaan 3 (to go) HYPERONYM=> komen 1 (to come) arriveren 1 (to arrive) SYNONYM => aankomen 1 (to arrive) HYPERONYM=> komen 1 (to come) The missing antonymy relations have to be find using information from the definitions. 4.2.1.4 ASSOCIATIVE The ASSOCIATIVE relation is used in Vlis when the hyperonym relation was not satisfying or when there seemed to be a strong association with another entry which could not be expressed otherwise. This relation can be cross-part-of-speech. Within the Vlis database the relation has been assigned 557 times to a verb. Within the scope of the experiment we found occurrences of an associative relation for the verb bewegen_6 (to move), namely voortbewegen_2 (to move on/ forward) . The associative voortbewegen_2 with the reflexive BEWEGEN is a reflexive itself and has no definition. The translation ‘move on/forward’ already illustrates that the verb expresses movement in a certain direction. This observation is also confirmed by the hyponyms of ‘voortbewegen_2’ (to move forward) which all more or less express a movement forward, e.g.: to skate, swim, glide, walk, with exception of the hyponym ‘to jump’. The hyponyms of the reflexive BEWEGEN express movement of which the direction is not specified, e.g.: to dance, to swarm. Just as we have seen with “weten 2” (to know) and its ASSOCIATIVE “begrijpen 1” (to understand) it seems to be the case that the associative relation actually denotes a hyponym relation. So voortbewegen_2 (move on) is a hyponym of bewegen_6 (move) based on the translation and the linked hyponyms. LE2-4003 EuroWordNet 68 EuroWordNet D006: Definition of the links and subsets for verbs Another interesting relation is the hyperonym of ‘voortbewegen_2’ (to move on) which is gaan_3 (to go). The definition of this verb is: to move oneself and by that change place (translated literally). This definition looks very similar to the definition of bewegen_1: to change place or position (literally translated). The only difference is that gaan_3 (to go) cannot apply to position but just to place, which makes it more specific then BEWEGEN (to move). However the verb gaan_3 (to go) is a top node in Vlis and is not related to the verb BEWEGEN (to move). To clarify the situation we show the relevant links in Vlis as they were and the new situation proposed: Old situation: veranderen (to change) gaan (to go) |HYP |HYP |HYP bewegen zich bewegen ASSOCIATIVE ‡ zich voortbewegen (to move on) (move) (to move oneself) (to move oneself on) New links: veranderen (to change) |HYP bewegen (to move) |HYP |HYP zich voortbewegen gaan (move on) (to go) The question is whether voortbewegen (to move on) is a hyponym of gaan (to go) or that we should talk of a sort of synonym relation. If the synonymy relations does fully apply it is also possible to create a NEAR_SYNONYM link between “gaan” (go) and “voortbewegen” (move forward) to make clear that these two co-hyponyms of “bewegen” (move) are close in meaning, as opposed to all the other co-hyponyms we may find below “bewegen” (move). Clear tests for synonymy and hyponymy for verbs are required here. CONCLUSION We looked at other examples for associative relations and came to the conclusion that in most cases the assigned relation can be converted to another relevant relation, which helps to improve the taxonomy. 4.2.1.5 REFERENCE In VLIS this relation denotes morphological relationships between all categories and it has been assigned 838 times to verbs. Within the subset of the experiment only reference relations for verbs LE2-4003 EuroWordNet 69 EuroWordNet D006: Definition of the links and subsets for verbs were found. Bewegen_1 (move) has a reference link to the noun 'beweging' (movement), bewegen_2 (move) has a reference link to the adjective 'beweegbaar' (movable). The relation is not assigned very consistently; since both verbs should have a reference link to the noun and adjective. The link with the adjective cannot be expressed semantically in EuroWordNet. There is no relation for acts and adjectives denoting capabilities to perform these acts. In the case of the verb-to-noun relation, “bewegen-beweging”, this can be expressed as a noun-to-verb synonymy relation in EuroWordNet. However, this is not always the case. A verb such as “aanplanten” (to plant out) has a reference relation to “aanplant” (the planting) which is not the event but the object of to plant (that what is undergoing the event). This has to be converted to a PATIENT_ROLE relation between the verb and the noun. All the reference relations thus have to be checked for this difference in interpretation. For this the information from the definitions will be very helpful. 4.2.1.6 CAUSATIVE In Vlis this relation has rarely been assigned; only 5 times in total. VlisSense neerlaten_1 def = "naar beneden laten, laten zakken" => VlisSense zakken_1 def = "dalen, naar beneden gaan" (let down) (go down) VlisSense onderwerpen_2 def = "een behandeling doen ondergaan" (put through) => VlisSense ondergaan_1 def = "doorstaan, verduren" (go through) VlisSense opmaken_1 def = "geheel opgebruiken" => VlisSense opraken_1 def = none LE2-4003 (use up) (run out) EuroWordNet 70 EuroWordNet D006: Definition of the links and subsets for verbs VlisSense tegenmaken_1 def = "maken dat iem. tegenzin in iets krijgt, ergens tegen is" (make unwanted) => VlisSense tegenstaan_1 def = "onaangenaam zijn, weerstand oproepen" (pall upon) VlisSense uitdoen_1 def = "uittrekken, afleggen" (take off) = > VlisSense uitgaan_7 def = "uitgetrokken kunnen worden" (come off) These relations all seem to be correct but in many cases causation information has been left out and thus has to be derived differently. 4.2.1.7 INCHOATIVE Another relation that occurs in VLIS but not in WN1.5 is the inchoative relation. According to the guidelines for Vlis this relation is used to encode that a verb marks the beginning of another action. In practice, however, it turns out to be used to relate verbs that cause a non-dynamic situation referred to by another verb: krijgen1 def = ontvangen (to get) => hebben def. = bezitten (to have) kamperen1 def ="ergens een tijdelijk verblijf buitenshuis opslaan, vooral in tent of caravan" (to camp) => legeren def = "zijn legerplaats opslaan" (to encamp) "dichtzitten1" def = "afgesloten zijn" (to be closed) => dichtgaan1 def = "gesloten worden" (to close) These relations have to be reinterpreted as CAUSES relations between verbs in EuroWordNet. 4.2.1.8 VERB This relation is specifically used in VLIS to link compound verbs to their base form, which highly occur in Dutch (used 723 times in the database). In many cases, however, the relation between the base and the compound form is not clear. For example the verb gaan_3 (to go) has 12 verb relations, among which: afgaan_2 uitgaan_9 overgaan_8 LE2-4003 def. = helemaal langsgaan def. = ten einde gaan def. = overschrijden (go along the line) (go to the end) (go over a line) EuroWordNet 71 EuroWordNet D006: Definition of the links and subsets for verbs In the first two examples the prefix ‘af-‘ and ‘uit-‘ denote resp. the concept of ‘all along’ ‘to the end’. Verbs can be combined with these prefixes quite productively: aflopen (walk along),afwerken (work along), afzoeken (search along), affietsen (cycle along); uitlopen (walk to the end), uitfietsen (cycle to the end). The prefix ‘uit-‘ is combined with movement verbs and the ‘to the end’ concept is the literal meaning. But as the examples show the prefix ‘af-‘ also combines with verbs denoting other sorts of actions/ processes, in which case the ‘all along’ concept is meant figuratively. There are other prefixes that have this property as well, e.g.:‘ door- ‘ (through): doorlopen (walk through), doorzoeken (search through) . The last case of ‘overgaan_8’ (cross) we are not dealing wit a new word form but with a wrong word form which does not occur in this meaning. In this meaning the preposition “over” should be separate from the base form and is productively combined with “gaan” (over) in its normal meaning. Two other examples of over-generated senses are: aanschoppen_1 definition = "een schop geven tegen" (to kick against) => schoppen_2 = een schop geven (to kick). aanzwellen 1 definition = "in omvang of sterkte toenemen" (swell, rise) => zwellen_2 = (van geluiden) geleidelijk sterker worden (swell, rise) In this last example “aanzwellen” is linked to “zwellen” which has a definition which fits “aanzwellen” but not “zwellen”. So the senses of ‘overgaan’, ‘aanschoppen’ and “zwellen 2” have to be omitted from the database. From this it follows that the VERB relations cannot systematically be interpreted in terms of a single EuroWordNet relation and they all have to be reinterpreted on the basis of other information. 4.2.1.9 PREFERENCE The PREFERENCE relation is used to refer to other entries which are preferred and where the relevant information should be found. For verbs it occurs 637 times, e.g.: aanbotsen_1 preference true is (bump against) botsen (bump) aanliggen_2 preference true (lie close to) liggen (lie) aanbranden_1 preference false is (burn on) aangebrand (burnt (on)) aantekenen_1 preference false aangetekend LE2-4003 EuroWordNet 72 EuroWordNet D006: Definition of the links and subsets for verbs (make an note of) (written down) There are many past participle forms of verbs that have special meanings, but which are nevertheless linked to their infinitive verb form. We also see some compound verb forms. For each of these case the relevant relations have to be extracted from other information. As such the relation is not useful. 4.2.1.10. Information in definitions We inspected the definitions for generating (missing) relations and improving less reliable relations in Vlis. In order to find these cases we inspected all senses with definitions in which inflections of “bewegen” and other keywords occur like: 'beweging' (movement), 'heen en weer bewegen’ (move/go back and forth), “gaan” (go). A search on the last keyword resulted in 24 items of which 9 are related as hyponyms of 'bewegen' (move) already. The remaining 15 items are potential hyponyms of 'bewegen' (move), but for some reason other relations are assigned. In many cases these candidates were assigned as synonyms of one of the hyponyms of BEWEGEN (move), so in fact these are hyponyms too. In other cases the candidates were related to another hyperonym, e.g.: zwaaien_3 trans_intr (to wave) "(iets) krachtig heen en weer bewegen" (to move something back and forth with power) HYPERONYM = groeten_1 (to greet) differentiae = "met de hand" (with the hand) The verb 'zwaaien_3' (to wave) is a hyponym of 'groeten_1' (to greet) which is an obscure assignment. The information that 'zwaaien_3' (to wave) sometimes is a sort of greeting is not in the definition, besides it is too specific and should therefore not be its (only) hyperonym. It should be a hyponym of 'bewegen' (move), which could have been found by its definition. likken_1 trans_intrans met de tong heen en weer gaan over (to lick) (move with the tongue back and forth over) HYPERONYM = aanraken_1 (to touch) differentiae = "met een heen en weer bewegende tong" (with a back and forth moving tongue) The verb 'likken_1' (to lick) is a hyponym of 'aanraken_1' (to touch). Which is not wrong, but it is not the only more generic term that applies; it could also be a sort of cleaning. Again the information in the definition is overruled by other non-explicit information. To extract synonyms we can use the following resources: LE2-4003 EuroWordNet 73 EuroWordNet D006: Definition of the links and subsets for verbs _ _ _ bilinguals: all words which have the same translation(s) monolinguals: all words with exactly the same definition, explicit synonymy relations WordNet1.5: translation of variants in the synsets For extracting synonyms from translations we use an Acquilex-tool which generates all possible English translations of a set of Dutch source words in the bilingual Dutch-English dictionary (Martin and Tops, 1986) and next looks up these English words in the bilingual English-Dutch dictionary to see what Dutch translations are generated. In those cases that one of the Dutch source words was found among the translations of the English word the translations have been proposed as possible candidate for translation. These cases form a so-called translation-circle: we return back to one of the source words via the bilingual dictionaries. If these Dutch translation of the English translation contains new words these are offered as potential synonyms of the Dutch source words. In the following lists we find the result of this process starting with “bewegen” (move): Following startword has been selected: ((bewegen 7112 (1) V) (bewegen 7113 (1 2) V) (bewegen 7114 (1) V)) Currently Processing item: (bewegen 7114 (1) V) Translation Circles: (budge (zich_roeren zich_ver_roeren zich_verplaatsen bewegen zich_bewegen)) (travel (lopen heen_en_weer_lopengaan bewegen zich_bewegen verschuiven)) (run (voortbewegen bewegen schuiven glijden rollen rijden hard_rijden heen_en_weer_rijdenvaren pendelen aflopen voorbijgaan draaien werken sijpelen druipen stromen weg_stromen uitlopen trekken zwemmen klimmen kruipen rondzwerven weiden gelden v._kracht_zijn zich_uitstrekken gaan lopen voortduren duren voortgaan)) (start (bewegen plotseling_bewegen losspringen aanslaan te_voorschijn_springen)) (stir (bewegen in_beweging_brengen roeren ontstellen verontrusten beroeren wakker_maken)) (stir (roeren zich_ver_roeren bewegen zich_bewegen)) (move (bewegen in_beweging_zetten in_bewegingberoering_brengen roeren ver_roeren)) (move (drijven aansporen bewegen aanzetten ertoe_zetten)) (move (bewegen zich_bewegen in_beweging_komen zich_in_beweging_zetten v._positiehouding_veranderen zich_verplaatsen)) Acceptable Equivalents from the translation Circles: (roeren 61572 (1) V) (roeren 61570 (2) V) (voortgaan 81599 (1) V) (gaan 20427 (1 2 3 4 5 6 7 8) V) (voortbewegen 81584 (1) V) LE2-4003 EuroWordNet 74 EuroWordNet D006: Definition of the links and subsets for verbs Each of the translation-circles consists of the English word followed by a list of the translations and translation phrases. Below the translation-circles a list of the equivalents which have been accepted (this can be determined using an interactive inter-face). All the other translations either could be found in the dictionary or are not acceptable (refer to another sense of “bewegen”). This strategy resulted in the new synonyms “roeren” and “voortgaan” and in “gaan” and “voortgaan” which are linked as associatives in Vlis. In the case of the monolingual resources we looked at words with the same definitions and for explicit synonymy links encoded in another Dutch dictionary (van Sterkenburg and Pijnenburg, 1984). The explicit synonym fields in Vlis only give “verroeren” (move) as a synonym which is also given as the only synonym in Vlis. An example of similar definitions within the movement field is the following pair: Examples of words with (exactly) the same definition beven_1 (intransitive) def. = met een korte snelle beweging heen en weer gaan (shake) (go back and forth wit a short quick movement) trillen_1 (intransitive) def. = zich snel heen en weer bewegen (tremble) (to move oneself quickly back and forth) The distributions of the verbs beven_1 (shake) and trillen_1 (tremble) are very similar as well: • • • de grond beeft/trilt onder zijn voeten (the ground shakes/ trembles under his feet) zijn handen trillen/beven (his hands are shaking/trembling) zijn ooglid trilt/beeft (his eyelid trembles) Not in all cases the verbs can be interchanged, e.g. not in the expression : met schrik en beven. (with fear and tremble). Nvertheless, in Vlis trillen_1 is a hyponym of 'beven_1' whereas it seems more a (near) synonym. A final method for finding new synonyms is by translating variants of the equivalent WordNet1.5 synsets: <move, make move, displace> (transitive) <move, locomote, travel, go> (reflexive) The first synset shows that ‘displace’ is a variant of move and make move. The translation of this variant is the Dutch verb ‘verplaatsen’, which has the definition: ‘elders plaatsen’ (place somewhere else). In Vlis this verb is a hyponym of ‘brengen’ (to bring) and it is not related to “bewegen” (move). The strategy therefore does not result in a new synonym but it does create a connection with hyponyms LE2-4003 EuroWordNet 75 EuroWordNet D006: Definition of the links and subsets for verbs that are not inter-linked in Vlis. The second synset results in “reizen” (travel) and “gaan” (go) as new candidates, where “reizen” is a hyponym of “voortbewegen” (move forward) in Vlis. As mentioned above “gaan” can also be seen as a hyponym of “voortbewegen” which suggests that “gaan” (go) and “reizen” (travel) are potential synonyms linked together as a hyponym-synset of “voortbewegen” (move forward). However, they hardly share the same distribution, since “reizen” (travel) is only used for longer trips and “gaan” (go) can be used in a much more general way: i.e. “reizen” (travel) should be a hyponym of “gaan” (go). As we have seen above other relations than hyponymy and synonymy are hardly present in the Vlis database and we therefore fully have to rely on the information extractable from the definitions. The next examples illustrate typical patterns found in the experiment subset BEWEGEN (move). Using relatively-simple techniques we have found many antonyms within the domain of movement verbs. The first strategy involved scanning the definitions with “bewegen” (move), “gaan” (go), “voortbewegen” (move forward) for the negation operator “niet” (does not). This is illustrated by the following examples: buitensluiten: 1 (to lock out) Relations: buiten een afgesloten plaats houden , niet binnenlaten (to keep outside a locked place, not let in) buitensluiten: 1 ANTONYM binnenlaten (let in) eindigen: 1 (to end) Relations: niet verder gaan (not on go) eindigen: 1 staan: 8 niet bewegen (stand) (not move) Relations: staan: 1 NEAR_ANTONYM gaan (go) NEAR_ANTONYM bewegen (move) wegblijven: 1 (stay away) Relations: niet komen waar men verwacht wordt of hoort te komen (not come where one is expected or should come) wegblijven: 1 ANTONYM komen (come) missen: 2 (miss) Relations: niet treffen , niet raken (not hit) missen:2 ANTONYM raken (hit) Not all these cases also result in a direct antonymy relation. In some cases the opposing event expressed in the definition is not a lexicalized word and therefore we cannot express the antonymy relation: opblijven: 1 (stay awake) binnenhouden: 1 LE2-4003 niet naar bed gaan (not to bed go) binnenshuis houden , niet uit laten gaan EuroWordNet 76 EuroWordNet D006: Definition of the links and subsets for verbs (to keep in) (insight keep, not let go out) The verb “gaan” (go) is much too general to be an antonym of “opblijven” (stay awake) or “binnenhouden” (to keep in). What is possible is to link these words as hyponyms of a direct antonym of “gaan” (go). However, such an antonym does not exist either. Possible candidates such as “stilliggen” (to lay still), “stilstaan” (to stand still), “stilzitten” (to sit still) are too specific. The least thing we can do then is to link it higher up the hierarchy to the antonym of “veranderen” (change), which is “toestand” (state). Other negative elements in definitions are e.g. “voorkomen” (prevent), “onmogelijk maken” (make impossible): ontwijken: 1 (to move aside) Relations: voor iets of iem. uit de weg gaan , om een botsing te voorkomen (to move aside for something or someone, to prevent a clash) ontwijken: 1 NEAR_ANTONYM botsen (clash) blokkeren: 4 (to block) Relations: de beweging onmogelijk maken van to make movement impossible blokkeren:4 NEAR_ANTONYM beweging (movement) Noun Antonyms can also be encoded in a more implicit way. Many verbs refer to a change of state. Those verbs that refer to opposing end-results can be seen as a kind of antonyms as well. Consider for example two large classes of movement verbs that are in an opposition relation: fill-events and empty-events. There are many verbs that result in an empty or full state where the manner of filling or emptying is specified. In Dutch many of these verbs are in fact compounds where the verb denoting the event is combined with the resulting adjective “vol” (full) and “leeg” (empty). These can be found easily in Dutch on the basis of their composite structure: LE2-4003 EuroWordNet 77 EuroWordNet D006: Definition of the links and subsets for verbs leegmaken: 1 to empty leegstorten: 1 to empty by dropping leeggooien: 1 to empty by throwing out leegstromen: 1 to empty by streaming leeglopen: 1 to empty by streaming leegbloeden: 1 to loose one’s blood leeggieten: 1 to empty by pouring leegschenken: 1 to empty by pouring leeghalen: 1 to empty by taking out leegpompen: 1 to empty by pumping leegdrinken: 1 to empty by drinking leegeten: 1 to empty by eating leegplunderen: 1 to empty by plundering leegroven: 1 to empty by robbing leegruimen: 1 to empty by cleaning leegscheppen: 1 to empty by digging leegschudden: 1 to empty by shaking leegstelen: 1 to empty by stealing leegzuigen: 1 to empty by sucking vullen: 1 to fill volstorten: 1 to fill by dropping volplempen: 1 to fill by dropping in an uncontrolled way volstromen: 1 to fill by streaming vollopen: ? to fill by streaming volgieten: ? to fill by pouring volschenken: 1 to fill by pouring volstoppen: 1 to fill completely volproppen: 1 to fill completely by pushing volpompen: 1 to fill using a pump volscheppen: ? to fill by digging Not every event has also an opposing event. It is simply impossible to fill by stealing, robbing, sucking, eating or drinking, etc.. The words followed by question marks are not in the database (for some reason). A final special case of negation is the following example in which we do not find a full opposition but in which the caused effect is negated: misschieten: 1 (to miss when shooting) bij het schieten niet raken (during shooting not hitting) We cannot say that “misschieten” (to