Download Semantic Predicative Analysis for Resolving Some Cases

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Embodied language processing wikipedia , lookup

Transcript
2015 17th UKSIM-AMSS International Conference on Modelling and Simulation
Semantic Predicative Analysis for Resolving Some Cases of Ambiguous Referents of
Pronoun “Nó” in Summarizing Meaning of Two Vietnamese Sentences
Trung Tran
Dang Tuan Nguyen
Faculty of Computer Science
University of Information Technology, Vietnam
National University – Ho Chi Minh City
Ho Chi Minh City, Vietnam
[email protected]
Faculty of Computer Science
University of Information Technology, Vietnam
National University – Ho Chi Minh City
Ho Chi Minh City, Vietnam
[email protected]
Discourse Representation Theory) denoting nouns, verbs or
adjectives and relates to indexes in U. After that, we
represent main predicates of the DRS structure (which are
semantic predicates of lexicons containing the main content
of the source pair of Vietnamese sentences) by a diagram.
This diagram helped for easily illustrating relationships and
building the syntactic structure of the new meaningsummarizing Vietnamese sentence.
Thus, the study approach in [18] is different from other
approaches in traditional researches in NLG field (cf. [3],
[4]) at three important points:
Abstract—In this paper, with the combination of text
summarization and generation ideas, we introduce a method
for summarizing some types of two-sentences based
Vietnamese paragraphs. The main characteristic of each pair
is that two sentences associate together by pronoun “nó”
appearing at the second sentence – the special one in this kind
of lexical category in Vietnamese, can indicate human,
animated or non-animated object depending on the context of
the paragraph. Our method idea is that we generate a new
Vietnamese sentence having the content which summarizes the
semantic of the original pair. This method composes following
tasks: i) find the object which is denoted by pronoun “nó”; ii)
model the semantic of the source sentence pair by a logical
form; iii) identify factors representing relationships between
nouns, verbs and adjectives; iv) generate the syntactic
structure of the new reducing sentence; v) build the lexical set
and complete the new sentence. Applying this method in the
experiment, we firstly identify the antecedence for each
anaphoric pronoun and build the logical form for testing
paragraphs. Then, we test the ability to generate the new
sentence from each logical form. The results show that new
Vietnamese sentences satisfy the given requirements.
TABLE I.
COMPARISON OF DIFFERENT APPROACHES FOR
GENERATING NEW SUMMARIZING SENTENCES
Point
The study approach in
[18]
1
Generate the new sentence
to summarize the original
pair of sentences.
2
The input and output of the
summarization model are
complete sentences.
3
Lexicons
used
for
generating
the
new
sentence are also lexicons
of the original pair of
sentences.
Keywords-sentence generation; inter-sentential anaphora;
discourse representation; meaning summarization
I.
INTRODUCTION
Generating a new complete sentence for summarizing the
meaning of the original pair of sentences is an essential
objective in Natural Language Generation (NLG) field.
Besides, this is also a new approach in Text Summarization
field (cf. [1], [2], [7], [8], [9], [10]). With this objective, a
solution was introduced in [18] to summarize four forms of
two-sentences paragraphs in Vietnamese. The primary
considered objects in [18] are pairs of Vietnamese sentences
which in each pair, there is the appearance of anaphoric
pronoun indicating human and standing with a demonstrative
adjective [“ta” / “ấy” / “này”]. With this characteristic, we
firstly applied method and techniques in [17] to find the noun
indicating person and taking the object role of the transitive
verb at the first sentence and identified it was the
antecedence of the pronoun. Secondly, we built a Discourse
Representation Structure – DRS (cf. [5], [12], [13], [16]) – to
represent the meaning of the original pair with a 2-tuple <U,
Con>: U is an ordered list includes indexes denoting noun;
and Con is an ordered list includes predicates (in the sense of
978-1-4799-8713-9/15 $31.00 © 2015 IEEE
DOI 10.1109/UKSim.2015.42
Traditional approaches ([3],
[4])
Generate text from some
nonlinguistic data for many
purposes.