miss when shooting) is an antonym of “schieten” (shoot) because it implies it. We cannot say either that it is an antonym of “raken” (hit) because it is much more specific. The information in the definition matches that of a cause relation. The best solution would therefore be to store the information as a cause relation where the result or caused event is negated: Relations LE2-4003 misschieten: 1 CAUSES NOT raken EuroWordNet 78 EuroWordNet D006: Definition of the links and subsets for verbs The negation label NOT would likewise function in the same way as factive and non-factive implications. Note that we can store the above verbs with a contrastive result-state in exactly the same way using this operator: leegpompen (to empty by pumping) HAS_HYPERONYM CAUSES volpompen (to fill by pumping) HAS_HYPERONYM CAUSES pompen (pump) leeg (empty) pompen (pump) NOT leeg (empty) In this way we can differentiate the antonymy relation in terms of an opposition in result and not as a full opposition of the events. It is also clear now that the manner of getting at the result is the same for both verbs. A negation operator would provide a very powerful and useful way of encoding all kinds of contrasts and implications in the database. The following examples illustrate definition patterns for CAUSE relations that have been found for movement verbs: _ doen (do) + Vinfinitive brengen: 3 (bring) Relations: <iets> doen toekomen (<something> do get) brengen: 3 CAUSES toekomen (get) The following example expresses a causation relation between verbs but the resulting verb does not refer to a dynamic event or process but to a state.: zetten: 1 (put) Relations: rechtop op de genoemde plaats doen zitten, rusten, staan (upright on the named place do sit, rest, stand) zetten: 1 CAUSES zitten d1 (sit) zetten: 1 CAUSES rusten d2 (rest) zetten: 1 CAUSES staan d3 (stand) Note also that the coordination is interpreted here as a disjunction of relations expressed by the disjunction indexes d1, d2, d3. LE2-4003 EuroWordNet 79 EuroWordNet D006: Definition of the links and subsets for verbs _ laten (let) + Vinfinitive druppelen: 2 (drip) Relations: _ om te (in order to) sjorren: 1 Relations: _ • druppels laten vallen (drops let fall) druppelen: 2 CAUSES vallen (fall) ingespannen aan iets trekken om het te verplaatsen of om het los of vast te maken (to pull fiercefully in order to move it or in order to untie it or tie it) sjorren:1 HAS_HYPERONYM trekken (pull) sjorren:1 CAUSES/HAS_HYPERONYM verplaatsen (move) sjorren:1 CAUSES los d1 (free, untied) sjorren: 1 CAUSES vast d2 (tied) door (te) (by means of) aftrekken: 5 (tear off) Relations: dichttrekken:2 Relations: wegkappen: 1 Relations: verwijderen door te trekken (remove by pulling) aftrekken:5 CAUSES/HAS_HYPERONYM verwijderen (remove) aftrekken:5 HAS_HYPERONYM trekken (pull) sluiten door te trekken (close by pulling) dichttrekken:2 CAUSES/ HAS_HYPERONYM sluiten (close) dichttrekken:2 HAS_HYPERONYM trekken (pull) <iets> verwijderen door te kappen (<something> remove by cutting) wegkappen:1 CAUSES/ HAS_HYPERONYM verwijderen (remove) wegkappen:1 HAS_HYPERONYM kappen (cut) Note that in most of these examples the events are not disjoint in time. Since they also fit the hyponymy test the HAS_HYPONYM relation should be preferred here above the CAUSES relation. LE2-4003 EuroWordNet 80 EuroWordNet D006: Definition of the links and subsets for verbs The next examples illustrate SUBEVENT relations: _ temporal relations between events expressed by prepositions such as “onder” (under), “gedurende” (during), “terwijl” (while), “tijdens (during). ontvoeren: 1 (kidnap) Relations: <iem.> onder dwang meenemen (<somebody> under pressure take away) ontvoeren:1 HAS_HYPERONYM meenemen(take ontvoeren:1 dwang (pressure) away) trekkebenen: (limb) Relations: SUBEVENT met een been trekken tijdens het lopen (to pull your leg during walking) trekkebenen:1 HAS_HYPERONYM trekkebenen:1 SUBEVENT trekken (pull) lopen (walk) _ conjunctive coordination, V en (and) V. aanhouden: 9 Relations: arresteren en voor verhoor meenemen (arrest and take away for interrogation) aanhouden:9 SUBEVENT aanhouden:9 SUBEVENT arresteren (arrest) meenemen (take with) _ met (with) higher-order noun + V fladderen: 1 Relations: LE2-4003 met ongelijkmatige bewegingen vliegen (with irregular movements fly) fladderen HAS_HYPERONYM fladderen SUBEVENT vliegen (fly) beweging N (movement) EuroWordNet 81 EuroWordNet D006: Definition of the links and subsets for verbs The following examples illustrate the cross-part-of-speech relations that can be extracted: _ doen (do), geven (give), beoefenen (perform), zijn (be) + higher-order noun slaan: 1 (hit) Relations: veroorzaken (to cause) Relations: een slag of slagen geven (a blow or blows give) slag N XPOS_HAS_SYNONYM (blow) oorzaak zijn van (cause be of) oorzaak N ROLE_AGENT (a cause) slaan (hit) V veroorzaken V (to cause) _ V + met first-order-noun carpoolen:1 (carpool) Relations: gezamelijk gebruik maken van een auto voor het woon-verkeer (common use of a car for home-travelling) carpoolen V INVOLVED_INSTRUMENT auto N (carpool) (car) _ Adjective maken (make), worden (become) vullen: 1 (fill) Relations: vol maken (full make) vullen: 1 CAUSES vol (full) Adj These examples are by no means complete. They only illustrate the patterns in the definitions that can be used to extract these relations. As general conclusion it shows that a lot of additional information can be extracted from the resources using semi-automatic techniques. 4.2.2. Verbs of knowing The Dutch equivalent for “know” is “weten” which has 5 senses in Vlis, one as a noun and 4 as a verb: SENSE 1. NOUN. kennis, wetenschap omtrent iets (knowledge of something) SENSE 2. VERB_transitive kennis hebben van (to have knowledge of) SENSE 3. VERB_transitive inzien, begrijpen (to see, understand) LE2-4003 EuroWordNet 82 EuroWordNet D006: Definition of the links and subsets for verbs SENSE 4. VERB_transitive zich bewust zijn (to be conscious of) SENSE 5. VERB_transitive kans zien, erin slagen, in staat zijn (to manage, succeed, be capable of) The first sense refers to the noun “weten” which is equivalent to the second sense of “weten”. Remarkably, this is not expressed by a so-called reference relation in Vlis which is used to mark such morphological correspondences. The 5th sense is clearly different and not relevant for our discussion here, the other senses are closely related to English KNOW. All the verbal senses of “weten” are coded as transitive so we cannot say there is over-differentiation due to different syntactic patterns (although it is not clear to what extent further differences in the complementation have triggered the senses distinction). From a semantic point of view we can say that the second sense of WETEN is clearly the most neutral and basic sense. This is confirmed by the Vlis data: • • WETEN_2 is a top or upper-leaf (it has no more general hyperonyms or other more general concepts to which it is related). the other verbal senses are directly or indirectly related to sense 2. The latter point is illustrated by taking the Vlis relations of sense 3 and 4 in upward direction: weten 4 (know) SYNONYM realiseren 2 (realize) HYPERONYM weten 2 LEAF (know) weten 3 SYNONYM begrijpen 1 (understand) ASSOCIATIVE weten 2 LEAF (know) Here we see that “weten 4” is a synonym of “realiseren 2” which in its turn is a hyponym of “weten 2”, and “weten 3” is a synonym of “realiseren 2” which has an association relation with “weten 2”. The neutral, basic sense of “weten 2” is correctly defined as “to have knowledge of”. The more special 3rd sense of “weten” refers to a special kind of knowledge: the solution to a problem, the structure of a complex condition, the working of a mechanism. It is not clear what is meant with the association relation here. It could have been linked as a SYNONYM or a HYPONYM of “weten 2”. The more special 4th sense refers to conscious knowledge as opposed to unconscious knowledge. It is not clear whether these more special meanings can be seen as pragmatic restrictions in the usage of “weten” or that they really denote distinct meanings. If we look at more coarse distinctions we see that none of the senses refers to an action or process: they all refer to the state of knowing something. It is also possible to explain the 3rd and 4th senses of “weten” by means of a pragmatic principle that a general word can often be used to refer more specific concepts as well. LE2-4003 EuroWordNet 83 EuroWordNet D006: Definition of the links and subsets for verbs 4.2.2.1 SYNONYMY As suggested above there are no synonyms for the abstract meaning of “weten 2”. The more specific meanings will be discussed below. 4.2.2.2 HYPERONYMY There are no hyperonyms for the abstract meaning of “weten 2” which is reasonable (see above), while the 3rd and 4th meaning are indirectly linked to the 2nd sense. In the case of the 3rd meaning we already mentioned that its synonym “begrijpen 1” should have been linked to “weten 2” as a hyponym rather than the unspecific relation association. 4.2.2.3. HYPONYMY In the case of the verb weten_2 (to know) we see that there are two explicit hyponyms in VLIS and one associative “begrijpen 1” which should be seen as a hyponym: kennen_2 (to know) door studie of oefening geleerd hebben (learned by study or training) realiseren_2 (realize) beseffen (be aware of) begrijpen_1 (to understand) de samenhang doorzien van (to see the coherence of) The last two have been explained above. The hyponym “kennen_2” refers to learned facts which is again a special subsets of facts in general. If we reinterpret the association relation as hyponymy we thus would get the following general picture for WETEN in Dutch, where “weten 3” and “weten 4” pop up as hyponyms of “weten 2”: LE2-4003 EuroWordNet 84 EuroWordNet D006: Definition of the links and subsets for verbs weten 2 {to have knowledge of} HYPONYM begrijpen 1 {knowledge of complex facts} (previously ASSOCIATIVE) SYNONYM bevatten 2 LEAF SYNONYM doorhebben 1 LEAF SYNONYM snappen 3 LEAF SYNONYM vatten 2 LEAF SYNONYM verstaan 2 LEAF SYNONYM weten 3 LEAF HYPONYM aanvoelen 2 {to know on intuition, without factual evidence} SYNONYM invoelen 1 LEAF SYNONYM voelen 6 LEAF HYPONYM voorvoelen 1 LEAF (to know in advance} HYPONYM doorzien 1 {to know something very complex} HYPONYM doorgronden 1 {to know something extremely complex} SYNONYM doorvorsen 1 LEAF HYPONYM misvatten 1 {to wrongly understand} SYNONYM misverstaan 1 LEAF HYPONYM kennen 2 HYPONYM realiseren 2 SYNONYM beseffen 1 SYNONYM inzien 2 LEAF SYNONYM weten 4 LEAF HYPONYM onderkennen 1 LEAF {to know despite one’s wishes} This hierarchy is acceptable accept for “misvatten” which is not a hyponym but an antonym of “begrijpen”. 4.2.2.4 ANTONYMY For WETEN (know) we do not find any cases of explicitly coded antonymy. We have already seen in the previous section that “misvatten” should have been coded as an antonym of the hyponym “begrijpen” (to know something complex) but was coded as a hyponym. 4.2.2.5 ASSOCIATIVE The association relations for WETEN (know) have been discussed above. In addition we also find an association relation between “realiseren 2” (realize) and “begrijpen 1” (understand) without really adding any information. LE2-4003 EuroWordNet 85 EuroWordNet D006: Definition of the links and subsets for verbs 4.2.2.6 REFERENCE We already suggested that the REFERENCE relation of “weten 1” (as a noun) and “weten 2” was missing. Other REFERENCE relations are hardly present in this domain. 4.2.2.7 CAUSATIVE There is no causative relation encoded for senses related to WETEN in Vlis. 4.2.2.8 INCHOATIVE In the case of WETEN there is one INCHOATIVE relation with the hyponym “begrijpen” (understand): begrijpen 1 (understand) INCHOATIVE=> doorkrijgen 1(to get to understand) The verb “doorkrijgen” refers to the event in which a person undergoes a change of mental state from not-understanding to understanding. 4.2.2.9 VERB and PREFERENCE Verb and preference relations have not been encoded in Vlis for verbs related to WETEN. 4.2.2.10. Information in definitions As we have seen above, WETEN (know) denotes a process which is not as differentiated as the concept of MOVEMENT. By looking at definitions in which the verbs “weten” (know), “begrijpen” (understand) and “realiseren” (realize) occur we tried to find more relations which have not been encoded in Vlis. The potential field of know-verbs has been enlarged by looking also at definitions in which the noun “kennis” (knowledge) occurs. This strategy did not give many new results with respect to hyponymy relations. The following list contains the concepts that come closest to WETEN: LE2-4003 EuroWordNet 86 EuroWordNet D006: Definition of the links and subsets for verbs (herinneren 33581332 (2) V) (ignoreren 16806482 (1) V) (impliceren 16806741 (1) V) (inzien 16808261 (1) V) (kennen 16809949 (1 4) V) (misvatten 16821041 (1) V) (misverstaan 16821043 (1) V) (onderkennen 16825609 (1) V) (ontkennen 16826586 (1) V) (vatten 50409046 (3) V) (vergeten 16855388 (1) V) (verleren 16855784 (1) V) (verstaan 33633593 (2 4) V) (remember) (to ignore) (to imply) (to see the point) (to know) (to misunderstand) (to misunderstand) (to recognize) (to deny) (to understand) (to forget) (to forget) (to understand) Some of these are already related to WETEN (know) in Vlis: “inzien”, “kennen”, “vatten”. Most of the new cases refer to dynamic changes which result in a state of knowing but cannot be seen as hyponyms or hyperonyms of the state knowing. We find also some antonyms which will be further discussed below. If we apply the Acquilex-tools for finding synonyms (described above) we get the following results: Extracting equivalents using the bilingual dictionaries Following startword has been selected: (weten 84581 (1 2 3 4) V) (begrijpen 5812 (1) V) ((kennen 32733 (2) V)) Currently Processing item: (weten 84581 (1 2 3 4) V) Translation Circles: (tell (weten uitmaken zeggen kennen)) (realize (beseffen weten zich_realiseren zich_bewust_zijnworden)) (know (weten bewust_zijn zich_bewust_zijn beseffen kennis_hebben kennis_hebben_van)) Currently Processing item: (begrijpen 5812 (1) V) LE2-4003 EuroWordNet 87 EuroWordNet D006: Definition of the links and subsets for verbs Translation Circles: (grasp (vatten omvatten begrijpen)) (comprehend (vatten be_vatten doorgronden doorzien begrijpen)) (take (begrijpen in_zich_opnemen beseffen doorzien)) (take (begrijpen beschouwen aannemen opvattennemen snappen)) (understand (begrijpen verstand_hebben_van verstaan snappen inzien vatten be_vatten)) (understand (begrijpen begrip_hebben_voor)) (understand (begrijpen vernemen aannemen uit_opmakenafleiden er_uit_opmakenafleiden)) (understand (begrijpen het_begrijpen het_snappen)) Search completed!!!!!! Currently Processing item: (kennen 32733 (2) V) Translation Circles: (know (kennen machtig_zijn bekendvertrouwd_zijn_met)) (know (kennen onderhevig_zijn_aan ervaren ondergaan)) Search completed!!!!!! Potential Equivalents from the Translation Circles: (vernemen 78759 (2) V) (hear) (verstaan 79161 (2) V) (understand) (aannemen 317 (4) V) (accept as a view) (beschouwen 6605 (1) V) (consider) (beseffen 6621 (1) V) (realise) (doorzien 15632 (1) V) (understand) (doorgronden 15427 (1) V) (understand) (snappen 67414 (2) V) (understand) (bevatten 6979 (2) V) (understand) (vatten 77398 (3) V) (understand) Most of the co-translations in the translation-circles are not entries or could not be found (due to the reflexives or the nominalized form). Most acceptable words represent synonyms of “begrijpen” in Vlis. Some of the new cases are not really synonyms but refer to events causing a state of knowing, e.g. “beschouwen” (consider), “vernemen” (hear). Only “verstaan” (understand) is a new synonym not yet included in Vlis. If we look at the explicit synonym fields in the Van Dale monolingual Dutch dictionary we again come across dynamic, event-denoting words which are wrongly presented as synonyms: “bespeuren” (notice), “gewaarworden” (perceive), “erkennen” (admit). This strategy does not result in new synonyms which have not been included in Vlis. LE2-4003 EuroWordNet 88 EuroWordNet D006: Definition of the links and subsets for verbs We also did not find cases of similar definitions (consisting of several words) for the concepts directly related to WETEN (one-word definitions are not considered as good candidates). Some overlapping definitions have been found for the following indirectly related verbs: • onderwijzen: 1 (educate) bijbrengen: 1 (teach) opleiden: 1 (train) bekwamen: 1 (train) kennis , begrip of vaardigheid overbrengen (to carry over knowledge, understanding or skill) <kennis> verschaffen , eigen laten maken (to provide <knowledge>,…) de nodige kennis en vaardigheid bijbrengen (to provide the necessary knowledge and skills) de nodige kennis en vaardigheid bijbrengen (to provide the necessary knowledge and skills) • misvatten: 1 (misunderstand) misverstaan: 1 (misunderstand) verkeerd opvatten , begrijpen (consider, understand wrongly) verkeerd begrijpen (understand wrongly) From these definitions we can at least derive the following synsets: {onderwijzen, bijbrengen, opleiden, bekwamen} {misvatten, misverstaan} which both are related to WETEN in an indirect way: causing it and denying it respectively. Starting from WordNet 1.5 did not result in any new synonyms. The synset for knowing in WN1.5. is {know cognize} which does not provide new translations. Concluding so far we can say that, a poorly lexicalized domain as knowing, apparently does not provide much material for extracting more hyponymy and synonymy relations. This may be due to the fact that there simply are not many related words or that the resources do not provide sufficient information in a consistent way to be able to extract these relations. However, as a non-dynamic, static verb we can expect that there are many more verbs denoting a change resulting in a state of knowing. Typically, we can think of the following types of events that denote such changes: LE2-4003 EuroWordNet 89 EuroWordNet D006: Definition of the links and subsets for verbs • communication verbs which result in the dissemination of information and therefore have a state of knowing this information as a result. • mental-action verbs leading to particular states of knowing • perception verbs leading to states of knowledge • educational actions leading to the acquisition of knowledge, such as reading, studying, learning, teaching, etc.. When we look at causative definition patterns of know-related events these groups clearly emerge. In the following examples this is evident for definitions in which “kennis” (knowledge) is the object of some transaction, by teaching and/or communication: inlichten: 1 (inform) Relations: de nodige kennis verschaffen , opheldering geven (to provide the necessary knowledge, to give clarification) inlichten: 1 CAUSES kennis onderwijzen: 1 (educate) Relations: kennis , begrip of vaardigheid overbrengen (to carry over knowledge, understanding or skill) onderwijzen: 1 CAUSES kennis leren: 2 (learn) Relations: bedrevenheid , kennis verwerven in (to acquire skills and knowledge in) leren: 2 CAUSES kennis When we look at the usual patterns for causation definitions we find the following cases: _ doen (do, make) + V infinitive uitleggen: 1 (explain) Relations: doen begrijpen (make understand) uitleggen: 1 CAUSES uitleggen: 1 bekendmaken: 1 (announce) Relations: HAS_HYPERONYM openbaar maken , doen weten (to make public, make known) bekendmaken: 1 CAUSES bekendmaken: 1 LE2-4003 begrijpen (understand) doen (do, act) weten (know) HAS_HYPERONYM doen (do, act) EuroWordNet 90 EuroWordNet D006: Definition of the links and subsets for verbs _ laten (let) + Vinfinitive melden: 1 (announce) Relations: berichten , laten weten notify, let know melden: 1 CAUSES melden: 1 waarschuwen: 2 (warn) Relations: HAS_HYPERONYM verwittigen , laten weten notify, let know waarschuwen:2 CAUSES waarschuwen:2 _ weten (know) doen (do, act) weten (know) HAS_HYPERONYM doen (do, act) om te (in order to) + Vinfinitive verkennen: 1 (explore) Relations: uitgaan om kennis trachten te verwerven over (to go out in order to try to gain knowledge about) verkennen:1 CAUSES verwerven (acquire) CAUSES kennis (knowledge) verkennen:1 HAS_HYPERONYM uitgaan (go out) aanleren: 1 zich <een kennis of vaardigheid> eigen maken (teach) (acquire <a knowledge or skill>) Relations: aanleren:1 CAUSES kennis (knowledge) aanleren:1 HAS_HYPERONYM eigen maken (acquire) aanleren: 2 (teach) Relations: <een kennis of vaardigheid> onderwijzen (teach <knowledge or skill>) aanleren:2 CAUSES aanleren:2 LE2-4003 HAS_HYPERONYM kennis (knowledge) onderwijzen (teach) EuroWordNet 91 EuroWordNet D006: Definition of the links and subsets for verbs _ te (to) Vinfinitive komen (get) uitvinden: 2 (find out) Relations: te weten komen (to get to know) uitvinden:2 CAUSES weten (know) achterhalen: 3 (find out) Relations: ontdekken , te weten komen (discover, to get to know) achterhalen:3 CAUSES weten (know) Typically, we see here that many causal patterns do not mention any of the know-verbs but refer to “kennis” (knowledge) as the object of some action. Only indirectly we can infer from this that a state of knowing is the result. Looking for conjunction of events resulted in the following examples of SUBEVENT relations: ontcijferen: 1 (decode) Relations: <iets onduidelijks> lezen en begrijpen (<something unclear) to read and understand) ontcijferen:1 SUBEVENT lezen (read) c1 ontcijferen:1 SUBEVENT begrijpen (understand) c2 bestuderen: 1 met aandacht lezen en overwegen om de betekenis ten volle te begrijpen (study) (to read and consider carefully in order to understand the full meaning) Relations: bestuderen:1 CAUSES begrijpen non-factive (understand) bestuderen:1 SUBEVENT lezen (read) c1 bestuderen:1 SUBEVENT overwegen (consider) c2 The first example is not fully straight forward. Although “ontcijferen” is represented here as a separate embedded state it is really the result of “lezen” (read). In the case of “bestuderen” this is done in a more correct way: “begrijpen” is the resulting state and the other verbs are subevents of embedding situations possibly (non-factive) leading to it. LE2-4003 EuroWordNet 92 EuroWordNet D006: Definition of the links and subsets for verbs When we look for definitions with negative elements we find the following potential antonyms: • achterhouden: 2 (to keep back) niet mededelen, geheimhouden (to not report, to keep as a secret) geheimhouden: 1 (keep as a secret) <iets> niet openbaren , ... (to not make<something> known, ...) Relations: • {achterhouden, geheimhouden} NEAR_ANTONYM {mededelen, openbaren} (report, make known) denken: 3 (remind) niet vergeten (not forget) vergeten: 1 (forget) uit het geheugen verliezen (to loose from memory) vergeten: 3 (forget) Relations: niet meer bewust zijn van , niet denken aan (to be no longer conscious of, not think of) denken: 3 NEAR_ANTONYM vergeten:1, 3 (forget) • verleren: 1 (unlearn) na enige tijd niet meer kennen of kunnen (after a while no longer being able or know) • misvatten: 1 (misunderstand) verkeerd opvatten , begrijpen (consider, understand wrongly) misverstaan:1 (misunderstand) verkeerd begrijpen (understand wrongly) Relations: {misvatten, misverstaan} NEAR_ANTONYM begrijpen (understand) Some of these have been described above. They all refer to events in which either a state of knowing or the negation of a state of knowing is the result. These events can be communication verbs or mental acts resulting in having access to knowledge. Some of them form pairs, but “verleren” (unlearn) is in an opposition with “kennen” (know) which is too general. We can expect that there will not be many cross-part-of-speech relations with “weten” (know). As a non-dynamic verb we can exclude concepts where “weten” is the cause of another event. The reverse situation that another verb represents the cause of knowing something has already been discussed above. What is left then are hyponymy and synonymy across part-of-speech. In this respect, LE2-4003 EuroWordNet 93 EuroWordNet D006: Definition of the links and subsets for verbs we already mentioned the nouns “weten” (the knowing) and “kennis” (knowledge). Scanning the typical definition patterns for these relations only yielded the following results: (begrijpelijk 16783027 (1) A) (onkenbaar 16826260 (1) A) (onbegrijpelijk 16825257 (1) A) (onbegrip 16825258 (1) N) (begrip 67114678 (4) N) (comprehensive) (unrecognizable) (uncomprehensive) (misunderstanding) (understanding) The first three cases are adjectives which refer to the capability to be understandable or not. This cannot be expressed using the current set of relations. The last two nouns are antonyms where “begrip” (understanding) is synonymous to “begrijpen” (to understand) and can be linked by the XPOS_HAS_SYNONYM relation. 4.2.3 Conclusion The general conclusion is that the Vlis relations can be taken as a starting point for deriving the EuroWordNet database. Where possible this information will be improved by: • • • extracting additional information from definitions using the bilingual dictionaries comparing wordnets cross-linguistically Special attention will be paid to relations which are unreliable or ambiguous, to phenomena such as transitive/intransitive alternation of verbs, or verb compounds of which we know that they may have lead to wrong decisions. Furthermore, some relations (e.g. entailment, roles) are missing in the Vlis database and have to be derived by systematically looking for patterns expressing it. In general we can apply the following techniques for improving data and adding links: • • • • • • extracting definition patterns and structure for relevant relations. check for the badly motivated ends in Vlis relations. check frequent combinations of hyperonyms for synonymy, disjunct or conjunct interpretations. If disjuncts are too extreme the synsets have to be differentiated again. Since multiple hyperonyms are not allowed in VLIS in our case we have to find these multiple cases by: • looking for deviating and coordinated genus words in definitions • looking at other senses • checking the association and reference relations • enlarging the synsets with bilinguals look for phrases in definitions to find missing children look for top-concepts to find potential combinations of hyperonyms look at classifications in other sources to find alternative hyperonyms LE2-4003 EuroWordNet 94 EuroWordNet D006: Definition of the links and subsets for verbs • • generate synsets using bilinguals to find new synonyms check the different meanings of the entries by looking for frequent pairs of hyperonyms that are applied to different meanings if the same word. The extraction of information from the definitions was done in an experimental way, but have shown that many regular patterns can be found that provide additional information. This information is certainly useful provided of course that it is clear what we are looking for and what the implications are. We have focused the discussion on the more general levels of knowing and moving. At this level we expect that the relations are least clear and most problems are concentrated. At more specific levels the given relations are more reliable and information is more explicit and therefore more easily extractable. LE2-4003 EuroWordNet 95 EuroWordNet D006: Definition of the links and subsets for verbs 4.3 First Results of work on Italian 4.3.1 Motion Verbs In order to try the encoding of relations for Italian motion verbs, we have first examined our Italian LDB aiming at extracting all the hyponyms (and their synsets) of the verb muoversi (= to move, indicating motion by the protagonist of the event, and roughly corresponding to the synset selected in WN 1.5 for the experiment). Before trying either to build synsets out of the data on synonymy or to build taxonomies, however, it was necessary to perform a sense disambiguation of all the synonym/hypernym verb homographs: actually, as already explained above (§ 3.1.1), within the LDB only relations among verb entries and verb homographs are identified, while the particular sense in which a verb is synonym/hyperonym of another is not coded. Moreover, some information had to be added or modified manually, in order to correct errors and inconsistencies found within the LDB, and some changes were made to the original source. A major problem of carrying out a work of this kind, however, is that of the subjectivity of the decisions taken: in fact, we must remind that, as expected in the area of semantics, our intuitions not only are sometimes different from data found in dictionaries but are also sometimes different from those of other speakers. In the following, we shall i) provide a general review of the problems encountered/solutions (preliminarily) identified; ii) provide examples of relations encoded. In the Appendix, the muoversi taxonomy is then reported for reference. 4.3.1.1 Major problems encountered / solutions identified 1) A problem we encountered several times was that of a not coherent distinction of senses for verbs referring either to motion along an unbounded path or to motion along a bounded path. This is a relevant distinction in Italian, given that not all Italian motion verbs may refer to motion to a goal/from a source (which is instead possible for all English motion verbs) (cf. Acquilex papers on this issue). On the other hand, as we have already noted above, data on cross linguistic differences in lexicalisation patterns are necessary for LE applications. Different dictionaries of Italian, which we examined, distinguish different senses for the same verb and also within the same dictionary there are verbs for which different senses (unbounded and bounded) are distinguished and others for which a similar distinction is not formulated, although they however display a similar syntactic behaviour/semantic reference. Therefore, we merged information coming from different sources and added other information manually when necessary. This was done, for instance, with respect to andare (to go) (but also in relation to other verbs: correre (to run), volare (to fly), etc.). In fact, this verb can be connected with two fundamental (motion)18 senses: one referring to undirected motion (or motion along an unbounded path) and the other referring to directed motion 18 Andare also has ‘non-motion’ senses similarly to the English to go. For the moment, however, only the fundamental motion senses have been taken into consideration. LE2-4003 EuroWordNet 96 EuroWordNet D006: Definition of the links and subsets for verbs (motion along a bounded path, i. e., motion to a goal or from a source: in other words, change of position). It is also indicated in dictionaries as a synonym of a number of verbs: - gire - ire - muovere (intr.) 0_1 (to move) - recarsi 0_2b (to go to a place) - portarsi 0_1 (to go to a place) (synonym of trasferirsi 0_0 = to change position) The first two verbs are archaic variants of andare itself and thus we have not considered them; (intransitive) muovere (to move) refers to undirected motion, whereas recarsi (to go to) and portarsi refer to motion to a goal (change of position). Two fundamental synsets have thus been created for andare: Sense 1 andare, muovere -- (to go along an unbounded path) Sense 2 andare, recarsi, portarsi, trasferirsi -- (to go to a place, to change position). The two synsets both appear to match with Sense 1 of to go in WordNet 1.5, since the difference between undirected/directed motion is not relevant in English, given that, as recalled above, all English motion verbs may refer to motion to a goal/from a source (have a derived ‘directed motion’ sense). In Italian, instead, the difference is relevant since only a (relatively small) subset of motion verbs may refer to motion to a goal (or from a source) and this has consequences on the syntactic properties of the different subsets of Italian motion verbs (cf. the difference between correre and nuotare: the former may either refer to motion along an unbounded path - Corse per ore (He/She ran for hours) - or along a bounded path - Corse a casa (He/She ran home) -; the latter only refers to motion along an unbounded path - Nuotò per 10 minuti / * a riva (He/She swam for 10 minutes / (*) to the bank)). 2) If andare and similar verbs have two senses that map to just one in WN 1.5, other verbs do not have the ‘directed motion’ sense: e. g., nuotare (to swim), camminare (to walk), etc. (which, in fact, cannot occur with directional phrases). These verbs were preliminarily connected with their ‘partially correspondent’ verbs in English (to swim and to walk), however we think that the link ‘Involved’ is necessary to clarify the semantic characteristics of them, i. e., the different lexicalisation patterns of semantic components within their roots with respect to their English counterparts. 3) In general Italian synsets have been built both using explicit data on synonymy within the LDB (words indicated as synonymous of an entry) and ‘implicit’ data (single words used to define an entry: e. g. correre 0_3 (to run) is defined as ‘accorrere’ (to run to a place) and we have considered them as synonyms). However, sometimes both our intuitions and the data from the bilingual (Italian-English) have led us to build synsets using verbs which were not related in the Italian dictionaries. For instance, girandolare, girellare, gironzolare and girovagare have very similar semantic references and, in fact, LE2-4003 EuroWordNet 97 EuroWordNet D006: Definition of the links and subsets for verbs can be all translated as “to saunter”, “to stroll”, thus we have decided to put them in a same synset, even if they were not indicated as synonyms within dictionaries. 4) Sometimes, within a dictionary, a verb is defined as a hyponym of another verb even if another dictionary or the speaker’s intuitions suggest that the two verbs are synonymous. This is the case of procedere (to go on) defined as “andare avanti, proseguire il cammino”, that is as hyponym of both andare and proseguire, whereas proseguire itself is defined as “andare avanti” and seems actually to be a synonym of procedere (and can, indeed, be translated also as ‘to go on’). In this and similar cases, synsets were built containing the verbs encountered. 5) Further problems were then encountered in trying to match our verb senses to WN 1.5. Some verbs, indeed, have a correspondent in WN 1.5 only for one (or some) of theis senses. For example, allontanarsi (to go away; to go far) can be put in two synsets: Sense 1 allontanarsi, discostarsi, muoversi, scostarsi -- (to go away) (Luigi si allontanò da quella casa - Luigi went away from that house) Sense 2 allontanarsi, andare lontano -- (to go far) (Non allontanarti troppo - don’t go too far) However there is a problem of correspondence with WN 1.5, since different synsets are found there corresponding to the first sense, but there is no synset exactly corresponding to the second: 3 senses of go away Sense 1 go, go away, depart, travel away -- (travel away from a place into another direction; "Go away before I start to cry"; "When did he go to work today?" "The train departs at noon") Sense 2 leave, go forth, go away => go, go away, depart, travel away -- (travel away from a place into another direction; "Go away before I start to cry"; "When did he go to work today?" "The train departs at noon") Sense 3 disappear, vanish, go away -- (get lost; "He disappeared without a trace") 2 senses of go far Sense 1 go far, go deep -- ("His accomplishments go far") => run, go, pass, lead, extend -- (run or extend between two points; "Service runs all the way to Cranbury"; "His knowledge doesn't go very far") LE2-4003 EuroWordNet 98 EuroWordNet D006: Definition of the links and subsets for verbs Sense 2 arrive, make it, get in, go far => succeed, come through -- (attain success) In still another case, some Italian verbs have no English correspondent and so it is not possible to indicate a direct connection with WN 1.5 (this happens, e. g., for coricarsi, which means “to go to bed”). 6) A problem we have not yet really dealt with is that of multi-words, even if in some cases they have been put within synsets. 4.3.1.2 Searching for the relevant relations The experiment carried out, besides being aimed at verifying problems in building the individual wordnets, had the goal of explicitly defining the relations needed. Thus, in the following points we shall summarize the results obtained with respect to each relation. 1) Those of synonymy/hyperonymy/hyponymy are, of course, the fundamental relations on which the structure of the database will be based. Indeed, they allow to make fundamental inferences: i) a property of a verb is also a property of all its synonyms (apart from the case of antonymy holding between variants, cf. above); ii) a property of a verb is inherited by all its hyponyms, even if in this case it is always somehow restricted, or in some cases overridden, by more specific characteristics displayed by hyponyms. This can be seen by the taxonomy provided in the Appendix: i.e., all the hyponyms of the sense 2 of andare (indicated for the time being as 0_1b), i.e., of the {andare, recarsi, portarsi, trasferirsi} synset have a basic reference to directed motion. The major problem which we shall have to deal with in order to extract and then encode data on these relations will be that of the sense disambiguation which needs to be carried out before extracting relations. However, since words found at low levels in hierarchies have more specific meanings, most manual work will be done to disambiguate top words, while then semi-automatic procedures can be used. 2) The INVOLVED relation seems also very important (as we have stated in section 2) in order to code more detailed semantic information on verbs. Unfortunately, for the time being, and within the scope of the project, the development of the complex procedures needed in order to extract data on it is not generally feasible (cf. § 2.5), however some data already (manually) extracted can be used in our case. For instance, within the Acquilex project an analysis of different taxonomies was carried out for encoding data of this kind; thus the results obtained can be used, by making the necessary changes connected with the characteristics of the relation identified in EuroWordNet. So, for instance, if we take again into consideration some verbs already reported in section 2, we can encode for them the relations indicated below: camminare = “andare a piedi” to walk to go on foot LE2-4003 EuroWordNet 99 EuroWordNet D006: Definition of the links and subsets for verbs INVOLVED_INSTRUMENT --> piedi cavalcare = “andare a cavallo” to ride to go on a horse INVOLVED_INSTRUMENT --> cavallo coricarsi = “andare a letto” to go to bed INVOLVED_TARGET_DIRECTION --> letto uscire = “andare, venire fuori” to go out to go, to come out INVOLVED_TARGET_DIRECTION --> fuori. 3) As far as the other relations are concerned, we have already remarked which strategies can be used to extract data on them (if they are indicated). However, by this first experiment on data some additional hypotheses in this respect have been formulated. For instance, we have found definitions in which a pattern is used to indicate the goal of an action referred to by a verb (per V= in order to V); when a pattern of this kind is present in a definition, a SUBEVENT relation may be identified. This seems the case of passare (to pass, defined as “muoversi attraversando un luogo per andare in un altro” - to move crossing a place to go to another), which could be seen as a subevent of andare (to go).19 Another pattern which can be used to detect a relation between words is per effetto di V/N (as an effect of), which indicates the ‘cause’ of a situation. For instance, beccheggiare (to pitch) is defined as “oscillare di una nave per effetto del moto ondoso” (to swing (said) of a ship as an effect of waves motion): therefore we may state a causal relation between the waves motion and the event. Also in these cases, however, data will be encoded only for those verbs for which either we have already extracted information or we may easily extract it. 19 And in fact also the pair buy/pay, which has been used above to give an example of the relation, could be connected by such construction: one has ‘to pay in order to buy’. LE2-4003 EuroWordNet 100 EuroWordNet D006: Definition of the links and subsets for verbs 4.2.2 Know verbs As far as ‘know verbs’ are concerned, in general, problems similar to those already reported with respect to motion verbs were encountered. Furthermore, we do not find much useful information within dictionaries, at least if we only consider synonymy and hyponymy relations. Sapere (to know) and its synonym conoscere, for instance, have generally essere (to be) and avere (to have) as genuses: these only transfer to their hyponyms the indication of a semantic reference to a state, whereas much important semantic information is conveyed by adjectives or phrases following them within definitions. Thus, in order to be able to indicate all the meaning facets of the two verbs, we need links other than simply synonymy or hyponymy. Furthermore, sapere and conoscere refer, in their fundamental sense, to a state, however what’s interesting of these verbs is their ability to refer to a change-of-state when inflected in certain tempo-aspectual forms: a. Giovanni sapeva/conosceva la verità. Giovanni knew the truth (meaning that he was conscious of the truth) b. Improvvisamente Giovanni seppe/conobbe la verità. Suddenly Giovanni knew (discovered) the truth. An indication of such a property is explicitally given in examples found within dictionary definition, but we think that it could be better given by connecting the two verbs to a genus referring to the inception of the awareness. Therefore, in order to extract synonymy and hyponymy relations we tried to intervene quite ‘heavily’ on the data found in dictionaries (either connecting sometimes taxonomies wich are not really connected within the dictionary or changing the status of the relations found: i.e. from hyponymy to synonymy and viceversa). As far as the other relations are concerned, we encountered problems of the kind already discussed for motion verbs. (Cf. the Appendix for the synsets identified and for the taxonomies built.) 