The input and output of the
research are not complete
sentence or paragraph, but other
forms for human understanding.
Lexicons used for generating
the output text are different
from lexicons of the input
information.
In this research, following the idea [18], we mainly
consider four other types of pairs of Vietnamese sentences
which have the main characteristic: pronoun “nó” appears at
the second sentence and can associates to a person or thing
(animate or non-animated object) appearing at the first
sentence. To summarize the meaning of each sentence pair
belonging to these types, our method in this study contains
following steps:
x Step 1: Apply method and techniques in [17] with
an improvement to find exactly the person or
animated or non-animated object which is the
associated object of pronoun “nó”.
x Step 2: Build the DRS structure for each sentence
pair. With the above improvement we can identify
factors representing relationships between nouns,
verbs and adjectives through indexes in list U and
semantic predicates in list Con of the DRS structure.
340
x
x
Step 3: Generate the syntactic structure of the new
meaning-summarizing Vietnamese sentence after
having relationship factors.
x Step 4: Define the object class so that we can build a
set of lexicons which is used to complete the new
sentence.
The structure of this paper: in section II we present the
considered types of pairs of Vietnamese sentences in this
research and their DRS structures; in section III we present
how to generate the new reducing Vietnamese sentence; and
finally, in section IV we present experiment and analysis.
II.
Example 1: “Con chim tới. Nó ăn bánh mì.”
(English: The bird arrives. It eats the bread.)
Ö The DRS structure of this example:
U: [1,2]
Con: con_chim(1,[con,chim],noun,common)
tới(1,[tới],verb,intransitive)
bánh_mì(2,[bánh,mì],noun,common)
ăn(1,2,[ăn],verb,transitive)
TYPES OF PAIRS OF VIETNAMESE SENTENCES AND
THEIR DRS STRUCTURES
Figure 1. DRS structure of the pair of sentences “Con chim tới. Nó ăn
bánh mì.”.
In this section, we present pairs of Vietnamese sentences
which are mainly considered in this research and building the
DRS structure for sentence pair of each type.
In [17], we presented a method for resolving intersentential anaphoric pronouns including pronoun “nó” and
building structure DRS of pairs of simple Vietnamese
sentences. We use the method of [17], only perform a small
improvement suitable for the aim of the research: when
describing grammatical characteristics of lexicons, add some
information into the lexical semantic predicate. The purpose
of adding these information is when analyzing structure DRS
of the pair of simple Vietnamese sentences, we exactly
identify objects (which are nouns), actions (which are verbs),
properties (which are adjectives) and the morphology content
of the lexicon. The added information according to each
category is:
x Noun: morphology content of the lexicon, category –
noun, noun classification – proper / common.
Consider proper noun “Nhân”, describe grammatical
characteristics: {semantic Æ [named]}; {index Æ
I};
{content
Æ
[nhân]};
[noun]}; {class Æ [proper]}.
Ö The
{category
semantic
x
U: [1,2]
Con: named(1,[tín],noun,proper)
căn_hộ(2,[căn,hộ],noun,common)
có(1,2,[có],verb,transitive)
ngăn_nắp(2,[ngăn,nắp],adjective)
Figure 2. DRS structure of the pair of sentences “Tín có một căn hộ. Nó
ngăn nắp.”.
Æ
x
predicate:
Verb: morphology content of the lexicon, category –
verb, verb classification – transitive / intransitive.
Consider transitive verb “ăn” (English: eat), describe
grammatical characteristics: {semantic Æ [ăn]};
{arg1 Æ Arg1}; {arg2 Æ Arg2}; {content Æ
[ăn]};
{categoy
Æ
[verb]};
{class
Æ
[transitive]}.
Ö The
semantic
predicate:
U: [1,2]
Con: named(1,[nhân],noun,proper)
căn_hộ(2,[căn,hộ],noun,common)
có(1,2,[có],verb,transitive)
thuê(1,2,[thuê],verb,transitive)
Adjective: morphology content of the lexicon,
category – adjective. Consider adjective “ngăn nắp”
(English: neat), describe grammatical characteristics:
{semantic
Æ
{content
Æ
[adjective]}.
Ö The
[ngăn_nắp]};
[ngăn,nắp]};
semantic
Type 3: The first sentence has: one human referent
takes the subject role of the transitive verb, one thing
(animate or non-animate object) takes the object role
of the transitive verb. The second sentence has: one
pronoun indicates person, stands alone and takes the
subject role of the transitive verb, one pronoun “nó”
takes the object role of the transitive verb.
Example 3: “Nhân có một căn hộ. Anh thuê nó.”
(English: Nhân has an apartment. He hires it.)
Ö The DRS structure of this example:
ăn(Arg1,Arg2,[ăn],verb,transitive).
x
Type 2: The first sentence has: one human referent
takes the subject role of the transitive verb, one thing
(animate or non-animate object) takes the object role
of the transitive verb. The second sentence has: one
pronoun “nó” takes the subject role of the
intransitive verb or adjective.
Example 2: “Tín có một căn hộ. Nó ngăn nắp.”
(English: Tín has an apartment. It is neat.)
Ö The DRS structure of this example:
named(I,[nhân],noun,proper).
x
Type 1: The first sentence has: one thing (animate or
non-animate object). The second sentence has: one
pronoun “nó” takes the subject role of the transitive
verb.
{arg
Æ
I};
{category
Æ
Figure 3. DRS structure of the pair of sentences “Nhân có một căn hộ.
Anh thuê nó.”.
predicate:
x
ngăn_nắp(I,[ngăn,nắp],adjective).
With techniques in [17] and the above improvements, we
apply for following types of Vietnamese sentences in this
research:
Type 4: The first sentence has: two identical human
referents. The second sentence has: one pronoun
“nó” takes the subject role of the transitive verb.
Example 4: “Nhân là một đứa bé. Nó tới trường.”
341
(English: Nhân is a little boy. He goes to school.)
Ö The DRS structure of this example:
U: [1,2]
Con: named(1,[nhân],noun,proper)
căn_hộ(2,[căn,hộ],noun,common)
có(1,2,[có],verb,transitive)
thuê(1,2,[thuê],verb,transitive)
x
Figure 4. DRS structure of the pair of sentences “Nhân là một đứa bé. Nó
tới trường.”.
III.
GENERATING THE NEW REDUCING VIETNAMESE
SENTENCE
in which arg takes the value is the index of one
object in list U.
o Show that the object takes the subject role of the
action or property in the sentence.
The relation between two semantic predicates of
noun:
o Represented by the identical semantic predicate,
which has form arg1=arg2, in which arg1 and
arg2 take the values are the indexes of two
objects in list U.
o Show that these two objects are identical.
In Table II, we present representations of these relations
based on each structure DRS in section II. Based on these
representations, we also give the expected syntax structure of
the new reducing sentence.
In this section, we present techniques for generating the
new reducing Vietnamese sentence. The new reducing
Vietnamese sentence will summarize the meaning of the
input pair of Vietnamese sentences. The proposed techniques
have following main points:
x Analyze objects in list U and semantic predicates in
list Con of structure DRS.
x With above analyzing, represent relations: between
semantic predicates of nouns (corresponding objects
in list U); between the semantic predicate of noun
and the semantic predicate of verb or adjective.
x With above representing, generate the syntactic
structure of the new reducing sentence with the
appropriate algorithm according to each type of pair
of sentences.
x Build the set of lexicons, in which each lexicon has
grammatical characteristics corresponding to the
information in the semantic predicate of lexicon.
x Combine the syntactic structure with the set of
lexicons, complete the new reducing Vietnamese
sentence.
These main points are presented in details as follows:
TABLE II.