4.4 First results of work on Spanish The Spanish approach is mainly based on extraction of information by means of automatic methods. We have two main sources: an MRD: VOX, and WN1.5. As in the case of nouns (see D005 for details), Spanish words will be automatically mapped to WN 1.5 synsets, thus finally obtaining a Spanish Verb lattice somewhat parallell to that of WN 1.5. As an effect, we hypothesize that verb-to-verb relations applying in WN1.5 will be adequate for its Spanish counterpart. Nevertheless a process of manual validation will be obviously needed. This way, WN1.5 will be taken as the main source of information while that extracted from MRD-VOX will be used to refine, improve and modify this basic source. LE2-4003 EuroWordNet 101 EuroWordNet D006: Definition of the links and subsets for verbs 4.4.1 Subset selection of motion verbs and ‘know verbs’ In the Spanish experiment MRDs and derived taxonomies have been used as source for testing potential both knowledge and movement subsets. In principle, verbs of moving should be those being hyponyms of either 'ir', 'mover' and 'moverse'. On the other hand, ‘saber' and 'conocer' have proved to be the best sources for deriving taxonomies to be used as verbs of knowing. These subsets obtained in this first step pose some problems concerning their semantic boundaries specially in the move taxonomy. Often, actions involve a movement of some kind but are not properly "movement actions". In these cases is very difficult to decide even manually if a sense belongs or not to a taxonomy. For movement verbs we have defined a basic criterium in order to help the decision task: A sense belongs to movement taxonomy iff the main purpose of the action is movement This criterium helps us to reject senses like: 'hojear', (to turn the etc.). This criterium has to be applied manually. pages of), 'fregar' (to rub, to wash, 4.4.2. Extraction of relations Hyponym The source for the extraction of hyponym and hyperonym relations is derived from taxonomies from MRD-VOX. We assume that this source is sufficient enough for obtaining these relations. The problem mentioned in the selection of the subset has an effect on the extraction of hyponym relations: a manual process of senses removal is needed and this process will reduce the number of hyponym relations given in the dictionary. Synonym MRD-VOX has a labeled field for this relation, and it can be extracted automatically, but frequently this information does not appear explicitly. For establishing exhaustively this relation others methods had to be used. The best solution should be to have a synonym dictionary available in MRD format, but currently we don't have access to this kind of sources. Another source for extracting synonymy will derive of our approach of automatically mapping Spanish words to WN1.5 synsets. Obviously, words which map onto the same WN1.5 synset ought to be considered synonyms but also in this case all the relations extracted would had to be checked and validated. Antonymy LE2-4003 EuroWordNet 102 EuroWordNet D006: Definition of the links and subsets for verbs Like synonymy, antonymy can be extracted automatically from MRD. There is a specific field in MRD for this relation, but on the contrary of nouns, in verb entries there are very few antonyms. For example, in the selected subsets there are only 3 antonym relations encoded in MRD. This fact forces us to raise again how to extract this relation. A possible solution could be to apply a morphological analysis. The derivational Spanish process allows to deduce antonym relations from pairs like 'conocer' / 'desconocer' (to know / not to know), where the preffix 'des-' (and also 'a-', 'in-', etc.) means the opposite of 'conocer'. Has_Subevent & Cause Has_Subevent and Cause are the most difficult relations for automatic extraction. For Spanish there is not information about them in the sources available. It seems that for detecting these relations some world knwoledge is needed, therefore the main way for establishing them has to be manually. As pointed out above it must be noticed that Has_Subevent and Cause relations which apply between two English synsets are expected to keep on working between the Spanish synsets which are their synonyms. Therefore, the main approach for building the Spanish WordNet, namely mapping from Spanish to WN 1.5 synsets, will itself be a source of detection of Entailment and Cause relations. Notwithstanding, relations thus stated will need to undergo a process of manual validation Involvement relations The Involvement_agent can be extracted from several definitions patterns. The patterns we find for agent are the following: 'Persona que' +V 'Individuo que' +V 'Hombre que' +V 'Mujer que' +V (person that) (person that) (man that) (woman that) In these patterns the nouns (Persona, individuo,etc.) are the genus term. In this cases, the entry is the agent of the action referred by the verb. The Involvement_instrument relation can be extracted from intrument taxonomy. The senses of this taxonomy are in role instrument relationship with the verb that appears in their definitions after the pattern: LE2-4003 EuroWordNet 103 EuroWordNet D006: Definition of the links and subsets for verbs 'que sirve para' +V (used for) 'que se usa para' +V (") 'para' +V (") 'usado para' +V (") This verb has always a finite form. Summary of motion verbs relations: tops senses hyponyms IR MOVER MOVERSE synonyms 227 226 163 154 TOTAL antonyms 37 162 153 544 20 16 541 0 0 0 73 0 Summary of know verbs relations tops senses saber conocer 18 TOTAL LE2-4003 hyponyms 3 21 synonyms antonyms 2 0 0 17 2 2 19 2 2 EuroWordNet 104 EuroWordNet D006: Definition of the links and subsets for verbs 5. Appendix 5.1 List of tests for identifying the relations in each language 5.1.2 Tests for Dutch Dutch test 1 Comment: Score yes a yes b Conditions: Example: a b Effect: Dutch test 2 Comment: Score yes a yes b Conditions: Example: a b Effect: LE2-4003 Synonymy between verbs Test sentence als iets/iemand/het Xt dan Yt iets/iemand/het (if something/someone/it Xs the something/someone/it Ys) als iets/iemand/het Yt dan Xt iets/iemand/het (if something/someone/it Ys the something/someone/it Xs) - X is a verb in the third person singular form - Y is a verb in the third person singular form - there are no specifying PPs that apply to the X-phrase or the Y-phrase als iets/iemand/het begint dan start iets/iemand/het als iets/iemand/het start dan begint iets/iemand/het X = begint (begins) Y = start (starts) synset variants {beginnen V, starten V} Synonymy of nouns and verbs denoting events and states Test sentence Als er sprake is van een X dan Yt er iets/iemand/het (if there is a case of a/an X then something/someone/it Ys) als iets/iemand/het Yt dan is er sprake van een X (if something/someone/it Ys then there is a case of a/an X) - X is a noun in the singular - Y is a verb in the third person singular form - there are no specifying PPs that apply to the X-phrase or the Y-phrase - preferably there is a morphological link between the noun and the verb Als er sprake is van een beweging dan beweegt er iets als iets beweegt dan is er sprake van een beweging X = beweging (movement) Y = beweegt (move) beweging N XPOS_NEAR_SYNONYM bewegen V bewegen V XPOS_NEAR_SYNONYM beweging N EuroWordNet 105 EuroWordNet D006: Definition of the links and subsets for verbs The distinction between hyponymy and synonymy is not always clear-cut. Sometimes concepts can be very close showing only a very limited specialization. In the case of relations across part-of-speech we can at least formulate the extra conditions that a strong morphological link between the two words is preferred, which is the case of “beweging” (movement) and “bewegen” (move). A specific test elicits a stronger intuition than a more general test. In general we therefore try to formulate more specific tests in addition to the more general test. As a rule of thumb first the specific test should be applied. If this yields a positive score than the relation can be assigned. If it yields a negative score we can apply the general test. Whereas the above test works both for non-dynamic states and dynamic events, the next test only applies to dynamic events: Dutch test 3 Comment: Score yes a yes b Conditions: Example: a b Effect: LE2-4003 Synonymy between event denoting nouns and verbs Test sentence als er een X plaatsvindt dan Yt er iets/iemand/het (if an X takes place then something/somebody/it Ys) als iets/iemand/het Yt dan vindt er een X plaats (if something/somebody/it Ys then a/an X takes place) - X is a noun in the singular - Y is a verb in the third person singular form - there are no specifying PPs that apply to the X-phrase or the Y-phrase - preferably there is a morphological link between the noun and the verb als er een beweging plaatsvindt dan beweegt er iets/iemand als iets/iemand beweegt dan vindt er een beweging plaats X = beweging (movement) Y = beweegt (move) beweging N XPOS_NEAR_SYNONYM bewegen V bewegen V XPOS_NEAR_SYNONYM beweging N EuroWordNet 106 EuroWordNet D006: Definition of the links and subsets for verbs The next test only applies to non-dynamic states expressed by nouns and verbs: Dutch test 4 Comment: Score yes a yes b Conditions: Example: a b Effect: Example: a b Effect: Synonymy between state-denoting nouns and verbs Test sentence Als er sprake is van een toestand van X dan Yt iets/iemand/het (if there is a state of X then something/someone/it Ys) als iets/iemand/het Yt dan is er sprake van een toestand van X (if something/something/it Ys then there is a state of a/an X applies) - X is a noun in the singular - Y is a verb in the third person singular form - there are no specifying PPs that apply to the X-phrase or the Y-phrase - preferably there is a morphological link between the noun and the verb Als er sprake is van een toestand van slaap dan slaapt er iets als iets slaapt dan is er sprake van een toestand van slaap X = slaap (sleep) Y = slaapt (sleeps) slaap N XPOS_NEAR_SYNONYM slapen V slapen V XPOS_NEAR_SYNONYM slaap N Als er sprake is van een toestand van gelijkenis dan lijkt er iets als iets lijkt dan is er sprake van een toestand van gelijkenis X = gelijkenis (equality, similarity) Y = lijkt (equals, is similar) gelijkenis N XPOS_NEAR_SYNONYM lijken V lijken V XPOS_NEAR_SYNONYM gelijkenis N The next test elicits synonymy between adjectives and verbs that denote non-dynamic states: LE2-4003 EuroWordNet 107 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 5 Comment: Score yes a yes b Conditions: Example: a b Effect: Synonymy between state-denoting verbs and adjectives Test sentence Als iets/iemand/het Xt dan is iets/iemand/het Y (if something/someone/it Xs then something/someone/it is Y) als iets/iemand/het Y is dan Xt iets/iemand/het (if something/someone/it is Y then something/someone/it Xs) - X is a verb in the third person singular - Y is an adjective - there are no specifying PPs that apply to the X-phrase or the Y-phrase - preferably there is a morphological link between the noun and the verb Als iets/iemand/het vreest dan is iets/iemand/het bang als iets/iemand/het bang is dan vreest iets/iemand/het X = bang (afraid) Y = vreest (fear) vrezen V XPOS_NEAR_SYNONYM bang A bang Aq XPOS_NEAR_SYNONYM vrezen V For hyponymy relations between verbs a separate test is used which makes the notion of troponym (a more specific way of) explicit as defined by Fellbaum (1991): Dutch test 6 Comment: Score yes a no b Conditions: Example: a b Effect: Hyponymy-relation between verbs as troponyms Test sentence als iets/iemand/het Xt dan Yt iets/iemand/het op een bepaalde manier (if something/someone/it Xs the something/someone/it Ys in a particular way) als iets/iemand/het Yt dan Xt iets/iemand/het op een bepaalde manier (if something/someone/it Ys the something/someone/it Xs in a particular way) - X is a verb in the third person singular form - Y is a verb in the third person singular form - there is a manner-PP that has to be added to the Y-phrase to match it with the X-phrase als iets/iemand/het zwemt dan beweegt iets/iemand/het op een bepaalde manier als iets/iemand/het beweegt dan zwemt iets/iemand/het op een bepaalde manier X = zwemt (swims) Y = beweegt (moves) zwemmen V HAS_HYPERONYM bewegen V bewegen V HAS_HYPONYM zwemmen V In addition to specific manners of events verbs can also refer to more specific events covering any other features. These features may be expressed by any additional PP or specification. A verb such as “vertrekken” (to leave) should rather be defined as “to move away from something” than by “to move LE2-4003 EuroWordNet 108 EuroWordNet D006: Definition of the links and subsets for verbs in a certain way”. Following this the Dutch test sentence 6 can also be re-formulated in a more general way where the additional PP can express any aspect and not just manner: Dutch test 6 (revised) Comment: Hyponymy-relation between verbs reflecting any specification Score Test sentence yes a als iets/iemand/het Xt dan Yt NP (NP) PP (if something/someone/it Xs the something/someone/it Ys PP) no b als iets/iemand/het Yt dan Xt iets/iemand/het PP (if something/someone/it Ys the something/someone/it Xs PP) Conditions: - X is a verb in the third person singular form - Y is a verb in the third person singular form - there are is at least one specifying NP or PP that has to be added to the Y-phrase to match it with the X-phrase Example: a als iets/iemand/het vertrekt dan beweegt iets/iemand/het van iets vandaan (if something/somebody/it leaves then it moves from something away) b als iets/iemand/het beweegt dan vertrekt iets/iemand/het van iets vandaan (if something/someone/it moves then it leaves from something away) X = vertrekt (leaves ) Y = beweegt (moves) Effect: vertrekken V HAS_HYPERONYM bewegen V bewegen V HAS_HYPONYM vertrekken V LE2-4003 EuroWordNet 109 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 7 Comment: Score yes/no a no/yes b Conditions: Example: a b Effect: Example: a b Effect: Hyponymy of nouns and verbs denoting events and states Test sentence als er sprake is van een X dan Yt (er) iets/iemand/het + NP, PP (op een bepaalde wijze) (if there is a case of a/an X then something/someone/it Ys + NP, PP (in a certain way)) als iets/iemand/het Yt dan is er sprake van een bepaalde X (if something/someone/it Ys then there is a case of a/an certain X) - X is a noun in the singular - Y is a verb in the third person singular form - there should be at least one specifying NP or PP that makes the Y-phrase equivalent to the X-phrase or the other way around. - preferably there is no morphological link between the noun and the verb als er sprake is van een moord dan doodt iets/iemand/het op ‘illegale’ wijze ?als iets/iemand/het doodt dan is er sprake van een moord X = moord (murder) Y = doodt(kills) moord N HAS_XPOS_HYPERONYM doden V doden V HAS_XPOS_HYPONYM moord N ?als er sprake is van een sport dan voetbalt er iets op een bepaalde wijze als iets/iemand/het voetbalt dan is er sprake van een bepaalde sport X = sport (sport) Y = voetbalt (plays football) sport N HAS_XPOS_HYPONYM voetballen V voetballen V HAS_XPOS_HYPERONYM sport N Here we see that the reverse of the score leads to a reversion of the hyponymy as well: noun-to-hyperonym-verb or verb-to-hyperonym-noun. As long as one direction has a clear positive score and the other direction has a clear negative score we are dealing with a hyponymy relation. The standard “zijn” (be) construction for hyponymy can be extended with a manner-specification to elicit troponymy relations between nouns and verbs: LE2-4003 EuroWordNet 110 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 8 Comment: Score yes/no a no/yes b Conditions: Example: a b Effect: Example: a b Effect: Manner-hyponymy or troponymy between nouns and verbs Test sentence (een/het/de) X is een wijze/manier/vorm van Y ((a/an/the) X is a way/manner of Y) Y is een bepaalde X (Y is a certain X) - X is a noun in the singular - Y is a verb in the infinitive form - preferably there is no morphological link between the noun and the verb voetbal is een manier van sporten ?sporten is een bepaald voetbal X = voetbal (football) Y = sporten (sport) voetbal N HAS_XPOS_HYPERONYM sporten V sporten V HAS_XPOS_HYPONYM voetbal N ?een sport is een manier van voetballen voetballen is een bepaalde sport X = sport (sport) Y = voetballen (play football) sport NHAS_XPOS_HYPONYM voetballen V voetballen V HAS_XPOS_HYPERONYM sport N Whereas the previous tests works both for non-dynamic states and dynamic events, the next test only applies to dynamic events: LE2-4003 EuroWordNet 111 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 9 Comment: Score yes/no a no/yes b Conditions: Example: a b Effect: Hyponymy between event denoting nouns and verbs Test sentence als er (een/het/de) X plaatsvindt dan Yt (er) iets/iemand/het + NP, PP (op een bepaalde wijze) (if (a/an/the) X takes place then something/somebody/it Ys + NP, PP (in a certain way)) als iets/iemand/het Yt dan vindt er (een/het) bepaald(e) X plaats (if something/somebody/it Ys then (a/an/the) certain X takes place) - X is a noun in the singular - Y is a verb in the third person singular form - there should be at least one specifying NP or PP that makes the Y-phrase equivalent to the X-phrase or the other way around. - preferably there is no morphological link between the noun and the verb als er gemompel plaatsvindt dan spreekt er iets/iemand op een bepaalde manier ?als iets/iemand spreekt dan vindt er bepaald gemompel plaats X = gemompel (murmuring) Y = spreekt (speaks) gemompel N HAS_XPOS_HYPERONYM spreken V sprekenV HAS_XPOS_HYPONYM gemompel N The next test only applies to non-dynamic states expressed by nouns and verbs: LE2-4003 EuroWordNet 112 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 10 Comment: Score yes/no a no/yes b Conditions: Example: a b Effect: Hyponymy between state-denoting nouns and verbs Test sentence als er sprake is van een toestand van X dan Yt (er) iets/iemand/het + NP, PP (op een bepaalde wijze) (if there is a state of X then something/someone Ys + NP, PP (in a certain way)) als iets/iemand/het Yt dan is er sprake van een toestand van bepaalde X (if something/something Ys then there is a state of a/an certain X applies) - X is a noun in the singular - Y is a verb in the third person singular form - there should be at least one specifying NP or PP that makes the Y-phrase equivalent to the X-phrase or the other way around. - preferably there is no morphological link between the noun and the verb als er sprake is van een toestand van clubliefde dan houdt iermand van een club (if there is a state of club-loving then somebody loves a club) ?als iemand van een club houdt dan is er sprake van een toestand van bepaalde clubliefde (?if someone loves a club then there is a state of certain club-loving) X = clubliefde (love for a club N) Y = houden van (love V) clubliefde N HAS_XPOS_HYPERONYM houden van V houden van V HAS_XPOS_HYPONYM clubliefde N The next test elicits hyponymy between adjectives and verbs that denote non-dynamic states: LE2-4003 EuroWordNet 113 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 11 Comment: Score yes/no a no/yes b Conditions: Example: a b Effect: Hyponymy between state-denoting verbs and adjectives Test sentence Als iets/iemand/het Xt dan is iets/iemand/het Y (if something/someone/it Xs then something/someone/it is Y) als iets/iemand/het Y is dan Xt iets/iemand/het + NP, PP (op een bepaalde wijze) (if something/someone/it is Y then something/someone/it Xs + NP, PP (in a certain way)) - X is a verb in the third person singular - Y is an adjective -- there is at least one specifying adverb, NP or PP that applies to the X-phrase or the Y-phrase - preferably there is no morphological link between the noun and the verb Als iets/iemand/het spaart dan is iets/iemand/het zuining met geld (sparing with respect to money) ?als iets/iemand/het zuining is dan spaart iets/iemand/het op een bepaalde wijze X = spaart (save up) Y = zuinig (sparing) sparen V HAS_XPOS_HYPONYM zuinig A zuinig A HAS_XPOS_HYPERONYM sparen V The next test elicits antonymy with an additional clause to verify that the word-pairs are within a reasonable denotational range: LE2-4003 EuroWordNet 114 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 12 Comment: Score yes a yes b Conditions: Example: a b Effect: Antonymy of verbs and nouns Test sentence (een/de/het) X en Y zijn beide een soort/vorm van Z maar (een) X is het tegenovergestelde van (een) Y (X and Y are both a kind of Z but X is the opposite of Y) the converse of a) - X and Y are singular or plural nouns, or verbs in the infinitive form - Z is a hyperonym of both X and Z beginnen en stoppen zijn beide een vorm van doen maar beginnen is het tegenovergestelde van stoppen (to begin and to stop are both a way of acting but to begin is the opposite of to stop) stoppen en beginnen zijn beide een vorm van doen maar stoppen is het tegenovergestelde van beginnen (to stop and to begin are both a way of acting but to stop is the opposite of to begin ) X = begint(begins) Y = stopt (stop) Z = doen (act) beginnen V ANTONYM stoppen V stoppen V ANTONYM beginnen V No separate tests are needed for NEAR_ANTONYM (which holds between synsets). XPOS_NEAR_ANTONYM can be tested by compensating for the difference in part-of-speech. The above test can at least be used for cross-part-of-speech relations between nouns and verbs. LE2-4003 EuroWordNet 115 EuroWordNet D006: Definition of the links and subsets for verbs The next test elicits a factive causation relation (Lyons 1977), where the reversal of the relation is automatically generated (not necessarily implied, hence the label reversed): Dutch test 13 Comment: Score yes a no b Conditions: Example: a b Effect: LE2-4003 Factive causation relation between verbs Test sentence X veroorzaakt Y (X causes Y) X heeft Y tot gevolg (X has Y as a consequence) X leidt tot Y (X leads to Y) the converse of a) X is a verb in the infinitive form Y is a verb in the infinitive form doden veroorzaakt sterven doden heeft sterven tot gevolg doden leidt tot sterven ?sterven veroorzaakt doden ?sterven heeft doden tot gevolg ?sterven leidt tot doden X = doden (kill) Y = sterven (die) doden V CAUSES sterven V sterven V IS_CAUSED_BY doden V reversed EuroWordNet 116 EuroWordNet D006: Definition of the links and subsets for verbs Non-factive causation is elicited by tests that express a more complex modal relation. In that case both relations receive the label non-factive. Note that one of the relations can still be the result of reversing the relation, hence it also has the label reversed: Dutch test 14 Comment: Score yes a no b Conditions: Example: a b Effect: LE2-4003 Non-factive causation relation between verbs using a modal auxiliary Test sentence X kan Y veroorzaken (X may cause Y) X kan Y tot gevolg hebben (X may have Y as a consequence) X kan leiden tot Y (X may lead to Y) the converse of a) X is a verb in the infinitive form Y is a verb in the infinitive form zoeken kan vinden veroorzaken zoeken kan vinden tot gevolg hebben zoeken kan leiden tot vinden ? vinden kan zoeken veroorzaken ? vinden kan zoeken tot gevolg hebben ? vinden kan zoeken tot vinden X = zoeken (search) Y = vinden (find) zoeken V CAUSES vinden V non-factive vinden V IS_CAUSED_BY zoeken V reversed, non-factive EuroWordNet 117 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 15 Comment: Score yes a no b Conditions: Example: a b Effect: Example: a b Effect: LE2-4003 Non-factive causation relation between verbs using a modal constructions Test sentence X is doen/laten Y (X is do/let Y) X is Y doen/laten plaatsvinden (X is Y do/let take place) X is Y mogelijk maken (X is to make Y possible) the converse of a) X is a verb in the infinitive form Y is a verb in the infinitive form wijsmaken is doen/laten geloven wijsmaken is geloven doen/laten plaatsvinden wijsmaken is geloven mogelijk maken ?geloven is doen/laten wijsmaken ?geloven is wijsmaken doen/laten plaatsvinden ?geloven is wijsmaken mogelijk maken X = wijsmaken (fool, make believe) Y = geloven (believe) wijsmaken V CAUSES geloven V non-factive geloven V IS_CAUSED_BY wijsmaken V non-factive, reversed tonen is doen/laten zien tonen is zien doen/laten plaatsvinden tonen is zien mogelijk maken ?zien is doen/laten tonen ?zien is tonen doen/laten plaatsvinden ?zien is tonen mogelijk maken X = tonen (show) Y = zien (see) tonen V CAUSES zien V non-factive zien V IS_CAUSED_BY tonen V non-factive,reversed EuroWordNet 118 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 16 Comment: Score yes a no b Conditions: Example: a b Effect: Non-factive causation relation between verbs using try Test sentence X is proberen/trachten te Y (X is to try Y) the converse of a) X is a verb in the infinitive form Y is a verb in the infinitive form zoeken is proberen/trachten te vinden ?vinden is proberen/trachten te zoeken X = zoeken (search) Y = vinden (find) zoeken V CAUSES vinden V non-factive vinden V IS_CAUSED_BY zoeken V non-factive, reversed As explained above, causal relations can hold between a dynamic cause and a result which can either be dynamic or static. Information on the dynamicity of the result is however not expressed in terms of different cause relations but can be inferred from the hyponymy-relations that hold for the result. In this respect we can thus have nouns, verbs and adjectives as results of CAUSES, where the latter can only denote non-dynamic results and nouns and verbs can both denote dynamic or non-dynamic results. The tests are therefore not differentiated for the dynamicity of the result. On the other hand, they do have to be differentiated for the part-of-speech to get well-formed test-sentences. The next test can be used to elicit a causal relation between a verb or noun on the one hand and a resulting adjective/adverb on the other hand: LE2-4003 EuroWordNet 119 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 17 Comment: Score yes a Conditions: Example: a Effect: Example: Effect: LE2-4003 a Actions and Events resulting in a state denoted by an adjective Test sentence als er sprake is van X dan wordt iets/iemand/het Y (if it is the case that X then something/someone/it becomes Y) X is a noun or verb denoting an event Y is an adjective, possibly in a comparative form als er sprake is van genezen dan wordt iets/iemand/het beter X = genezen (heal) Y = beter (better, healthy) genezen V CAUSES beter A beter A IS_CAUSED_BY genezen V reversed als er sprake is van activeren dan wordt iets/iemand/het actief X = activeren (activate) Y = actief (active) activeren V CAUSES actief A actief A IS_CAUSED_BY activeren V reversed EuroWordNet 120 EuroWordNet D006: Definition of the links and subsets for verbs In those cases that the resulting state is denoted by a noun a separate test is needed to deal with the nominal form. Since the nominalization also makes the dynamicity explicit we need a separate test for nominal results which are static and nominal results which are dynamic: Dutch test 18 Comment: Score yes a Actions and Events resulting in a state denoted by a noun Test sentence als er sprake is van X dan komt iets/iemand/het in een toestand van Y (if it is the case that X then something/someone/it gets in a state of Y) X is a noun or verb denoting an event Y is an noun denoting a state als er sprake is van doden dan komt iets/iemand/het in een toestand van dood X = doden (kill) Y = dood (death) doden V CAUSES dood N dood N IS_CAUSED_BY doden V reversed Conditions: Example: a Effect: Dutch test 19 Comment: Score yes a Conditions: Example: Effect: a Actions and Events resulting in an event denoted by a noun Test sentence als er sprake is van X dan vindt Y plaats als gevolg. (if it is the case that X then Y takes place as a consequence) X is a noun or verb denoting an event Y is an noun denoting an event als er sprake is van doden dan vindt sterfte plaats als gevolg X = doden (kill) Y = sterfte (the dying) doden V CAUSES sterfte N sterfte N IS_CAUSED_BY doden V reversed Note that the difference in dynamicity between the nouns “sterfte” (the dying) and “dood” (death) is not expressed by the CAUSE relation but has to follow from the hyponymy-relations with “verandering” (change) and “toestand” (state), respectively. Another interesting aspect of the Dutch situation is that “dood” is also an adjective (meaning “dead”) which is synonymous with the noun. LE2-4003 EuroWordNet 121 EuroWordNet D006: Definition of the links and subsets for verbs The following tests elicit verb-meronymy relations, which are stored as HAS_SUBEVENT/ IS_SUBEVENT_OF relations. Automatically reversed relations which do not necessarily hold have again been labelled with the the label reversed: Dutch test 20 Comment: Score yes a no b Conditions: Example: a b Effect: Subevent relation between verbs using a sub-process construction Test sentence om te X moet je Y (in order to X you have to Y) the converse of a) X is a verb in the infinitive form Y is a verb in the infinitive form om te kopen moet je betalen ?om te betalen moet je kopen X = kopen (buy) Y = betalen (pay) kopen V HAS_SUBEVENT betalen V betalen V IS_SUBEVENT_OF kopen V reversed Another way of eliciting subevent relations is by temporal inclusion of events within the embedding event: LE2-4003 EuroWordNet 122 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 21 Comment: Score yes a no b Conditions: Example: a b Effect: Example: a b Effect: Subevent relation between verbs using temporal inclusion Test sentence X vindt plaats tijdens/gedurende/onder Y (X occurs during/while Y) the converse of a) X is a verb in the infinitive form Y is a verb in the infinitive form bedreigen vindt plaats tijdens/gedurende afpersen ?afpersen vindt plaats tijdens/gedurende het bedriegen X = bedreigen (threaten) Y = afpersen (blackmail) afpersen V HAS_SUBEVENT bedreigen V bedreigen V IS_SUBEVENT_OF afpersen V reversed dromen vindt plaats tijdens/gedurende slapen ?slapen vindt plaats tijdens/gedurende het dromen X = dromen (dream) Y = slapen (sleep) dromen V IS_SUBEVENT_OF slapen V slapen V HAS_SUBEVENT dromen V reversed In the case of the tempotral inclusion test, the semantic anomaly of the converse sentence b) is not very strong. To make the relation more explicit this test should be combined with an implicational test such as the following if-then test: LE2-4003 EuroWordNet 123 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 22 Comment: Score yes a no b Conditions: Example: a b Effect: Example: a b Effect: LE2-4003 Subevent relation between verbs using if/then implication Test sentence als X plaatsvindt dan vindt ook Y plaats (if X takes place then also Y takes place) the converse of a) X is a verb in the infinitive form Y is a verb in the infinitive form als afpersen plaatsvindt dan vindt ook bedreigen plaats ?als bedreigen plaatsvindt dan vindt ook afpersen plaats X = afpersen (blackmail) Y = bedreigen (threaten) afpersen V HAS_SUBEVENT bedreigen V bedreigen V IS_SUBEVENT_OF afpersen V reversed als dromen plaatsvindt dan vindt ook slapen plaats ?als slapen plaatsvindt dan vindt ook dromen plaats X = dromen (dream) Y = slapen (sleep) dromen V IS_SUBEVENT_OF slapen V slapen V HAS_SUBEVENT dromen V reversed EuroWordNet 124 EuroWordNet D006: Definition of the links and subsets for verbs Parallel to the meronymy tests for nouns we can use compositional structures for eliciting subevent relations: Dutch test 23 Comment: Score yes a Subevent relation between verbs and nouns using compositional structures Test sentence (een/de/het) X bestaat uit (een/de/het) Y en andere gebeurtenissen/handelingen (an/a/the X consists of an/a/the Y and other events/actions) the converse of a) X and Y are dynamic nouns or verbs in the infinitive form autorijden bestaat uit sturen en andere handelingen ?sturen bestaat uit autorijden en andere handelingen X = autorijden (drive a car) Y = sturen (steer) autorijden V HAS_SUBEVENT sturen V sturen V IS_SUBEVENT_OF autorijden V reversed vergaderen bestaat uit besluitvorming en andere handelingen ? besluitvorming bestaat uit vergaderen en andere handelingen X = vergaderen (to assemble) Y = opening decison-making) vergaderen V HAS_SUBEVENT besluitvorming N besluitvorming N IS_SUBEVENT_OF vergaderen V reversed no b Conditions: Example: a b Effect: Example: a b Effect: The next test elicits a general involvement relation, irrespective of the specific role. The example of “hamer” (hammer) and “timmeren” (to hammer) shows that specific involvements also pass the general test: Dutch test 24 Comment: Score yes a Conditions: Example: Effect: a Involvement/role relation in general Test sentence (een) X is degene/dat die/wat bij het Y betrokken is ((a/an) X is the one/that who/which is involved in Ying) X is a noun Y is a verb in the infinitive form een hamer is dat wat bij het timmeren betrokken is X = hamer (hammer) Y = timmeren (hammer) hamer N ROLE timmeren V timmeren V INVOLVED hamer N reversed Note that in the case of these role-relations it does not make sense to reverse the test because a verb can never occupy the position of the concrete noun nor can the concrete noun take the position of the verb. The b) sentences have therefore been omitted. Furthermore, as with the previous cases we see here that LE2-4003 EuroWordNet 125 EuroWordNet D006: Definition of the links and subsets for verbs the reversal of the relation again leads to a generated implication which may not necessarily hold: a “hamer” (hammer) is not necessarily involved in an event of “timmeren” (hammer). The label reversed can thus automatically be inserted for the reverse relations of these roles. The next tests can then be used to elicit more specific roles/involvements. Dutch test 25 Comment: Score yes a Conditions: Example: a Effect: Example: a Effect: Dutch test 26 Comment: Score yes a Conditions: Example: a Effect: Dutch test 27 Comment: Score yes a LE2-4003 Agent Involvement/role Test sentence (een) X is degene/dat die/wat het Y uitvoert/doet ((a/an) X is the one/that who/which executes/does the Ying) X is a noun Y is a verb in the infinitive form een timmerman is degene die het timmeren uitvoert (a carpenter is the one who executes the hammering) X = timmerman (carpenter) Y = timmeren (hammer) timmerman N ROLE_AGENT timmeren V timmeren V INVOLVED_AGENT timmerman N reversed een bedreiging is datgene dat het dreigen doet (a threat is that which does the threatening) X = bedreiging (a threat) Y = dreigen (to threat) bedreiging N ROLE_AGENT dreigen V dreigen V INVOLVED_AGENT bedreiging N reversed Patient Involvement/role Test sentence (een) X is degene/dat die/wat het Y ondergaat ((a/an) X is the one/that who/which is undergoing the Ying) X is a noun Y is a verb in the infinitive form een spijker is dat wat het timmeren ondergaat X = spijker (nail) Y = timmeren (hammer) spijker N ROLE_ PATIENT timmeren V timmeren V INVOLVED_ PATIENT spijker N reversed Instrument Involvement/role Test sentence (een) X is het instrument waarmee het Y wordt gedaan ((a/an) X is the instrument with which the Ying happens) EuroWordNet 126 EuroWordNet D006: Definition of the links and subsets for verbs Conditions: Example: a Effect: X is a noun Y is a verb in the infinitive form (een) hamer is het instrument waarmee het timmeren wordt gedaan X = hamer (hammer) Y = timmeren (hammer) hamer N ROLE_ INSTRUMENT timmeren V timmeren V INVOLVED_ INSTRUMENT hamer N reversed For those things which are used as instruments but which cannot be called an “instrument” another test can be used which avoids the phrase “instrument”: Dutch test 28 Comment: Score yes a Conditions: Example: a Effect: Dutch test 29 Comment: Score yes a Conditions: Example: Effect: LE2-4003 a Instrument Involvement/role without usinmg the phrase “instrument” Test sentence (een) X is wat gebruikt wordt voor het Y ((a/an) X is what is used for the Ying) X is a noun Y is a verb in the infinitive form inkt is wat gebruikt wordt voor het schrijven X = inkt (ink) Y = schrijven (write) inkt N ROLE_ INSTRUMENT schrijven V schrijven V INVOLVED_ INSTRUMENT inkt N reversed Location Involvement/role Test sentence (een) X is de plaats waar het Y plaatsvindt ((a/an) X is the place where the Ying happens) X is a noun Y is a verb in the infinitive form een school is de plaats waar het onderwijzen plaatsvindt X = school (school) Y = onderwijzen (teach) school N ROLE_ LOCATION onderwijzen V onderwijzen V INVOLVED_ LOCATION school N reversed EuroWordNet 127 EuroWordNet D006: Definition of the links and subsets for verbs Dutch test 30 Comment: Score yes a Conditions: Example: a Effect: Dutch test 31 Comment: Score yes a Conditions: Example: Effect: LE2-4003 a Source-Location Involvement/role Test sentence (een) X is de plaats waarvandaan het Y plaatsvindt ((a/an) X is the place where from the Ying happens) X is a noun Y is a verb in the infinitive form de start is de plaats waarvandaan het racen plaatsvindt X = start (start) Y = racen (race) start N ROLE_ SOURCE_ LOCATION racen V racen V INVOLVED_ SOURCE_ LOCATION start N reversed Target-Location Involvement/role Test sentence (een) X is de plaats waarnaartoe het Y plaatsindt ((a/an) X is the place where to the Ying takes place) X is a noun Y is a verb in the infinitive form de finish is de plaats waarnaartoe het racen plaatsvindt X = finish (finish) Y = racen (race) finish N ROLE_ TARGET _ LOCATION racen V INVOLVED_ TARGET_ LOCATION racen V finish N reversed EuroWordNet 128 EuroWordNet D006: Definition of the links and subsets for verbs The final test elicits properties of events: Dutch test 32 Comment: Score yes a Conditions: Example: a Effect: Actions and Events being in a state Test sentence als er sprake is van X dan gebeurt iets/iemand/het Y (if it is the case that X then something/someone/it happens Y X is a noun or verb denoting an event Y is an adjective, adverb or PP expressing a property of the event als er sprake is van haasten dan gebeurt iets/iemand/het snel/haastig X = haasten (hurry) Y = snel (quick, quickly) haasten V BE_IN_STATE snel A snel A STATE_OF haasten V reversed 5.1.2 Tests for Italian In the following specific criteria for identification of relations with respect to Italian are provided. In some cases, i.e. when tests are either identical to English ones or to similar to other tests for Italian, we may omit them. Italian test (1) Comment: Score yes a yes b Conditions: Example: a b Effect: LE2-4003 Synonymy between verbs Test sentence se qlco./qlcu. X allora qlco./qlcu. Y se qlco./qlcu. Y allora qlco./qlcu. X X is a verb in the third person singular form Y is a verb in the third person singular form there are no specifying PPs that apply to the X-phrase or the Y-phrase se qlco./qlcu. comincia allora qlco./qlcu. inizia se qlco./qlcu. inizia allora qlco./qlcu. comincia synset variants: {cominciare, iniziare} EuroWordNet 129 EuroWordNet D006: Definition of the links and subsets for verbs Italian Test (2) Comment: Score yes-probably a yes/yes yes-probably b yes/yes Conditions: Italian Test (3) Comment: Score yes a yes b Conditions: Example: a b Effect: LE2-4003 Near-Synonymy between verbs Test sentence se qlco./qlcu. X allora qlco./qlcu. Y e i) Xh e Yh sono differenti o ii) Hx e Hy sono differenti se qlco./qlcu. Y allora qlco./qlcu. X e i) Yh e Xh sono differenti o ii) Hy e Hx sono differenti X and Y are verbs in the third person singular form X and Y are not linked by any other relation Xh is the set of the hyponyms (if there are any) of X Yh is the set of the hyponyms (if there are any) of Y Hx is the set of the hyperonyms (if there are any) of X Hy is the set of the hyperonyms (if there are any) of Y Near-Synonymy between state-denoting verbs and adjectives Test sentence se qlco./qlcu. X allora qlco./qlcu. è Y se qlco./qlcu. è Y allora qlco./qlcu. X X is a verb in the third person singular form Y is an adjective there are no specifying PPs that apply to the X-phrase or the Y-phrase preferably there is a morphological link between the noun and the verb se qlco. vive allora qlco. è vivo se qlco. è vivo allora qlco. vive {vivere} XPOS_NEAR_SYNONYM {vivo} EuroWordNet 130 EuroWordNet D006: Definition of the links and subsets for verbs Italian Test (4) Comment: Score yes a no b Conditions: Example: a b Effect: Italian Test (5) Comment: Score yes a no b Conditions: Example: a b Effect: Italian Test (6) Comment: Score yes a yes b Conditions: Example: a b Effect: Italian Test (7) Hyperonymy/hyponymy between verb synsets Test sentence X è Y + phrase Y è X (phrase) X is a verb in the infinitive form Y is a verb in the infinitive form there is at least one specifying AdvP, AdjP, NP or PP that applies to the Y-phrase correre è andare (velocemente) *andare è correre (velocemente) {correre} HAS_HYPERONYM {andare} {andare} HAS_HYPONYM {correre} Hyponymy between state-denoting verbs and adjectives Test sentence X + phrase è/significa essere Y essere Y è/significa X X is a verb in the infinitive form Y is an adjective there is at least one specifying AdvP, AdjP, NP or PP that applies to the X-phrase preferably there is no morphological link between the noun and the verb Ammalarsi facilmente è essere cagionevole ? Essere cagionevole è ammalarsi {ammalarsi} XPOS_HYPERONYM_OF {cagionevole} {cagionevole} XPOS_HYPONYM_OF {ammalarsi} Antonymy between verb synsets (a)20 Test sentence se qlco./qlcu. X allora qlco./qlcu. non Y se qlco./qlcu. Y allora qlco./qlcu. non X X is a verb in the third person singular form Y is a verb in the third person singular form X and Y are co-hyponyms se qlcu. ingrassa allora qlcu. non dimagrisce se qlcu. dimagrisce allora qlcu. non ingrassa {ingrassare, prendere peso} NEAR_ANTONYM {dimagrire, perdere peso} 20 As discussed in § 2.4, three different antonymy relations can be encoded. However, since the tests developed are all similar, we are only going to provide translations for the ones for synsets antonymy. LE2-4003 EuroWordNet 131 EuroWordNet D006: Definition of the links and subsets for verbs Comment: Score yes a yes b Conditions: Antonymy between verb synsets (b) Test sentence se qlco./qlcu. X allora qlco./qlcu. non Y se qlco./qlcu. Y allora qlco./qlcu. non X X is a verb in the third person singular form Y is a verb in the third person singular form there is a hyperonym of X which is opposite to a hyperonym of Y se qlcu. vende allora qlcu. non compra se qlcu. compra allora qlcu. non vende {vendere} NEAR_ANTONYM {comprare} Example: a b Effect: Italian Test (8) Comment: Score yes a yes b Conditions: Example: a b Effect: Italian Test (9) Comment: Score yes a Conditions: Example: Effect: a Italian Test (10) Comment: Score yes a Conditions: Example: a Effect: Italian Test (11) Comment: Score LE2-4003 Antonymy between verb synsets (c) Test sentence se qlco./qlcu. X allora qlco./qlcu. non Y se qlco./qlcu. Y allora qlco./qlcu. non X X is a verb in the third person singular form Y is a verb in the third person singular form X has an addressee and the addressee is the protagonist of Y se qlcu. dà allora qlcu. non riceve se qlcu. riceve allora qlcu. non dà {dare} NEAR_ANTONYM {ricevere} Involvement relation in general Test sentence un/a X è chi/la cosa che è implicato/a nel Y X is a noun Y is a verb in the infinitive form Un martello è la cosa implicata nel martellare {martellare} INVOLVED {martello} Agent Involvement Test sentence un/a X è chi/la cosa che esegue l’Y-are ( /Y) X is a noun Y is a verb un insegnante è chi (esegue l’insegnare) insegna {insegnare} INVOLVED_AGENT {insegnante} Patient Involvement Test sentence EuroWordNet 132 EuroWordNet D006: Definition of the links and subsets for verbs yes a Conditions: Example: Effect: a Italian Test (12) Comment: Score yes a Conditions: Example (1): Effect: Example (2): Effect: Italian Test (13) Comment: Score yes a Conditions: Example: Effect: a Italian Test (14) Comment: Score yes a Conditions: Example: a Effect: LE2-4003 un/a X è chi/la cosa che è sottoposto all’Y-are (/Y) X is a noun Y is a verb un allievo è chi impara (è sottoposto all’imparare) {imparare} INVOLVED_PATIENT {allievo} Instrument Involvement (also valid for nouns indicating things, used for doing something, which cannot be really called ‘instruments’) Test sentence un/a X è lo strumento/ciò che è usato per Y X is a noun Y is a verb in the infinitive form un martello è lo strumento usato per martellare {martellare} INVOLVED_INSTRUMENT {martello} un natante è ciò che è usato per navigare {navigare} INVOLVED_INSTRUMENT {natante} Location Involvement Test sentence un/a X è il luogo dove si Y X is a noun Y is a verb in the third person singular form una scuola è il luogo dove si insegna {insegnare} INVOLVED_LOCATION {scuola} Direction Involvement Test sentence è possibile Y da o in/ad un luogo Y is a verb in the infinitive form è possibile correre da o in un luogo {correre} INVOLVED_DIRECTION {direzione} EuroWordNet 133 EuroWordNet D006: Definition of the links and subsets for verbs Italian Test (15) Comment: Score yes a Conditions: Example: Effect: a Italian Test (16) Comment: Score yes a Conditions: Example: Effect: a Italian Test (17) Comment: Score yes a no b Conditions: Example: a b Effect: LE2-4003 Source-Direction Involvement Test sentence il/lo/la X è il luogo dal quale si Y X is a noun Y is a verb la patria è il luogo dal quale si emigra {emigrare} INVOLVED_SOURCE {patria} Target-Direction Involvement Test sentence il/lo/la X è il luogo al/nel quale si Y X is a noun Y is a verb la casa è il luogo nel quale si rincasa {rincasare} INVOLVED_TARGET_DIRECTION Factive causation relation between verbs Test sentence X causa Y X ha Y come conseguenza