REPRESENT RELATIONS BASED ON DRS STRUCTURES OF
PAIRS OF SENTENCES ACCORDING TO TYPES IN SECTION II
Type
Represent relations based on DRS structure and the
expected syntactic structure of the new reducing sentence
x
x
1
2
A. Generating the Syntactic Structure of the New Reducing
Vietnamese Sentence
Analyzing structure DRS of pairs of sentences according
to types in section II, especially considering the semantic
predicate of verbs and adjectives, we see that there are
relations:
x The relation between the semantic predicate of noun
and the semantic predicate of transitive verb:
o Represented by the semantic predicate of
transitive
verb,
which
has
form
transitive_verb(arg1, arg2), in which arg1
and arg2 take the values are the indexes of the
two objects in list U.
o Show that one object takes the subject role, one
object takes the object role of the action in the
sentence.
x The relation between the semantic predicate of noun
and the semantic predicate of intransitive verb or
adjective:
o Represented by the semantic predicate of
intransitive verb or adjective, which has form
intransitive_verb(arg) or adjective(arg),
3
4
(1) Æ [tới]: (1) is the subject of action [tới].
(1) Æ [ăn] Æ (2): (1) is the subject of action
[ăn], (2) is the object of action [ăn].
Î Comment: (1) performs two continuous actions are [tới]
and [ăn], (2) is the object of one action [ăn].
Î The expected syntactic structure: (1) + [tới] +
[[ăn] + (2)].
x
(1) Æ [có] Æ (2): (1) is the subject of action
[có], (2) is the object of action [có].
x
(2) Æ [ngăn nắp]: (2) is the subject of property
[ngăn nắp].
Î The expected syntactic structure: (1) + [có] + (2) +
[ngăn nắp].
x
(1) Æ [có] Æ (2): (1) is the subject of action
[có], (2) is the object of action [có].
x
(1) Æ [thuê] Æ (2): (1) is the subject of action
[thuê], (2) is the object of action [thuê].
Î Comment: (1) performs two continuous actions are [có]
and [thuê], (2) is the object of these two actions.
Î The expected syntactic structure: (1) + [[có] and
[thuê]] + (2).
x
(1) = (2): (1) is identical with (2).
x
(2) Æ [tới] Æ (3): (2) is the subject of action
[tới], (3) is the object of action [tới].
Î The expected syntactic structure: [(1) is (2)] +
[tới] + (3).
Base on the above analysis, representation, comments
and expected syntactic structure of the new reducing
sentence, we propose the general algorithm for generating
the syntactic structure of the new reducing sentence as
follow:
Consider the first semantic predicate of verb or
adjective
If is the identical predicate then
Add object (1) into structure;
Add “là” (is) into structure;
342
Add object (2) into structure;
If is the predicate intransitive verb or
adjective then
Add object (1) into structure;
Add predicate intransitive verb or adjective
into structure;
If is the predicate transitive verb then
Add object (1) into structure;
Add predicate transitive verb into structure;
Consider the second semantic predicate of verb or
adjective
If is the predicate adjective then
Add object (2) into structure;
Add predicate adjective into structure;
If is the predicate transitive verb then
If this predicate relates to object (1) and
object (2) then
Add “và” (and) into structure;
Type
Add object (2) into structure;
Add the second predicate adjective
into structure;
Figure 7. The algorithm for type 2.
Î The general syntactic structure: [object (1)] +
[predicate transitive verb (1)] + [object
(2)] + [predicate adjective (2)]
Add object (1) into structure;
Add the first predicate transitive
verb into structure;
Add “và” (and) into structure;
Add the second predicate transitive
3
Î The general syntactic structure: [object (1)] +
[predicate transitive verb (1)] + [“và”] +
[predicate transitive verb (2)] + [object
(2)]
Add
Add
Add
Add
Add
Figure 5. The general algorithm for generating the syntactic structure for
the new reducing sentence.