X porta a Y the converse of (a) X is a verb in the infinitive form Y is a verb in the infinitive form uccidere causa (il) morire uccidere ha (il) morire come conseguenza uccidere porta (qualcuno) a morire *morire causa (l’) uccidere *morire ha (l’) uccidere come conseguenza *morire porta (qualcuno) a uccidere {uccidere} CAUSES {morire} {morire} IS_CAUSED_BY {uccidere} {casa} factive reversed EuroWordNet 134 EuroWordNet D006: Definition of the links and subsets for verbs Italian Test (18) Comment: Score yes a no b Conditions: Example: a b Effect: Italian Test (19) Comment: Score yes a no b Conditions: Example: a b Effect: LE2-4003 Non-factive causation relation between verbs using a modal auxiliary Test sentence X può causare Y X può avere Y come conseguenza X può portare a Y the converse of (a) X is a verb in the infinitive form Y is a verb in the infinitive form cercare può causare trovare cercare può avere trovare come conseguenza cercare può portare a trovare ?trovare può causare cercare ?trovare può avere cercare come conseguenza ?trovare può portare a cercare {cercare} CAUSES {trovare} (non-factive) {trovare} IS_CAUSED_BY {cercare} (non-factive) Has_subevent/Is_subevent_Of relation between verbs (/nouns) (a) Test sentence Se X sta avendo luogo allora Y può avere luogo the converse of a) X is a verb in the infinitive form Y is a verb in the infinitive form Se (il) dormire sta avendo luogo allora (il) russare può aver luogo * Se (il) russare sta avendo luogo allora (il) dormire può aver luogo {russare} IS_SUBEVENT_OF {dormire} {dormire} HAS_SUBEVENT {russare} reversed EuroWordNet 135 EuroWordNet D006: Definition of the links and subsets for verbs Italian test (20) Comment: Score yes a no b Conditions: Example: a b Effect: Has_subevent/Is_subevent_Of relation between verbs (/nouns) (b) Test sentence Se Y sta avendo luogo allora X deve aver luogo the converse of a) Y is a verb in the infinitive form X is a verb in the infinitive form Se (il) comprare sta avendo luogo allora (il) pagare deve aver luogo * Se (il) pagare sta avendo luogo allora (il) comprare deve aver luogo {comprare} HAS_SUBEVENT {pagare} {pagare} IS_SUBEVENT_OF {comprare} reversed 5.1.3 Tests for Spanish Spanish test 1 Comment: Score yes a yes b Conditions: Example: a b Effect: Spanish test 2 Comment: Score yes a yes b Conditions: Example: a b Effect: Synonymy between verbs Test sentence Si algo/alguien (se) X, entonces ese algo/alguien (se) Y Si algo/alguien (se) Y, entonces ese algo/alguien (se) X X is a verb in the third person singular form Y is a verb in the third person singular form there are no specifying PPs that apply to the X-phrase or the Y-phrase Si algo empieza, entonces ese algo comienza Si algo comienza, entonces ese algo empieza synset variants: {comenzar V,empezar V} Synonymy of nouns and verbs denoting events and states Test sentence Si se da el caso de un X, entonces algo/alguien (se) Y Si algo/alguien (se) Y, entonces se da el caso de un X. - X is a noun in the singular - Y is a verb in the third person singular form - there are no specifying PPs that apply to the X-phrase or the Y-phrase - preferably there is a morphological link between the noun and the verb Si se da el caso de un movimiento, entonces algo se mueve Si algo se mueve, entonces se da el caso de un movimiento X = movimiento (movement) Y = moverse (move) movimiento N XPOS_NEAR_SYNONYM moverse V moverse V XPOS_NEAR_SYNONYM movimiento N Spanish test 3 LE2-4003 EuroWordNet 136 EuroWordNet D006: Definition of the links and subsets for verbs Comment: Score yes a yes b Conditions: Example: a b Effect: Spanish test 4 Comment: Score yes a yes b Conditions: Example: a b Effect: LE2-4003 Synonymy between event denoting nouns and verbs Test sentence Si tiene lugar un X, entonces algo/alguien (se) Y Si algo/alguien (se) Y, entonces tiene lugar un X - X is a noun in the singular - Y is a verb in the third person singular form - there are no specifying PPs that apply to the X-phrase or the Y-phrase - preferably there is a morphological link between the noun and the verb Si tiene lugar un movimiento, entonces algo/alguien se mueve Si algo se mueve, entonces tiene lugar un movimiento X = movimiento (movement) Y = moverse (move) movimiento N XPOS_NEAR_SYNONYM moverse V moverse V XPOS_NEAR_SYNONYM movimiento N Synonymy between state-denoting nouns and verbs Test sentence Si se da un estado de X, entonces algo/alguien Y Si algo/alguien Y, entonces se da un estado de X - X is a noun in the singular - Y is a verb in the third person singular form - there are no specifying PPs that apply to the X-phrase or the Y-phrase - preferably there is a morphological link between the noun and the verb Si se da un estado de sueño, entonces alguien duerme Si alguien duerme, entonces se daun estado de sueño X = sueño (sleep) Y = dormir (sleep) sueño N XPOS_NEAR_SYNONYM dormir V dormir V XPOS_NEAR_SYNONYM sueño N EuroWordNet 137 EuroWordNet D006: Definition of the links and subsets for verbs Spanish test 5 Comment: Score yes a yes b Conditions: Example: a b Effect: Spanish test 6 Comment: Score yes a no b Conditions: Example: a b LE2-4003 Synonymy between state-denoting verbs and adjectives Test sentence Si alguien/algo X entonces alguien/algo es/está Y Si alguien/algo es/está Y entonces algo/alguien X - X is a verb in the third person singular - Y is an adjective - there are no specifying PPs that apply to the X-phrase or the Y-phrase - preferably there is a morphological link between the noun and the verb Si alguien teme entonces alguien está temeroso Si alguien está temeroso entonces alguien teme X = temer (fear) Y = temeroso (affraid) temer V XPOS_NEAR_SYNONYM temeroso A temeroso A XPOS_NEAR_SYNONYM temer V Hyperonymy/hyponymy between verbs Test sentence Si algo/alguien (se) X, entoces ese algo/alguien (se) Y Si algo/alguien (se) Y, entonces ese algo/alguien (se) X X is a verb in the third person singular form Y is a verb in the third person singular form there is at least one specifying adverb, NP or PP that applies to the X-phrase or the Y-phrase Si alguien corre, entonces ese alguien se mueve (rápidamente) *Si alguien se mueve, entonces ese alguien corre EuroWordNet 138 EuroWordNet D006: Definition of the links and subsets for verbs Spanish test 7 Comment: Score yes/no a no/yes b Conditions: Example: a b Effect: Spanish test 8 Comment: Score yes a no b Conditions: Example: a b Effect: LE2-4003 Hyponymy between nouns and verbs Test sentence Si se da el caso de un X, entonces es que algo/alguien (se) Y + NP, PP (in a certain way) Si algo/alguien (se) Y, entonces se da el caso de un X - X is a noun in the singular - Y is a verb in the third person singular form - there should be at least one specifying NP or PP that makes the Y-phrase equivalent to the X-phrase or the other way around. - preferably there is no morphological link between the noun and the verb Si se da el caso de un asesinato, entonces es que alguien mata (con alevosía) ?Si alguien mata entonces se da el caso de un asesinato X = asesinato (murder) Y = matar (kill) asesinato N HAS_XPOS_HYPERONYM matar V matar V HAS_XPOS_HYPONYM asesinato N Hyponymy between state-denoting verbs and adjectives Test sentence X es ser/estar Y (phrase) ser/estar Y es X (phrase) X is a verb in the infinitive form Y is an adjective there is at least one specifying adverb, NP or PP that applies to the X-phrase or the Y-phrase preferably there is no morphological link between the noun and the verb prevaricar es ser injusto (a propósito al dictar una resolución) ?ser injusto es prevaricar (a propósito al dictar una resolución) X = prevaricar (pervert the course of justice) Y = injusto (unjust) {prevaricar} XPOS_HYPONYM_OF {injusto} {injusto} XPOS_HYPERONYM_OF {prevaricar} EuroWordNet 139 EuroWordNet D006: Definition of the links and subsets for verbs Spanish test 9 Comment: Score yes a yes b Conditions: Example: a b Effect: Spanish test 10 Comment: Score yes a b Conditions: Example: a b Effect: LE2-4003 Antonymy between verbs Test sentence Que algo/alguien (se) X es lo contrario de que ese algo/alguien (se) Y Que algo/alguien (se) Y es lo contrario de que ese algo/alguien (se) X X is a verb in the third person singular form Y is a verb in the third person singular form X and Y are co-hyponyms Que alguien se engorde es lo contrario de que ese alguien adelgace Que alguien adelgace es lo contrario de que ese alguien engorde X = engordarse (get fat) Y = adelgazar (get thin) engordarse V ANTONYM adelgazar V adelgazar V ANTONYM engordarse V Factive causation relation between verbs Test sentence Que alguien/algo X causa que alguien/algo Y Que alguien/algo X tiene como consecuencia que alguien/algo Y Que alguien/algo X lleva a que alguien/algo Y the converse of (a) X is a verb in in the third person singular form of subjunctive Y is a verb in in the third person singular form of subjunctive Que alguien/algo asesine causa que alguien/algo muera Que alguien/algo asesine tiene como consecuencia que alguien/algo muera Que alguien/algo asesine lleva a que alguien/algo muera *Que alguien/algo muera causa que alguien/algo asesine *Que alguien/algo muera tiene como consecuencia que alguien/algo asesine *Que alguien/algo muera lleva a que alguien/algo asesine {asesinar} CAUSES {morir} {morir} IS_CAUSED_BY {asesinar} reversed, non-factive EuroWordNet 140 EuroWordNet D006: Definition of the links and subsets for verbs Spanish test 11 Comment: Score yes a no b Conditions: Example: a b Effect: Spanish test 12 Comment: Score yes a no b Conditions: Example: Effect: a b Non-factive causation relation between verbs using a modal auxiliary Test sentence Que alguien/algo X puede causar que alguien/algo Y Que alguien/algo X puede tener como consecuencia que alguien/algo Y Que alguien/algo X puede llevar a que alguien/algo Y the converse of (a) X is a verb in in the third person singular form of subjunctive Y is a verb in in the third person singular form of subjunctive Que alguien/algo busque puede causar que alguien/algo encuentre Que alguien/algo busque puede tener como consecuencia que alguien/algo encuentre Que alguien/algo busque puede llevar a que alguien/algo encuentre ?Que alguien/algo encuentre puede causar que alguien/algo busque ?Que alguien/algo encuentre puede tener como consecuencia que alguien/algo busque ?Que alguien/algo encuentre puede llevar a que alguien/algo busque {buscar} CAUSES {encontrar} non-factive {encontrar} IS_CAUSED_BY {buscar} reversed, non-factive Subevent relation between verbs Test sentence Para X tienes que Y Para Y tienes que X X is a verb infinitive Y is a verb in infinitive X and Y are not connected by a hyponymy/hyperonymy relation Para comprar tienes que pagar *Para pagar tienes que comprar {comprar} HAS_SUBEVENT {pagar} {pagar} IS_SUBEVENT_OF {comprar} Verb test 13 Comment: Involvement and Role relation in general This test is replaced by accomplishing one of the specific Involvement tests below (14 to 20) LE2-4003 EuroWordNet 141 EuroWordNet D006: Definition of the links and subsets for verbs Spanish test 14 Comment: Score yes a Conditions: Example: Effect: a Spanish test 15 Comment: Score yes a Conditions: Example: Effect: a Spanish test 16 Comment: Score yes a Conditions: Example: Effect: LE2-4003 Agent Involvement/Role Test sentence Un X es un X porque es aquel/aquello que Y X is a noun Y is a verb in present third person singular form Un educador es un educador por que es aquel que educa {educador} ROLE_AGENT {educar} {educar} INVOLVED_AGENT {educador} reversed Patient Involvement/Role Test sentence Un X es un X porque es aquel/aquello que es/puede ser/ha (sido) Y-do X is a noun Y is a verb in participle form Un educando es un educando porque es aquel que es educado {educando} ROLE_PATIENT {educar} {educar} INVOLVED_PATIENT {educando} reversed Instrument Involvement/Role (also valid for nouns indicating things, used for doing something, which cannot be really called ‘instruments’) Test sentence Un X es un X porque es una cosa que sirve para Y X is a noun Y is a verb in the infinitive form Un martillo es un martillo porque sirve para martillear {martillo} ROLE_INSTRUMENT {martillear} {martillear} INVOLVED_INSTRUMENT {martillo} reversed EuroWordNet 142 EuroWordNet D006: Definition of the links and subsets for verbs 5.2 Data obtained by the analysis performed on Dutch 5.2.1 List of Base Concepts beinvloeden 00435835-v 1 2629 influence 24 7 VlisNoEnd 01436060-v aanbieden 01307317-v 2 2511 offer 7 7 VlisEnd aanbrengen 00783910-v 1 2082 puton 25 6 VlisNoEnd bedenken 00935525-v 2 3700 think up 13 2 VlisEnd 00396499-v beginnen 00207968-v 6 36461 begin 16 1 VlisNoEnd begrijpen 00330150-v 1 18184 15 understand 1 VlisNoEnd behandelen 00048767-v 5 4787 treat 17 2 VlisNoEnd bekijken 01216027-v 1 5688 look 15 8 VlisNoEnd bepalen 00393722-v 1 13245 determine 15 2 VlisNoEnd 00428728-v beschermen 00639004-v 1 3091 protect 18 1 VlisNoEnd beschouwen 00388394-v 2 7764 consider 23 1 VlisNoEnd betalen 01281885-v 2 5332 pay 4 30 VlisNoEnd bevinden 01501697-v 2 5376 be 9 6 VlisEnd bevinden 01501697-v 3 5376 be 9 14 VlisEnd bewegen 01043075-v 1 6157 move 23 14 VlisNoEnd 01046072-v LE2-4003 influence 8 think up 1 determine 5 move 15 EuroWordNet 143 EuroWordNet D006: Definition of the links and subsets for verbs bewegen 01055491-v 2 6157 move 20 16 VlisNoEnd bewegen 01046072-v 6 6157 move 15 15 VlisNoEnd binnenkomen 1 2295 01152122-v enter 17 5 VlisNoEnd blijven 00068138-v 5 52837 remain 1 1 VlisEnd brengen 00078946-v 4 36536 bring 5 1 VlisEnd 00827521-v bring 5 controleren 01226798-v 2 2087 check 21 25 VlisNoEnd 01421427-v check 28 dansen 00978339-v 2 2919 dance 17 4 VlisNoEnd delen 00897572-v 2 3233 divide 16 3 VlisNoEnd 01161526-v divide 5 denken 00341396-v 3 58538 think 8 2 VlisEnd denken 00387631-v 4 58538 think 9 6 VlisEnd doden 00758542-v 1 2496 kill 5 16 VlisNoEnd doen 00980842-v 5 71277 do 6 20 VlisEnd dragen 01537537-v 3 11963 carry 1 27 VlisEnd drijven 01935511-a 3 3360 soaked 1 1 VlisEnd drukken 01067117-v 2 6555 push 4 8 VlisEnd eindigen 01475351-v 2 2041 end 16 2 VlisEnd 00211850-v end 14 ervaren 01008772-v 2 3633 24 experience 6 LE2-4003 VlisNoEnd 01203891-v experience 7 01434123-v check 29 01204902-v experience 8 EuroWordNet 144 EuroWordNet D006: Definition of the links and subsets for verbs eten 00662381-v 3 8283 eat 2 gaan 01513147-v 1 137589 11 concern 6 VlisEnd gaan 01046072-v 3 137589 43 go 14 VlisNoEnd gaan 01100714-v 5 137589 7 head 25 VlisEnd gaan 00089026-v 8 137589 3 go 4 VlisEnd gebeuren 00204516-v 2 16006 happen 17 1 VlisEnd 00206903-v gebruiken 00658243-v 1 17227 use 8 43 VlisEnd gedragen 00007021-v 3 2091 behave 18 1 VlisNoEnd 01425745-v gelden 01513610-v 3 8411 apply 6 10 VlisEnd geven 01255335-v 3 54472 give 1 18 VlisEnd geven 01271194-v 4 54472 give 29 19 VlisNoEnd gooien 00867132-v 1 4594 throw 39 9 VlisNoEnd halen 00823804-v 1 17748 fetch 16 1 VlisNoEnd 01279432-v handelen 01341700-v 2 4414 act 12 29 VlisEnd hebben 01257491-v 1 474965 15 have 13 VlisNoEnd hebben 01487374-v 2 474965 2 have 19 VlisEnd hebben 00041140-v 3 474965 2 have 3 VlisEnd helpen 1 VlisNoEnd LE2-4003 11749 33 VlisNoEnd 00663538-v eat 3 22 00670058-v eat 4 happen 2 00207186-v happen 3 behave 2 01426576-v behave 3 EuroWordNet 145 EuroWordNet D006: Definition of the links and subsets for verbs 01442355-v help 7 herhalen 00206373-v 1 2853 repeat 20 2 VlisNoEnd hoeven 00675532-v 1 10197 require 1 3 VlisEnd houden 01517254-v 8 46508 keep 1 17 VlisEnd kapotmaken 00900879-v 1 60 destroy 16 3 VlisEnd 00928634-v kennen 00333177-v 1 21514 know 1 2 VlisEnd kiezen 00379073-v 1 7555 choose 17 1 VlisNoEnd kijken 01216027-v 1 51564 look 25 8 VlisNoEnd klinken 01241976-v 1 8091 sound 48 13 VlisNoEnd komen 01144761-v 1 135035 33 come 7 VlisNoEnd 01054590-v kopen 01259481-v 1 6942 buy 15 3 VlisNoEnd krijgen 01260836-v 1 45929 receive 21 9 VlisNoEnd 00307705-v kunnen 01539155-v 2 209812 6 can 8 VlisEnd lachen 00020446-v 1 12773 laugh 20 3 VlisNoEnd leggen 00859635-v 1 18068 lay 3 16 VlisNoEnd leiden 01518088-v 2 10955 lead 3 23 VlisEnd leiden 01381333-v 4 10955 lead 18 21 VlisNoEnd leven 01477879-v 8 26762 live2 1 VlisEnd LE2-4003 destroy 4 come 6 receive 2 EuroWordNet 146 EuroWordNet D006: Definition of the links and subsets for verbs lezen 00351672-v 1 12834 read 22 6 VlisEnd liggen 00890801-v 1 40957 lie 4 16 VlisNoEnd lijken 01217877-v 2 24119 seem 15 1 VlisNoEnd lopen 01084973-v 1 35539 walk 62 11 VlisNoEnd loslaten 00719229-v 1 1003 20 come off 2 VlisEnd maken 00926361-v 2 86305 make 49 13 VlisNoEnd noemen 00581359-v 1 18351 call 20 6 VlisEnd 00547315-v call 19 onderzoeken 00442867-v 2 4017 34 investigate 2 VlisNoEnd ontbreken 01488598-v 1 3849 lack 5 3 VlisEnd ontstaan 01484012-v 1 10944 become 16 2 VlisNoEnd openen 00772933-v 3 4849 open 37 8 VlisEnd opnemen 00111108-v 7 4801 include 15 1 VlisNoEnd opvallen 01511883-v 1 2190 excel 9 1 VlisEnd overeenkomen 1 737 19 01503041-v correspond 3 VlisEnd overtreffen 00624716-v 1 558 exceed 16 1 VlisEnd 01509366-v passen 01504707-v 1 6194 fit 10 4 VlisEnd plaatsen 00859635-v 1 3183 place 44 23 VlisNoEnd LE2-4003 exceed 2 01509494-v exceed 3 EuroWordNet 147 EuroWordNet D006: Definition of the links and subsets for verbs praten 00542186-v 4 14570 talk10 6 VlisEnd proberen 01432563-v 2 17811 try 9 15 VlisEnd raken 00089026-v 1 9419 become 9 1 VlisEnd raken 00806352-v 2 9419 hit 15 12 VlisEnd 00704074-v hit 13 regelen 00942868-v 1 2497 arrange 18 6 VlisEnd rijden 01114042-v 2 8591 ride 27 8 VlisNoEnd roepen 04737713-n 1 15111 cry 5 25 VlisNoEnd schrijven 00559904-v 3 17958 write 13 1 VlisEnd slaan 00806352-v 1 13752 hit 15 12 VlisEnd slaan 00809580-v 6 13752 hit 16 27 VlisNoEnd 00806352-v hit 15 slapen 00009337-v 1 8739 sleep 17 4 VlisNoEnd spelen 00605818-v 5 16115 play 54 21 VlisNoEnd spreken 00530290-v 2 25142 speak 50 1 VlisNoEnd springen 01118288-v 1 5030 jump 28 14 VlisNoEnd steken 00700678-v 3 10772 stab 2 2 VlisEnd sterven 00216283-v 1 6524 die 6 23 VlisNoEnd stoppen 00211850-v 8 4691 stop 6 12 VlisEnd sturen 2 22 VlisEnd LE2-4003 4284 EuroWordNet 148 EuroWordNet D006: Definition of the links and subsets for verbs 00548636-v 00826451-v send send toenemen 00093597-v 1 3168 25 increase 7 VlisEnd tonen 00373148-v 2 5565 show VlisNoEnd 00573040-v uitvoeren 00981680-v 4 3650 8 perform 2 vallen 01122509-v 1 26001 fall 15 19 VlisNoEnd 01558020-v fall 26 vallen 01122509-v 2 26001 fall 15 26 varen 01053250-v 1 2550 sail 3 24 VlisNoEnd 01108062-v sail 6 vastmaken 00768642-v 1 240 fasten 50 3 vaststellen 00287933-v 00517007-v 1 3381 24 VlisNoEnd measure 9 00392710-v determine determine 6 veranderen 00064108-v 1 10109 change 117 VlisEnd 11 veranderen 00071241-v 4 10109 change 101 VlisEnd 12 verbergen 01224184-v 1 3117 hide 9 5 VlisEnd verbergen 01224184-v 3 3117 hide 5 5 VlisEnd verbinden 00778333-v 1 2923 connect 18 4 VlisNoEnd verdwijnen 00253904-v 1 10596 36 disappear 1 VlisNoEnd verkopen 01546360-v 1 4326 sell 7 1 VlisEnd verkopen 01277199-v 2 4326 sell 4 23 VlisNoEnd 00430622-v sell 2 LE2-4003 1 5 16 6 00583238-v 01110654-v send send 2 6 show 11 00602249-v send 3 00393722-v determine 2 VlisEnd VlisEnd VlisEnd 1 EuroWordNet 149 EuroWordNet D006: Definition of the links and subsets for verbs verlangen 01028208-v 2 4070 long 15 1 VlisNoEnd 01042002-v long 2 verliezen 01301277-v 1 6852 lose 11 7 VlisEnd 01302104-v lose 10 vermoorden 00125250-n 1 2026 murder 22 1 VlisNoEnd veroorzaken 00941367-v 1 4285 cause 25 7 VlisEnd verschijnen 00251107-v 1 7396 appear 16 2 VlisNoEnd vertrouwen 00386671-v 2 3436 trust 10 8 VlisEnd 00406425-v trust 9 verwachten 00405636-v 1 8345 expect 6 4 VlisEnd verwijderen 00104355-v 2 2076 remove 77 2 VlisNoEnd verzamelen 00794237-v 2 2618 collect 29 2 VlisEnd 00796914-v collect 3 verzetten 00631049-v 2 1265 resist 17 2 VlisEnd vliegen 01104809-v 1 4232 fly 15 16 VlisNoEnd voorstellen 00469225-v 3 5177 15 VlisNoEnd represent 2 00556972-v represent 3 voorzien 00671827-v 3 2301 provide 76 2 VlisEnd 01323715-v provide 3 vormen 00949570-v 1 11573 form 11 14 VlisEnd 00083270-v form 12 vragen 00504861-v 2 34058 ask 4 6 VlisEnd vragen 00422854-v 3 34058 ask 1 12 VlisEnd vullen 00268884-v 1 3245 fill 5 18 VlisNoEnd LE2-4003 EuroWordNet 150 EuroWordNet D006: Definition of the links and subsets for verbs waarnemen 01202814-v 2 2468 10 perceive 1 VlisEnd wachten 01492762-v 1 14839 wait 4 5 VlisEnd 01494897-v weggaan 01147140-v 1 2070 leave 52 10 VlisNoEnd weigeren 00258338-v 1 3480 fail 2 6 VlisEnd weigeren 00448333-v 2 3480 refuse 2 weigeren 01274026-v 3 3480 refuse 2 4 VlisEnd werken 01364691-v 1 17080 work 29 20 VlisNoEnd 01366212-v weten 00333362-v 2 66735 know 4 3 VlisEnd willen 01040073-v 1 76697 want 10 8 VlisEnd winnen 00620486-v 1 3383 win 8 3 VlisEnd zeggen 00569629-v 1 134802 37 say 8 VlisNoEnd zetten 00862576-v 1 17652 17 put down 4 VlisNoEnd zijn 01471536-v 2 852027 5 be 3 VlisEnd zijn 01501697-v 3 852027 15 be 9 VlisNoEnd zijn 01472320-v 7 852027 4 be 4 VlisEnd zingen 00989166-v 1 4234 sing 16 3 VlisEnd 00990063-v zitten 00888240-v 1 45810 sit 1 16 VlisNoEnd zorgen 2 9 VlisEnd LE2-4003 7608 wait 6 work 21 sing 4 VlisEnd EuroWordNet 151 EuroWordNet D006: Definition of the links and subsets for verbs 01443838-v care for 5 01443257-v see to zullen 01539804-v 2 221958 2 shall 1 zwijgen 00273772-v 1 5191 4 VlisEnd fall silent 1 01249616-v fall silent aanduiden 00433815-v 1 1996 16 indicate 1 Level01 aanpassen 00181048-v 2 1060 adapt 17 2 Level01 aanraken 00685874-v 1 1175 touch 21 17 Level01 00686113-v aanrichten 00941367-v 1 395 2 give rise to 1 Level01 aansporen 00429831-v 1 247 urge 26 2 Level01 aanstellen 01356031-v 1 450 appoint 16 2 aanvallen 00633037-v 2 686 attack 1 VlisEnd 2 touch 18 Level01 01401683-v appoint 3 19 13 Level01 00633619-v attack 14 achteruitgaan 2 104 00122638-v decline 24 5 Level01 afhalen 00104355-v 3 424 take 26 4 Level01 afkraken 00462569-v 1 21 15 Level01 criticize harshly 1 afmaken 00212590-v 2 703 finish 29 6 finish 8 afnemen 01318941-v 6 1557 rob 1 15 Level01 01320281-v rob 2 afsluiten 00772512-v 2 1102 close 15 5 Level01 afvoeren 00268395-v 1 260 drain 9 6 Level01 beeindigen 00213455-v 1 802 end 15 35 Level01 LE2-4003 Level01 00285198-v EuroWordNet 152 EuroWordNet D006: Definition of the links and subsets for verbs bedekken 00763269-v 1 1275 cover 43 16 Level01 bedriegen 00479186-v 1 692 deceive 19 1 Level01 01456537-v beginnen 00207968-v 3 36461 begin 11 1 Level01 bekendmaken 1 113 18 00242637-v makepublic 1 Level01 00549256-v bekrachtigen 01403135-v 1 171 14 declarevalid 1 Level01 belemmeren 01447000-v 1 678 hinder 16 3 Level01 benadelen 01422880-v 1 175 15 disadvantage Level01 3 beoordelen 00547761-v 1 1668 judge 10 6 Level01 00377820-v beoordelen 00376571-v 2 1668 judge 21 2 Level01 berekenen 00358556-v 1 737 calculate 15 1 Level01 beschadigen 00154558-v 1 371 damage 15 6 Level01 besteden 01531792-v 1 2795 spend 14 3 Level01 bestrijken 01192082-v 2 435 spread 17 15 Level01 01519515-v betekenen 00537777-v 1 11718 mean 14 6 Level01 bevestigen 00132823-v 2 2955 confirm 4 1 Level01 bewerken 00302695-v 1 935 treat 90 3 Level01 bezighouden 01367366-v 3 1599 occupy 21 3 Level01 01366212-v LE2-4003 deceive 2 announce 2 judge 5 spread 16 exert oneself 00549373-v announce 3 2 EuroWordNet 153 EuroWordNet D006: Definition of the links and subsets for verbs bezorgen 00941367-v 2 1756 cause 11 7 Level01 brengen 00823804-v 1 36536 bring 13 2 Level01 00824200-v dalen 01122509-v 1 2332 11 go down 3 Level01 denken 00354465-v 1 58538 think 13 4 Level01 denken 00341396-v 3 58538 think 8 2 Level01 dichtmaken 00772512-v 1 38 close 35 5 Level01 doodgaan 00216283-v 1 585 die 6 3 Level01 doorgaan 00210630-v 3 1875 40 Level01 continue 2 01517254-v continue dwingen 01418102-v 2 3451 force 13 16 Level01 ergeren 01018552-v 2 1196 irritate 13 2 Level01 gebruiken 00656714-v 2 17227 11 consume 2 Level01 goedkeuren 00452960-v 2 737 agree 12 2 Level01 haasten 01177978-v 1 987 hurry 11 7 Level01 01175685-v handhaven 01515519-v 1 1943 15 maintain 9 Level01 houden 01515519-v i nformeren 00467082-v 9 46508 keep 13 16 Level01 2 1169 inform 9 2 Level01 inspannen 01343200-v 3 400 exert 9 3 Level01 01343364-v instemmen 1 13 Level01 LE2-4003 395 bring 3 7 hurry 6 overexert 1 EuroWordNet 154 EuroWordNet D006: Definition of the links and subsets for verbs 00452960-v agree 2 kaarten 00652908-v 1 165 play 25 24 Level01 kalmeren 01005235-v 1 810 calm 6 4 Level01 kapotgaan 00201526-v 1 24 break 23 20 Level01 kletsen 00587229-v 2 594 chaffer 21 1 Level01 kletsen 00586278-v 4 594 15 twaddle 2 Level01 leegmaken 00267845-v 1 43 empty 30 4 Level01 losmaken 00104355-v 1 844 remove 29 2 Level01 meedelen 00467082-v 2 596 inform 16 2 Level01 meedoen 01387083-v 1 716 14 participate 2 Level01 meemaken 01204902-v 1 2062 10 experience 8 Level01 merken 00297919-v 2 9381 mark 14 14 Level01 mislukken 01431735-v 1 1412 fail 7 13 Level01 musiceren 00986807-v 1 83 play 22 34 Level01 naderen 01174613-v 1 2204 14 approach 11 Level01 nadoen 00995557-v 1 121 copy 16 6 Level01 01512597-v omgaan 01378917-v 3 1475 manage 10 1 Level01 omgeven 01531353-v 1 848 13 surround 6 Level01 LE2-4003 copy 8 EuroWordNet 155 EuroWordNet D006: Definition of the links and subsets for verbs omhooggaan 01121367-v 1 36 rise 17 18 Level01 omvallen 01122509-v 1 253 fall 15 19 Level01 ontdoen 00117617-v 1 832 strip 59 9 Level01 00104355-v opengaan 00773597-v 1 485 open 17 9 Level01 opgeven 00611702-v 00611702-v 1 1578 quit give up 15 1 2 Level01 01269413-v ophouden 00211850-v 1 2715 stop 11 12 Level01 opschrijven 00576052-v 1 730 16 writedown 1 Level01 optillen 00095155-v 1 399 lift 9 17 Level01 01121367-v lift 25 ordenen 00416049-v 1 637 arrange 15 2 Level01 protesteren 01427604-v 1 988 protest 14 5 Level01 00512172-v protest 3 samenkomen 1 163 01374778-v meet 16 10 Level01 01408129-v meet 11 schoonmaken 1 676 00023287-v clean 00881979-v clean 52 2 7 Level01 00106393-v clean 4 schoppen 00788650-v 2 983 kick 24 6 Level01 spelen 00987477-v 8 16115 play 7 35 Level01 stoppen 00218979-v 5 4691 stop 13 13 Level01 01061046-v stop 16 streven 01433327-v 3 2147 strive 13 2 Level01 strijden 1 5 Level01 LE2-4003 845 remove 2 00721752-v strip 14 give up 5 01515268-v give up 12 00109110-v clean 5 EuroWordNet 156 EuroWordNet D006: Definition of the links and subsets for verbs 00615347-v fight 5 toevoegen 00110396-v 1 1737 add 1 16 Level01 00580469-v add 4 uitdoen 00868381-v 2 170 turn off 6 1 uiteenvallen 00237247-v 1 226 10 break intoparts Level01 1 uiten 00554586-v 1 1412 utter 20 3 Level01 00529407-v uiten 00531321-v 2 1412 express 20 6 Level01 uitleggen 00528672-v 1 1807 explain 11 2 Level01 uitoefenen 00662183-v 2 2237 exert 3 1 Level01 01257388-v uitstellen 01495560-v 1 538 14 postpone 1 Level01 uitzoeken 00357043-v 3 577 solve 13 1 Level01 vastbinden 00768642-v 1 413 fix 11 14 Level01 vastleggen 00563886-v 3 1392 15 putdown 1 Level01 vastzitten 01045270-v 1 307 be fixed 1 10 Level01 00131922-a stuck vechten 00615347-v 1 2651 fight 14 5 Level01 verbeteren 00123459-v 1 1762 improve 1 8 Level01 verbeteren 00123997-v 2 1762 improve 2 18 Level01 verdwijnen 00253904-v 1 10596 disappear 36 Level01 vergissen 00347169-v 1 1304 err 1 15 Level01 LE2-4003 Level01 utter 1 exert 2 1 EuroWordNet 157 EuroWordNet D006: Definition of the links and subsets for verbs verkleinen 00262983-v 1 337 8 makesmaller 2 Level01 verknoeien 01431414-v 1 288 botch 19 2 Level01 11 Level01 verminderen 00090574-v 1 1804 22 decrease 5 Level01 verminderen 00262983-v 2 1804 23 decrease 6 Level01 vernielen 00900879-v 1 507 destroy 20 3 Level01 vernietigen 00900879-v 1 1876 destroy 17 3 Level01 veroorzaken 00432532-v 1 4285 cause 25 6 Level01 00941367-v verplaatsen 01055491-v 1 1166 displace 3 43 Level01 verschaffen 01323715-v 1 1982 supply 18 6 Level01 versieren 00959417-v 1 919 32 decorate 2 Level01 verspreiden 00792958-v 1 1781 spread Level01 00546238-v vervoeren 00834152-v 1 702 11 Level01 transport 7 01110654-v transport 9 verwarmen 00222990-v 1 702 heat 10 7 Level01 00223859-v warm 1 verwerven 01301819-v 1 1960 gain 18 9 Level01 verwonden 00043545-v 1 187 injure 20 1 Level01 verzorgen 00048767-v 1 1451 care for 17 1 Level01 00026120-v neaten 1 vermeerderen 2 464 00093597-v increase 7 LE2-4003 15 11 cause 7 distribute 1 01112683-v transport 10 EuroWordNet 158 EuroWordNet D006: Definition of the links and subsets for verbs voorbereiden 1 1808 00241323-v prepare 7 2 Level01 voorbijgaan 01172741-v 22 3 Level01 1 1015 go by voortbewegen 2 260 50 Level01 01136714-v moveon 2 01046072-v locomote vrijen 00820052-v 2 882 37 make love 1 Level01 weergeven 00966090-v 1 1051 12 represent 5 Level01 werven 01275844-v 1 167 recruit 14 4 Level01 zien 01215314-v 4 98375 see 14 12 Level01 zuiveren 00881979-v 1 346920 13 makeclean 1 Level01 LE2-4003 1 EuroWordNet 159 EuroWordNet D006: Definition of the links and subsets for verbs 5.3 Data obtained by the analysis performed on Italian 5.3.1 List of Base Concepts In the following, the results obtained so far are reported (cf. above). In particular we indicate: 1) the verb homograph; 2) the preliminarily established correspondences among the genus senses and WN 1.5 synsets; 3) the levels in WN taxonomies; 4) the levels in our (preliminary) taxonomies; 5) the frequency of each non-disambiguated genus (i.e., of its senses altogether). Note that: i) some Italian verb senses which, for the time being, have not been connected, match with the same WN synset, thus, either a) we shall insert them in a same synset in a further stage of our work (i.e., when the sense disambiguation will be complete); or b) we shall state that in Italian more granular distinctions are operated with respect to those found in WN; ii) when either two WN synsets match with one Italian verb sense or vice versa, we indicate it by means of the conjunction ‘and’ connecting the two senses corresponding to one sense in the other language (i.e., WN1 and WN2 = Italian verb, and vice versa). Top-concept WN corresponding synsets Level in WN Level in Italian Frequency as a genus fare Sense 1 of to do 01448761 Sense 3 of to do 01449587 Sense 3 of to make 00926361 Sense 1 of to put 00859635 ??? Sense 3 of to give 01254390 Sense 1 of to give 01317872 Sense 1 of to provide 01323715 Sense 1 of to become 00089026 Sense 1 of to be 01472320 Sense 1 of to take 00691086 Sense 1 of to become 00089026 Sense 29 of to take 01471041 Sense 10 of to take 01354977 Sense 1 of to have 01256853 Sense 1 of to feel 01008772 and Sense 3 of to feel 01202814 Sense 1 of remove 00104355 Sense 1 of to lose 01301401 ? Sense 1 of to go 01046072 Sense 1 of to go 01046072 and sense 2 of to go 201054314 ? Sense 1 of to change 00072540 Sense 3 of to pass 01172741 Sense 1 of to send 01110956 Sense 1 of to subject 00632326 top top-5 top top-1 top-1 top top-1 top top top top top-1 top-1 top top top top top top top top top top-1 top-1 top-1 top top-1 top top top top-1 top-1 top top top top top top-1 top top top 374 top top-1 top-1 top-1 135 110 107 top top-1 top top-1 92 80 76 74 mettere dare diventare essere prendere avere togliere perdere andare ridurre passare mandare sottoporre LE2-4003 288 286 221 206 164 163 EuroWordNet 160 EuroWordNet D006: Definition of the links and subsets for verbs privare colpire liberare portare cadere dire mettersi parlare emettere provare rendere tenere unire porre produrre disporre divenire provocare chiudere compiere eseguire tagliare coprire lavorare dividere lasciare cercare muovere stare apparire trasformare stabilire battere muoversi esporre LE2-4003 Sense 1 of to deprive 01316712 Sense 2 of to hit 00704074 Sense 1 of to strike 01007544 Sense 1 of to free 01370323 Sense 1 of to bring 01188762 Sense 2 of to bring 00827521 Sense 1 of fall down 01044689 Sense 2 of to fall 01122509 Sense 4 of to fall 01558020 Sense 1 of to say 00569629 and sense 8 of to say 00532392 ??? Sense 1 of to speak 00530290 Sense 1 of to emit 00062533 Sense 2 of to emit 01563115 Sense 3 of to emit 00554586 Sense 1 of to feel 01008772 and sense 3 of to feel 01202814 Sense 3 of to make 00926361 Sense 1 of to keep 01515519 Sense 2 of to keep 01517254 Sense 3 of to unite 01483441 Sense 1 of to put 00859635 Sense 6 of to produce 01222100 Sense 1 of to put 00859635 Sense 1 of to to become 00089026 Sense 1 of to provoke 01003070 Sense 1 of to close 00772512 Sense 1 of to accomplish 00938602 Sense 1 of to accomplish 00938602 Sense 1 of to cut 00894185 Sense 1 of to cover 00763269 Sense 3 of to cover 00687390 Sense 1 of to work 01366212 Sense 1 of to divide 01396914 Sense 4 of to divide 00898017 Sense 1 of to distinguish 00365740 Sense 5 of to let 00291924 Sense 1 of to try 01432563 Sense 1 of to look for 00754445 Sense 1 of to move 01046072 Sense 3 of to move 01043075 Sense 1 of to stay 00068138 Sense 2 of to stay 01060041 Sense 1 of to appear 01217877 Sense 2 of to transform 00229041 Sense 3 of to establish 00373148 Sense 2 of to hit 00704074 Sense 1 of to move 01046072 Sense 1 of to show 01225799 top-1 top-1 top top top-1 top-1 top-1 top top-1 top-1 top-1 top-1 top-1 top top top top top top top-1 top top-1 top-1 top-1 top top-1 top top-1 top-1 top-1 top top-1 top top-1 top-1 top-1 top-1 top-1 top top top top-1 top-1 top-1 top-1 top-1 top-1 top top-1 top-1 top top-1 top-1 top-1 top-1 top-2 70 68 top 53 top top-1 top-1 52 50 48 top 48 top top-1 48 45 top-1 top top-1 top top top-1 top top top top-2 top-2 44 43 42 41 40 40 39 39 37 37 36 top top top top top-1 top-1 36 35 top top-1 top 34 top top-1 top-1 top top top-1 32 31 30 29 29 28 64 60 54 35 34 33 EuroWordNet 161 EuroWordNet D006: Definition of the links and subsets for verbs farsi mostrare cambiare esprimere raccogliere riconoscere separare sostenere stringere subire rompere volgere cessare diminuire esercitare indurre manifestare scrivere tirare condurre dichiarare eccitare aumentare impedire riempire causare LE2-4003 Sense 1 of to become 00089026 Sense 2 of to cause 00432532 Sense 1 of to behave 00007021 Sense 4 of to show 01219939 Sense 5 of to show 01226169 Sense 1 of to change 00072540 Sense 2 of to change 00064108 Sense 1 of to express 00531321 Sense 1 of to collect 01311458 Sense 3 of to recognize 01251230 Sense 1 of to admit 00459649 Sense 1 of to declare 00570287 Sense 4 of to divide 00898017 Sense 1 of to distinguish 00365740 Sense 1 of to support 01446559 Sense 4 of to support 00693130 Sense 1 of to tighten 00249759 Sense 7 of to press 00799016 Sense 1 of to undergo 01203891 Sense 3 of to break 00787971 and sense 5 of to break 00201902 Sense 1 of to violate 01509094 and sense 2 of to violate 01451879 Sense 1 of to turn 01086483 Sense 1 of to stop 01515268 Sense 1 of to lessen 00090574 top-2 top-1 top top top top top top-4 top-3 top top-2 top-2 top-1 top-1 top top top-1 top-2 top top-1 top-1 top top-4 top-1 top top-2 Sense 1 of to use 00658243 Sense 1 of to do 01448761 Sense 2 of to induce 00432532 Sense 4 of to show 01219939 and sense 5 of to show 01226169 Sense 1 of to write 00972859 Sense 1 of to throw 00867132 Sense 1 of to lead 01141779 Sense 5 of to lead 01381333 Senses 1 of to declare 00570287 and sense 2 of to declare 00544887 Sense 6 of to excite 01004175 Sense 1 of to increase 00093597 Sense 2 of to increase 00091455 Senses 1 of to prevent 01388675 and sense 2 of to prevent 01387332 Sense 1 of to fill 00268884 Sense 1 of to cause 00941367 and sense 2 of to cause 00432532 top top top top top top-2 top-2 top top-3 top-2 top-2 top-2 top-2 top-1 top top-1 top-1 top top top top top-1 28 top 27 top-1 top-1 top-1 top top-1 top top top-1 26 26 26 top-2 top-1 top top-2 25 28 26 25 25 24 top-1 top-1 top top-1 and top-1 top top top-1 top-1 24 23 23 top-2 top-1 top-2 top top-1 23 23 22 top-1 top-1 top-1 top-1 22 21 top-1 top 21 20 23 23 23 22 21 EuroWordNet 162 EuroWordNet D006: Definition of the links and subsets for verbs 5.3.2 Motion verb synsets and taxonomy In the following, the synsets and taxonomy built for motion verbs are reported. For some verbs (the most interesting from the point of view of the issues raised) we indicate: a) the LDB sense considered; b) the corresponding synset built on the basis both of data from dictionaries and of personal intuitions/decisions; the corresponding synset(s) in WordNet 1.5 (indicated as WN). muoversi Sense 1 muoversi -- (to move - no specification either of direction or manner) [WN: Sense 3 of to move] => agitarsi 0_1 agitarsi, tempestare (to be agitated) [WN: Sense 1 of to be agitated] => dibattersi (to struggle, to flounder) => sciaguattare (to move in a container - said of liquids) => sguazzare (to wallow) => spagliare (to scatter straw) => andare 0_1a andare, muovere (to go along an unbounded path) [WN: ? Sense 1 of to go] => accompagnare 0_1a accompagnare, guidare, portare, andare insieme, scortare (to accompany, to go with) [WN: Sense 2 of to accompany] => aggirarsi (to wander about, to go about) => arrancare (to plod along, to trudge) => avanzare 0_1b avanzare, andare avanti, procedere, proseguire (to move forward, to advance into, to progress) [WN: Sense 1 of to advance] => bighellonare (to lounge, to loaf) => camminare 0_1 camminare, andare a piedi (to walk, to go on foot) [WN: Sense 1 of to walk] => cavalcare 0_0 cavalcare, andare a cavallo (to ride) [WN: Sense 1 of to ride] => ciondolare (to dangle) => circolare (to circulate, to move round) => concorrere (to come together, to assemble) => correre 0_1 (to run along an unbounded path) [WN: ?Sense 1 of to run] => derivare (to drift) => errare 0_1 vagare, vagabondare, peregrinare, girovagare (to wander about, around) [WN: Sense 1 of to wander] LE2-4003 EuroWordNet 163 EuroWordNet D006: Definition of the links and subsets for verbs => foraggiare (to go around the country to look for foraging) => galoppare (to gallop) => girandolare 0_0 girandolare, girellare, gironzolare, girovagare (to saunter, to stroll) [WN: Sense 1 of to saunter] => girare (0_2a) girare, percorrere, visitare (to travel, to tour) [WN: ?? Sense 2 and Sense 3 of to travel] => girellare 0_0 girellare, girandolare, gironzolare, girovagare (to saunter, to stroll) [WN: Sense 1 of to saunter] => gironzolare 0_0 gironzolare, girandolare, girellare, girovagare (to saunter, to stroll) [WN: Sense 1 of to saunter] => girovagare 0_1 girovagare, girandolare, girellare, gironzolare (to saunter, to stroll) [WN: Sense 1 of to saunter] => navigare 0_0 navigare, viaggiare (to sail, to navigate, to go, to travel - said of ships or airplanes) [WN: Sense 1 and sense 3 of to navigate] => orzare (to luff) => pattugliare (to patrol) => peregrinare 0_1 peregrinare, errare, vagabondare, vagare, girovagare 0_2 (to wander, to roam) [WN: Sense 1 of to wander] => precedere (to go ahead) => procedere 0_1 avanzare, andare avanti, procedere, proseguire (to move forward, to advance into, to progress) [WN: Sense 1 of to advance] => proseguire 0_0 avanzare, andare avanti, procedere, proseguire (to move forward, to advance into, to progress) [WN: Sense 1 of to advance] => scarrozzare (to drive about (in a carriage)) => seguire (to follow) => sgonnellare (to wiggle) => slittare (to slide) => trottare (to trot) => trotterellare (to trot along, to toddle) => uccellare (to go fowling) => vagabondare 0_1b vagabondare, girovagare, peregrinare, vagare, errare (to wander (a place), to rove (a place)) [WN: Sense 1 of to wander] => vagare 0_0 LE2-4003 EuroWordNet 164 EuroWordNet D006: Definition of the links and subsets for verbs vagare, vagabondare, girovagare, peregrinare, errare (to wander (a place), to rove (a place)) [WN: Sense 1 of to wander] => andare 0_1b andare, recarsi, portarsi, trasferirsi (to go to a place, to change position) [WN: ? Senses 1 and 2 of to go] => accompagnare 0_1a accompagnare, guidare, portare, andare insieme, scortare (to accompany, to go with) [WN: Sense 2 of to accompany] => affondare (to sink) => affrontare (to go against someone) => allontanarsi 0_0a allontanarsi, discostarsi, muoversi (0_2), scostarsi (to move/go away) [WN: Senses 1 and 2 of to go away] => allontanarsi 0_0b allontanarsi, andare lontano (to go far) [WN: ???] => cadere 0_1 cadere, cascare (to fall down) [WN: Sense 1 of to fall down] => cascare 0_0 cascare, cadere (to fall down) [WN: Sense 1 of to fall down] => congedarsi (to take one’s leave) => coricarsi (to go to bed) => correre 0_3 correre, accorrere, affrettarsi (to run along bounded path, to run to a place) [WN: ? Sense 1 of to run] => discendere (to go down) => dividersi (to go towards different directions) => entrare 0_1b entrare, andare dentro (to go into) [WN: Sense 1 of to go into] => entrare 0_1a entrare, addentrarsi, inoltrarsi, penetrare (to go into, to penetrate, to get into) [WN: Sense 2 of to go into] => esiliarsi 0_0 esiliarsi, esulare (to go into exile) => esulare 0_0 esulare, esiliarsi (to go into exile) => naufragare (to be wreck) => riandare (to go to a place again) => ribaltarsi (to upset) => rilevare (to call for) => ritornare (to go back again) => salire (to go up) => scendere (to go down) => sloggiare (to clear out) LE2-4003 EuroWordNet 165 EuroWordNet D006: Definition of the links and subsets for verbs => spargersi (to scatter, to spread) => spatriare (to emigrate) => tornare (to go/come back) => uscire (to go out) => brulicare (to swarm) => dimenarsi (to move about restlessly) => dondolare (to swing) => formicolare (to swarm) => girare (0_1) (to turn) [WN: Sense 1 of to turn] => gravitare (to gravitate) => guizzare (to wriggle, to frisk) => marciare (to march) => navigare (to sail, to navigate) => nuotare (to swim) => ondeggiare (to rock, to wave, to sway, to flutter, to ripple, to flicker, to blow ...) [WN: Sense 2 of to rock; Sense 2 of to flutter; Sense 1 of to ripple ...] => ondeggiare, barcollare (to sway, to vacillate) [WN: Sense 2 of to sway; Sense 2 of to vacillate, which is a hyponym of Sense 2 of to sway ???] => oscillare, tentennare, dondolare, ciondolare, vacillare (to swing, to sway, to flicker, to waver, to oscillate, to rock, to vacillate ...) [WN: Sense 2 of to swing and of to sway (same synset); the same as for ondeggiare above] => passare, transitare (to pass (through a place), to go along) [WN: Sense 3 of to pass] => scatenarsi (?? to move with a great energy, or with violence) => scorrere (to run, to glide, to slide, to flow) [WN: Sense 7 of to run; Sense 1 of to slide] => scrollarsi (to shake) => spaziare (to space, to range) => sventolare (to wave, to flutter) => tramestare (to move in a noisy way) => vacillare, pencolare, traballare, tentennare (to vacillate) [WN: cf. above] => vibrare (to vibrate, to quiver) => volare (to fly) [WN: Sense 1] 5.3.3 Know verb synsets and taxonomies sapere Sense 1 sapere, conoscere (to know, to be aware of something) [WN: Sense 1 of to know] (essere a conoscenza di qlc., avere notizia di qlc.) Sense 2 LE2-4003 EuroWordNet 166 EuroWordNet D006: Definition of the links and subsets for verbs sapere, conoscere, avere familiarità con, intendere, capire, comprendere (to know, to have a familiarity with, to understand) [WN: Sense 2 of to know] => padroneggiare (to master) => digerire, assimilare (to understand well) => penetrare (to penetrate) => decodificare Sense 3 sapere, conoscere, discernere, distinguere, riconoscere, ravvisare (to be aware of, to discern, to distinguish) [WN: sense 2 of to distinguish] Sense 4 sapere, conoscere, apprendere, divenire consapevole, capire (to know, to become aware of, to realize) [WN: Sense 2 of to realize] => imparare (to learn) => impadronirsi (to begin to master) => memorizzare (to memorize) => afferrare, intuire (to realize, to understand immediately without reasoning) Sense 5 sapere, essere in grado, essere capace Sense 6 sapere, presagire conoscere Sense 1 sapere, conoscere (to know, to be aware of something) [WN: Sense 1 of to know] Sense 2 sapere, conoscere, avere familiarità con, intendere, capire, comprendere (to know, to have a familiarity with, to understand) [WN: Sense 2 of to know] => padroneggiare (to master) => digerire, assimilare (to understand well) => penetrare (to penetrate) => decodificare (to decode) Sense 3 LE2-4003 EuroWordNet 167 EuroWordNet D006: Definition of the links and subsets for verbs sapere, conoscere, discernere, distinguere, riconoscere, ravvisare (to be aware of, to discern, to distinguish) [WN: sense 2 of to distinguish] Sense 4 sapere, conoscere, apprendere, divenire consapevole, capire (to know, to become aware of, to realize) [WN: Sense 2 of to realize] => imparare (to learn) => impadronirsi (to begin to master) => memorizzare (to memorize) => afferrare, intuire (to realize, to understand immediately without reasoning) Sense 5 conoscere, avere familiarità con, sapere chi è qualcuno (to know someone, to have a familiarity with someone) [WN: Sense 4 of to know] 5.4 Data obtained by the analysis performed on Spanish 5.4.1 List of Base Concepts 104 synsets Structure of the file: WNsynsetId Lexicographersfile 1stEnglishwordofWNsynset SpanishSynset freqasGenus freqinMRD freqinCorpus 00021823 29 look {parecer} 14 635 2267 00033668 29 wear {llevar,usar} 273 996 2780 00037593 29 bear {llevar} 192 932 2400 00048767 29 treat {tratar} 122 579 1759 00068499 30 keep_up {seguir} 62 311 1599 00079704 30 leave {dejar} 402 655 2139 00104355 30 remove {quitar,sacar} 873 1167 906 00113224 30 insert {poner,colocar} 1552 2597 3008 00204516 30 happen {pasar} 308 724 2573 00210630 30 go_on {seguir} 62 311 1599 00217088 30 leave {dejar} 402 655 2193 00311709 30 issue {salir} 220 552 1357 LE2-4003 EuroWordNet 168 EuroWordNet D006: Definition of the links and subsets for verbs 00331061 31 understand {ver} 80 230 5185 00333362 31 know {saber,conocer} 47 301 4296 00333754 31 know {saber} 26 119 2868 00334020 31 know {saber} 26 119 2868 00341396 31 remember {recordar} 14 60 1005 00341811 31 think {pensar} 8 56 1598 00342479 31 remember {pensar} 15 56 1598 00343126 31 remind {recordar} 14 60 1005 00343621 31 remember {recordar} 14 60 1005 00343810 31 commemorate {recordar} 14 60 1005 00345074 31 abandon {dejar} 402 655 2193 00354465 31 think {pensar} 8 56 1598 00359802 31 get {obtener,sacar} 407 1098 1129 00366972 31 identify {llamar} 68 383 1659 00405636 31 expect {esperar} 16 76 1153 00416049 31 arrange {poner} 1478 2049 2671 00432532 32 induce {hacer,causar} 3054 6095 9045 00451248 32 permit {dejar} 402 655 2193 00535241 32 consider {contar} 38 139 1069 00551959 32 pronounce {decir} 286 719 7424 00557762 32 address {hablar} 204 760 2097 00611702 33 drop_out {dejar} 402 655 2193 00620218 33 lose {perder} 304 417 1020 00656714 34 consume {tomar} 426 699 1210 00658243 34 use {usar,emplear} 94 2827 683 00661760 34 use {usar,emplear} 94 2827 683 00675686 34 need {querer} 18 119 2499 00691086 35 take {tomar,obtener,sacar} 620 1797 2338 00763269 35 cover {cubrir} 334 937 330 00848136 35 free {quitar} 1002 640 328 00877616 35 function {ir} 148 779 11701 00926188 36 create {crear,causar} 222 685 1812 00926361 36 make {crear,realizar,causar} 264 977 2848 01008772 37 feel {sentir} 130 227 1049 01011032 37 love {querer} 18 119 2499 01040073 37 desire {querer} 18 119 2499 01046072 38 travel {ir} 148 779 11701 01054314 38 go {ir,salir} 294 1331 13058 01141779 38 lead {llevar} 192 932 2400 01144761 38 arrive {llegar} 196 343 2720 01146905 38 stay {quedar} 96 522 1957 01152122 38 enter {entrar} 96 1775 963 01154482 38 reach {llegar} 154 343 2720 LE2-4003 EuroWordNet 169 EuroWordNet D006: Definition of the links and subsets for verbs 01156236 38 meet {encontrar} 14 277 2026 01202814 39 feel {sentir} 130 227 1049 01203891 39 experience {tener} 546 3946 8289 01214474 39 miss {perder} 304 417 1020 01214832 39 watch {mirar} 50 102 1447 01216027 39 look {mirar} 50 102 1447 01226339 39 look_at {mirar} 50 102 1447 01227418 39 watch {mirar} 50 102 1447 01230190 39 produce {producir} 238 1227 1386 01255335 40 give {dar} 1942 2509 5054 01256853 40 have {tener} 546 3946 8289 01257491 04 own {tener} 546 3946 8289 01261345 40 get {tomar,obtener,sacar} 620 1797 2338 01267839 40 get_rid_of {quitar} 1002 640 328 01280035 40 find {encontrar} 14 277 2026 01283168 40 owe {deber} 2 418 2057 01286642 40 trade {tratar} 122 579 1759 01289475 40 account {contar} 38 139 1069 01301401 40 lose {perder} 304 417 1020 01301721 40 lose {perder} 304 417 1020 01302104 40 lose {perder} 304 417 1020 01317872 40 give {dar} 1942 2509 5054 01350197 41 install {poner} 1478 2049 2671 01360902 41 remove {quitar,sacar} 873 1167 906 01361498 41 take_out {quitar,sacar} 873 1167 906 01371393 41 let {dejar} 402 655 2193 01448761 41 effect {hacer} 2854 5522 8818 01469362 41 meet {conocer} 42 182 1428 01471536 42 exist {haber,existir} 54 1711 21521 01472320 42 be {ser} 260 3452 48737 01475351 42 end {acabar} 38 104 1009 01476675 42 be {haber} 40 1588 20355 01477977 42 live {vivir} 62 543 1434 01480699 42 survive {vivir} 62 543 1434 01482115 42 constitute {ser} 260 3452 48737 01492762 42 wait {esperar} 16 76 1153 01501697 42 be {estar} 342 900 5542 01506899 42 equal {ser} 260 3452 48737 01513147 42 refer {tratar} 122 579 1759 01515268 42 discontinue {dejar} 402 655 2193 01531792 42 spend {pasar} 308 724 2573 01535432 42 postdate {seguir} 62 311 1599 01539155 42 can {poder,saber} 33 2184 10619 LE2-4003 EuroWordNet 170 EuroWordNet D006: Definition of the links and subsets for verbs 01539245 42 can {poder} 20 2065 7751 01539308 42 must {deber} 2 418 2057 01539689 42 must {deber} 2 418 2057 01539895 42 should {deber} 2 418 2057 01546360 42 sell {vender} 68 695 278 01548592 42 persist {quedar} 96 522 1957 01556433 42 have {tratar} 122 579 1759 5.4.2 Motion verb synsets and taxonomies VERBS OF MOVEMENT EXTRACTED SO FAR: taxonomies of 'ir', 'mover' and 'moverse' 4.4.1 IR TAX ir_1_1 ir_1_1 acudir_1_1 ir_1_1 acudir_1_1 acorer_1_2 ir_1_1 acudir_1_1 afluir_1_1 ir_1_1 acudir_1_2 ir_1_1 andar_2_1 ir_1_1 andar_2_1 acular_1_4 ir_1_1 andar_2_1 agilar_1_1 ir_1_1 andar_2_1 amblar_1_1 ir_1_1 andar_2_1 amblar_1_1 anquear_1_1 ir_1_1 andar_2_1 ambular_1_1 ir_1_1 andar_2_1 anadear_1_1 ir_1_1 andar_2_1 anadear_1_1 nanear_1_1 ir_1_1 andar_2_1 apeonar_1_1 ir_1_1 andar_2_1 atrochar_1_1 ir_1_1 andar_2_1 bandear_2_2 ir_1_1 andar_2_1 bodegonear_1_1 ir_1_1 andar_2_1 cabalgar_1_1 ir_1_1 andar_2_1 cabalgar_1_1 subir_1_2 ir_1_1 andar_2_1 cabalgar_1_1 trotar_1_2 ir_1_1 andar_2_1 cabalear_1_1 ir_1_1 andar_2_1 calejear_1_1 ir_1_1 andar_2_1 calejear_1_1 rutiar_1_1 ir_1_1 andar_2_1 caminar_1_2 ir_1_1 andar_2_1 camotear_1_1 ir_1_1 andar_2_1 cantonear_1_1 ir_1_1 andar_2_1 cazcalear_1_1 LE2-4003 EuroWordNet 171 EuroWordNet D006: Definition of the links and subsets for verbs ir_1_1 andar_2_1 cazcalear_1_1 andorear_1_1 ir_1_1 andar_2_1 cazcalear_1_1 andorear_1_1 andareguear_1_1 ir_1_1 andar_2_1 cejar_1_1 ir_1_1 andar_2_1 cejar_1_1 acular_1_5 ir_1_1 andar_2_1 cejar_1_1 recular_1_1 ir_1_1 andar_2_1 cejar_1_1 recular_1_1 recejar_1_1 ir_1_1 andar_2_1 cejar_1_1 recular_1_1 retrechar_1_1 ir_1_1 andar_2_1 circular_2_1 ir_1_1 andar_2_1 circular_2_1 corer_1_9 ir_1_1 andar_2_1 cojear_1_1 ir_1_1 andar_2_1 cojear_1_1 chenquear_1_1 ir_1_1 andar_2_1 coretear_1_1 ir_1_1 andar_2_1 coretear_1_1 hopear_1_2 ir_1_1 andar_2_1 culebrear_1_1 ir_1_1 andar_2_1 danzar_1_3 ir_1_1 andar_2_1 deambular_1_1 ir_1_1 andar_2_1 deambular_1_1 divagar_1_1 ir_1_1 andar_2_1 deambular_1_1 divagar_1_1 erar_1_4 ir_1_1 andar_2_1 deambular_1_1 divagar_1_1 trafagar_1_3 ir_1_1 andar_2_1 deambular_1_1 rular_1_3 ir_1_1 andar_2_1 discurir_1_1 ir_1_1 andar_2_1 discurir_1_1 pasear_1_6 ir_1_1 andar_2_1 discurir_1_1 travesear_1_2 ir_1_1 andar_2_1 erar_1_3 ir_1_1 andar_2_1 escarabajear_1_1 ir_1_1 andar_2_1 ganadear_1_1 ir_1_1 andar_2_1 gatear_1_2 ir_1_1 andar_2_1 gatear_1_5 ir_1_1 andar_2_1 jinetear_1_1 ir_1_1 andar_2_1 ladear_1_3 ir_1_1 andar_2_1 lanear_1_1 ir_1_1 andar_2_1 marchar_1_1 ir_1_1 andar_2_1 marchar_1_1 andonear_1_1 ir_1_1 andar_2_1 marchar_1_1 desfilar_1_1 ir_1_1 andar_2_1 marchar_1_1 desfilar_1_3 ir_1_1 andar_2_1 marchar_1_1 rodar_1_9 ir_1_1 andar_2_1 montear_1_5 ir_1_1 andar_2_1 nagualear_1_1 ir_1_1 andar_2_1 navegar_1_2 ir_1_1 andar_2_1 navegar_1_3 ir_1_1 andar_2_1 navegar_1_4 ir_1_1 andar_2_1 pasear_1_1 ir_1_1 andar_2_1 pasear_1_1 carocear_1_1 LE2-4003 EuroWordNet 172 EuroWordNet D006: Definition of the links and subsets for verbs ir_1_1 andar_2_1 pasear_1_1 rondar_1_4 ir_1_1 andar_2_1 pasear_1_1 ruar_1_2 ir_1_1 andar_2_1 pasear_1_3 ir_1_1 andar_2_1 pasear_1_7 ir_1_1 andar_2_1 patear_1_4 ir_1_1 andar_2_1 patojear_1_1 ir_1_1 andar_2_1 peregrinar_1_1 ir_1_1 andar_2_1 peregrinar_1_4 ir_1_1 andar_2_1 pernear_1_2 ir_1_1 andar_2_1 recorer_1_1 ir_1_1 andar_2_1 recorer_1_1 bajar_1_5 ir_1_1 andar_2_1 recorer_1_1 caminar_1_4 ir_1_1 andar_2_1 recorer_1_1 campear_1_6 ir_1_1 andar_2_1 recorer_1_1 corer_1_13 ir_1_1 andar_2_1 recorer_1_1 corer_1_24 ir_1_1 andar_2_1 recorer_1_1 pampear_1_1 ir_1_1 andar_2_1 recorer_1_1 patear_1_5 ir_1_1 andar_2_1 recorer_1_1 peregrinar_1_5 ir_1_1 andar_2_1 recorer_1_1 rodar_1_6 ir_1_1 andar_2_1 recorer_1_1 subir_1_10 ir_1_1 andar_2_1 recorer_1_1 viajar_1_5 ir_1_1 andar_2_1 reptar_1_1 ir_1_1 andar_2_1 rodear_1_1 ir_1_1 andar_2_1 rodear_1_1 arodear_1_1 ir_1_1 andar_2_1 rondar_1_2 ir_1_1 andar_2_1 rondar_1_3 ir_1_1 andar_2_1 rondar_1_6 ir_1_1 andar_2_1 ruar_1_1 ir_1_1 andar_2_1 rumbear_1_2 ir_1_1 andar_2_1 talonear_1_1 ir_1_1 andar_2_1 terear_1_2 ir_1_1 andar_2_1 tesar_1_2 ir_1_1 andar_2_1 trafagar_1_1 ir_1_1 andar_2_1 trafagar_1_4 ir_1_1 andar_2_1 trajinar_1_2 ir_1_1 andar_2_1 travesear_1_1 ir_1_1 andar_2_1 trotar_1_3 ir_1_1 andar_2_1 tumbear_1_1 ir_1_1 andar_2_1 tunar_1_1 ir_1_1 andar_2_1 tunar_1_1 lampar_1_2 ir_1_1 andar_2_1 vagabundear_1_1 ir_1_1 andar_2_1 vagabundear_1_1 bundear_1_2 ir_1_1 andar_2_1 vagabundear_1_1 vagamundear_1_1 LE2-4003 EuroWordNet 173 EuroWordNet D006: Definition of the links and subsets for verbs ir_1_1 andar_2_1 vagar_3_1 ir_1_1 andar_2_1 vagar_3_2 ir_1_1 andar_2_1 vagar_3_3 ir_1_1 andar_2_1 ventear_1_12 ir_1_1 andar_2_1 ventear_1_4 ir_1_1 andar_2_1 voltear_1_8 ir_1_1 andar_2_1 zancajear_1_1 ir_1_1 andar_2_1 zancajear_1_1 zanquear_1_2 ir_1_1 andar_2_1 zanganear_1_1 ir_1_1 andar_2_1 zarcear_1_3 ir_1_1 andar_2_1 zigzaguear_1_1 ir_1_1 aribar_1_3 ir_1_1 aribar_1_5 ir_1_1 atrechar_1_1 ir_1_1 barquear_1_1 ir_1_1 bordear_1_1 ir_1_1 bordear_1_1 bordejar_1_1 ir_1_1 bordonear_1_1 ir_1_1 cabalgar_1_3 ir_1_1 caminar_1_1 ir_1_1 caminar_1_1 arear_1_5 ir_1_1 caminar_1_1 chuequear_1_1 ir_1_1 caminar_1_1 corer_1_1 ir_1_1 caminar_1_1 corer_1_1 acosar_1_3 ir_1_1 caminar_1_1 corer_1_1 adelantar_1_1 ir_1_1 caminar_1_1 corer_1_1 adelantar_1_1 ascender_1_2 ir_1_1 caminar_1_1 corer_1_1 adelantar_1_1 avanzar_1_4 ir_1_1 caminar_1_1 corer_1_1 adelantar_1_1 avanzar_1_6 ir_1_1 caminar_1_1 corer_1_1 adelantar_1_1 corer_1_17 ir_1_1 caminar_1_1 corer_1_1 adelantar_1_1 salir_1_35 ir_1_1 caminar_1_1 corer_1_1 carerear_1_2 ir_1_1 caminar_1_1 corer_1_1 circular_2_2 ir_1_1 caminar_1_1 corer_1_1 coretear_1_2 ir_1_1 caminar_1_1 corer_1_1 descender_1_2 ir_1_1 caminar_1_1 corer_1_1 descorer_1_3 ir_1_1 caminar_1_1 corer_1_1 desempedrar_1_2 ir_1_1 caminar_1_1 corer_1_1 despatilar_1_5 ir_1_1 caminar_1_1 corer_1_1 escapar_1_7 ir_1_1 caminar_1_1 corer_1_1 fluir_1_1 afluir_1_3 ir_1_1 caminar_1_1 corer_1_1 fluir_1_1 discurir_1_2 ir_1_1 caminar_1_1 corer_1_1 fluir_1_1 efluir_1_1 ir_1_1 caminar_1_1 corer_1_1 gambetear_1_2 ir_1_1 caminar_1_1 corer_1_1 lanzar_1_6 LE2-4003 EuroWordNet 174 EuroWordNet D006: Definition of the links and subsets for verbs ir_1_1 caminar_1_1 corer_1_1 lanear_1_2 ir_1_1 caminar_1_1 corer_1_1 mundear_1_1 ir_1_1 caminar_1_1 cruzar_1_13 ir_1_1 caminar_1_1 despezuÒarse_1_2 ir_1_1 caminar_1_1 faldear_1_1 ir_1_1 caminar_1_1 faldear_1_1 enfaldar_1_3 ir_1_1 caminar_1_1 ir_1_5 ir_1_1 caminar_1_1 marchar_1_4 ir_1_1 caminar_1_1 samurear_1_1 ir_1_1 caminar_1_1 surcar_1_3 ir_1_1 caminar_1_1 tanguear_1_1 ir_1_1 caminar_1_1 venir_1_1 ir_1_1 caminar_1_1 venir_1_1 caer_1_1 ir_1_1 caminar_1_1 venir_1_1 caer_1_1 aterizar_1_3 ir_1_1 caminar_1_1 venir_1_1 caer_1_1 deslizar_1_5 ir_1_1 caminar_1_1 venir_1_1 caer_1_1 rodar_1_14 ir_1_1 caminar_1_1 venir_1_1 caer_1_1 rodar_1_3 ir_1_1 caminar_1_1 venir_1_1 embestir_1_1 ir_1_1 caminar_1_1 venir_1_1 embestir_1_1 acometer_1_1 ir_1_1 caminar_1_1 venir_1_1 embestir_1_1 acometer_1_1 entrar_1_10 ir_1_1 caminar_1_1 venir_1_1 embestir_1_1 acometer_1_1 entrar_1_6 ir_1_1 caminar_1_1 venir_1_1 embestir_1_1 acometer_1_1 partir_1_6 ir_1_1 caminar_1_1 venir_1_1 embestir_1_1 avanzar_1_2 ir_1_1 caminar_1_1 venir_1_1 revolotear_1_2 ir_1_1 caminar_1_1 venir_1_1 salir_1_21 ir_1_1 caminar_1_1 volar_1_5 ir_1_1 corer_1_34 ir_1_1 destorentarse_1_2 ir_1_1 galopar_1_1 ir_1_1 galopar_1_1 galopear_1_1 ir_1_1 galopar_1_1 galuchar_1_1 ir_1_1 legar_1_10 ir_1_1 pasar_1_23 ir_1_1 peregrinar_1_2 ir_1_1 proceder_2_1 ir_1_1 proceder_2_1 descender_1_4 ir_1_1 proceder_2_1 salir_1_19 ir_1_1 proceder_2_1 tunear_1_2 ir_1_1 rastrear_1_6 ir_1_1 rodear_1_2 ir_1_1 rolar_1_2 ir_1_1 seguir_1_1 ir_1_1 seguir_1_1 cabestrear_1_1 LE2-4003 EuroWordNet 175 EuroWordNet D006: Definition of the links and subsets for verbs ir_1_1 seguir_1_1 cabestrear_1_1 ramalear_1_1 ir_1_1 seguir_1_1 cabestrear_1_4 ir_1_1 seguir_1_1 caminar_1_3 ir_1_1 seguir_1_1 curvear_1_1 ir_1_1 seguir_1_1 ir_1_3 ir_1_1 seguir_1_1 pasar_1_9 ir_1_1 seguir_1_1 perseguir_1_1 ir_1_1 seguir_1_1 perseguir_1_1 acosar_1_1 ir_1_1 seguir_1_1 perseguir_1_1 acosar_1_1 corer_1_18 ir_1_1 seguir_1_1 perseguir_1_1 acosar_1_1 seguir_1_4 ir_1_1 seguir_1_1 perseguir_1_1 acosar_1_1 sitiar_1_2 ir_1_1 seguir_1_1 perseguir_1_1 arinconar_1_2 ir_1_1 seguir_1_1 perseguir_1_1 asenderear_1_2 ir_1_1 seguir_1_1 perseguir_1_1 bandear_2_3 ir_1_1 seguir_1_1 perseguir_1_1 carabritear_1_1 ir_1_1 seguir_1_1 perseguir_1_1 colear_1_8 ir_1_1 seguir_1_1 perseguir_1_1 encorer_1_1 ir_1_1 seguir_1_1 perseguir_1_1 sabanear_1_4 ir_1_1 seguir_1_1 rumbar_1_4 ir_1_1 seguir_1_1 siguetear_1_1 ir_1_1 seguir_1_3 ir_1_1 seguir_1_7 ir_1_1 subir_1_8 ir_1_1 trotar_1_1 ir_1_1 trotar_1_1 trochar_1_1 ir_1_1 trotar_1_1 trotear_1_1 ir_1_1 trotar_1_1 trotinar_1_1 ir_1_1 volar_1_4 MOVER TAX mover I (1) aballar I (1) abrir I (7) adelantar I (1) agitar I (1) chapotear I (3) guachapear I (1) menear I (1) hopear I (1) hurgar I (1) hurgonear I (1) LE2-4003 EuroWordNet 176 EuroWordNet D006: Definition of the links and subsets for verbs rabear I (1) revolver I (1) agitar I (3) arrebujar I (3) batir I (6) mazar I (1) merengar I (1) trabajar I (5) buscar I (2) cacear I (1) enrehojar I (1) escaramuzar I (2) papelear I (1) trastear II (1) tabalear I (1) zalear I (1) revolver I (1) silbar I (2) tabalear I (1) traquetear I (2) ventilar I (2) vibrar I (5) alear I (1) aletear I (1) aletear I (2) apalancar I (1) bailar I (1) bailotear I (1) danzar I (1) polcar I (1) tripudiar I (1) valsar I (1) valsar I (1) bandear I (1) batir I (5) batir I (6) befar I (1) blandir I (1) blandear II (1) doblegar I (2) bocezar I (1) bocear I (1) bornear I (3) LE2-4003 EuroWordNet 177 EuroWordNet D006: Definition of the links and subsets for verbs bracear I (1) cambiar I (5) contrabracear I (1) cabalgar I (2) cabecear I (1) cavar I (1) agostar I (2) agostar I (3) azadonar I (1) entrecavar I (1) escavanar I (1) escavar I (1) escopetar I (1) hornaguear I (1) jirpear I (1) rebinar I (1) cerner I (7) codear I (1) colear I (1) contonearse I (1) campanear I (4) remenearse I (1) zarandear I (3) florear I (3) goncear I (1) ijadear I (1) levantar I (1) alfar II (1) alzar I (1) elevar I (1) bombear I (3) remontar I (2) piafar I (1) atabalear I (1) sobrealzar I (1) sofaldar I (1) apalancar I (1) arrezagar I (2) aupar I (1) upar I (1) cavar I (1) desvenar I (4) empinar I (1) LE2-4003 EuroWordNet 178 EuroWordNet D006: Definition of the links and subsets for verbs empingorotar I (1) enarbolar I (1) arbolar I (1) tremolar I (1) encampanar I (3) encaramar I (1) engarriar I (1) encumbrar I (1) engallarse I (1) enhestar I (1) erguir I (1) escalar I (4) hozar I (1) hocicar I (1) incorporar I (2) realzar I (1) rebotar I (5) solevantar I (1) solevar I (2) sopesar I (1) sompesar I (1) sospesar I (1) suspender I (1) lomear I (1) menear I (1) motorizar I (1) mecanizar I (1) nalguear I (1) orejear I (1) palear I (2) patalear I (1) pernear I (1) pestaÒear I (1) rabear I (2) remar I (1) bogar I (1) bogar I (2) ciar I (1) paletear I (1) proejar I (1) singar I (1) remecer I (1) remolinear I (1) LE2-4003 EuroWordNet 179 EuroWordNet D006: Definition of the links and subsets for verbs rizar I (3) ronzar II (1) arronzar I (1) sacudir I (1) desgargolar I (1) despolvorear I (1) espolvorear I (1) escodar II (1) zamarrear I (1) teclear I (1) mover I (3 4 5) compungir I (1) conmover I (1) decidir I (2) despertar I (4) enternecer I (2) incitar I (1) inducir I (1) ocasionar I (2) quebrantar I (8) mover I (7) mover I (8) mover I (9) MOVERSE TAX moverse I (1) agarrotar I (5) andar II (1) abordonar I (1) acuartillar I (2) amblar I (1) anadear I (1) nanear I (1) apeonar I (1) atrochar I (1) barbear I (4) caminar I (2) arrear I (5) cruzar I (13) faldear I (1) ladear I (3) venir I (1) LE2-4003 EuroWordNet 180 EuroWordNet D006: Definition of the links and subsets for verbs advenir I (1) llegar I (1) revolotear I (2) sobrevenir I (2) cejar I (1) garrar I (1) garrear I (1) recular I (1) recejar I (1) retrechar I (1) cerner I (6) chancletear I (1) circular II (1) roldar I (1) cochear I (3) cojear I (1) culebrear I (1) danzar I (3) deambular I (1) rular I (3) discurrir I (1) errar I (3) escarabajear I (1) gatear I (2) jinetear I (1) juntar I (5) ladear I (3) llanear I (1) lustrar I (3) navegar I (1) navegar I (3) noctambular I (1) parrandear I (1) pasear I (1) deambular I (1) ruar I (2) pasear I (3) pecorear I (3) peregrinar I (1) lustrar I (3) raquear I (1) ratear III (1) renquear I (1) LE2-4003 EuroWordNet 181 EuroWordNet D006: Definition of the links and subsets for verbs reptar I (1) retrasar I (3) desacelerar I (1) rodear I (1) abarcar I (5) cercar I (1) cercar I (3) circuir I (1) circundar I (1) circunvalar I (1) rondar I (2) rondar I (3) roldar I (1) talonear I (1) tesar I (2) trafagar I (1) trafagar I (4) trajinar I (2) travesear I (1) vagar III (1) flanear I (1) vaguear I (1) vagar III (3) zancajear I (1) zigzaguear I (1) andar II (3) marchar I (1) navegar I (2) bailar I (3) escarabajear I (5) bailar I (6) bambolear I (1) bambalear I (1) bambanear I (1) bambonear I (1) blandir I (2) bullir I (3) bullir I (5) cabecear I (3) circular II (1) danzar I (2) girar I (1) arribar I (6) LE2-4003 EuroWordNet 182 EuroWordNet D006: Definition of the links and subsets for verbs bailar I (5) bornear I (5) campanear I (2) orbitar I (1) hervir I (1) borbotar I (1) borboritar I (1) brollar I (1) bullir I (1) hojear I (3) hornaguear I (2) mandar I (8) manejar I (5) marchar I (1) menear I (4) mimbrear I (1) ondular I (1) ondear I (3) undular I (1) oscilar I (1) bascular I (2) titubear I (1) tantear I (7) titubar I (1) palpitar I (2) rehilar I (2) relingar I (3) revolver I (14) serpentear I (1) serpear I (1) tartalear I (1) vacilar I (1) fluctuar I (1) trastrabillar I (2) venir I (1) vibrar I (1) florear I (3) volar I (1) circunvolar I (1) planear I (4) remontar I (12) revolear I (1) revolotear I (1) LE2-4003 EuroWordNet 183 EuroWordNet D006: Definition of the links and subsets for verbs revolar I (2) abejear I (1) voltijear I (1) sobrevolar I (1) zangolotear I (3) flanear I (1) zapatear I (7) zigzaguear I (1) 5.4.3 Know verb synsets and taxonomies VERBS OF KNOWING EXTRACTED SO FAR: taxonomies of 'conocer' and 'saber' CONOCER TAX adivinar_1_4 conocer_1_1 desconocer_1_3 determinar_1_4 discernir_1_1 distinguir_1_1 dominar_1_3 individualizar_1_1 individuar_1_1 individuar_1_2 palpar_1_3 prevenir_1_2 pronosticar_1_1 saber_2_9 separar_1_4 singularizar_1_1 subdistinguir_1_1 tocar_1_8 SABER TAX saber_2_1 ignorar_1_1 resaber_1_1 LE2-4003 EuroWordNet 184 EuroWordNet D006: Definition of the links and subsets for verbs References Alonge, A. (1992a) ‘Machine-Readable Dictionaries and Lexical Information on Verbs’, in Proceedings of the 5th Euralex International Congress, Tampere, Finland Alonge, A. (1992b) Acquisizione di informazioni semantico-sintattiche relative a verbi: analisi di dizionari di macchina, Tesi di Dottorato, Università di Pisa. Bloksma, L., P. L. Díez-Orzas, P. Vossen (1996) User Requirements and Functional Specification of the EuroWordNet Project, Deliverable D001, EuroWordNet, LE2-4003. Climent, S., H. Rodriguez, J. Gonzalo (1996) Definition of the Links and Subsets for Nouns of the EuroWordNet Project, Deliverable D005, EuroWordNet, LE2-4003. Cruse (1986) Lexical Semantics, Cambridge, Cambridge University Press. Díez-Orzas, P. L., P. Forest, M. Louw (1996) High-Level Architecture of the EuroWordNet Database. A Novell ConceptNet-Based Semantic Network, Deliverable D007, EuroWordNet, LE2-4003. Fellbaum (1993) ‘English Verbs as a Semantic Net’, ms. Gruber, J. (1976)Lexical Structures in Syntax and Semantics, Amsterdam, North-Holland. Haas, W. (1964) ‘Semantic Value’, in Proceedings of the IXth International Congress of Linguists,, The Hague, Mouton. Levin, B. (1993) English verb classes and alternations: a preliminary investigation, Chicago, The University of Chicago Press. Lyons, J. (1968) Introduction to Theoretical Linguistics, Cambridge, Cambridge University Press. Lyons, J. (1977) Semantics, London, Cambridge University Press. Miller, G, R. Beckwith, C. Fellbaum, D. Gross and K. Miller Five papers on WordNet, CSL Report 43. Cognitive Science Laboratory. Princeton University. 1990. Miller, G., R. Beckwith, C. Fellbaum, D Gross, K. Miller (1993) ‘Introduction to WordNet: An On-Line Lexical Database’, ms. Montemagni S., (1995) Subject and Object in Italian Sentence Processing, PhD Dissertation, Umist Manchester, UK. LE2-4003 EuroWordNet 185 EuroWordNet D006: Definition of the links and subsets for verbs Sanfilippo, A., T. Briscoe, A. Copestake, M. A. Martì Antonin, A. Alonge (1992) ‘Translation Equivalence and Lexicalization in the ACQUILEX LKB’, in Proceedings of the 4th International Conference on Theoretical and Methodological Issues in Machine Translation, Montreal, Canada. Talmy, L. (1985) ‘Lexicalization Patterns: Semantic Structure in Lexical Form’, in Shopen, T. (cur.), Language Typology and Syntactic Description: Grammatical Categories and the Lexicon, Cambridge, Cambridge University Press. Wright, G. H. von (1963) Norm and Action, New York, Humanities Press. LE2-4003 EuroWordNet