4
In Table III, we present in detail the algorithm for
generating the syntactic structure of the new reducing
sentence according to each type:
verb into structure;
Add object (3) into structure;
Î The general syntactic structure: [object (1)] +
[“là”] + [object (2)] + [“và”] +
[predicate transitive verb (2)] + [object
(3)]
B. Building the Set of Lexicons
To build the set of lexicons suitable for the research aim,
we define three lexical classes corresponding to three
categories which are noun, verb, and adjective. The defined
attributes in these classes corresponding to the information in
the semantic predicate noun, verb, and adjective (described
in Section II). The detailed descriptions of these classes are:
x The class Noun: {flagSemantic: predicate name –
[named] if is proper noun, morphology if is common
noun}; {flagIndex: unique index of each object};
{flagContent:
content};
{flagMorphology:
morphology}; {flagCategory: category – [noun]};
{flagClass: sub-category – [proper] if is proper
noun, [common] if is common noun}.
x The class Verb: {flagSemantic: predicate name –
morphology}; {flagArg1: index of the object taking
the subject role}; {flagArg2: index of the object
taking the object role}; {flagContent: content};
{flagMorphology: morphology}; {flagCategory:
Algorithm for generating the syntactic structure of the
reducing sentence
Add object (1) into structure;
Add the first predicate intransitive
verb into structure;
Add “và” (and) into structure;
Add the second predicate transitive
verb into structure;
Add object (2) into structure;
Figure 6. The algorithm for type 1.
Î The general syntactic structure: [object (1)] +
[predicate intransitive verb (1)] + [“và”]
+ [predicate transitive verb (2)] +
[object (2)]
2
object (1) into structure;
“là” (is) into structure;
object (2) into structure;
“và” (and) into structure;
the second predicate transitive
Figure 9. The algorithm for type 4.
TABLE III.
THE ALGORITHM FOR GENERATING THE SYNTACTIC
STRUCTURE OF THE NEW REDUCING SENTENCE ACCORDING TO EACH
TYPE
1
verb into structure;
Add object (2) into structure;
Figure 8. The algorithm for type 3.
Add predicate transitive verb into
structure;
Add object (2) into structure;
If this predicate relates to object (2) and
object (3) then
Add “và” (and) into structure;
Add predicate transitive verb into
structure;
Add object (3) into structure;
Type
Algorithm for generating the syntactic structure of the
reducing sentence
Add object (1) into structure;
Add the first predicate transitive
verb into structure;
343
In Table IV, with the algorithm for generating the
complete new reducing Vietnamese sentence, based on the
general syntactic structure of the new reducing sentence
according to each type in Table III, we present the new
reducing Vietnamese sentence for the example pair of
Vietnamese sentences in section II:
category – [verb]}; {flagClass: sub-category –
[transitive] if is transitive verb, [intransitive] if is
intransitive verb}.
x The class Adjective: {flagSemantic: predicate
name – morphology}; {flagArg: index of the object
taking the subject role}; {flagContent: content};
{flagMorphology: morphology}; {flagCategory:
category – [adjective]}.
To be consistent with the research aim, our approach
when building the set of lexicons is reuse the entire lexicons
which are described in [17]. To perform, we describe
lexicons according to lexical classes which are presented
above. The idea for this approach is based on representations
in Table II and syntactic structures in Table III, while just
reuse existing lexicons is enough to generate the new
reducing Vietnamese sentence. The description is illustrated
with lexicons are described in Section II as follows:
x Consider proper noun “Nhân” described in section
II. We define object nNhan belonging to class Noun
with attributes and values: {flagSemantic Æ
[named], flagIndex Æ [],
[nhân],
flagMorphology
flagCategory
Æ
[noun],
[proper]}.
x
TABLE IV.
Type
1
2
flagContent Æ
Æ
[Nhân],
flagClass
Æ
3
Consider transitive verb “ăn” described in section II.
We define object vAn belonging to class Verb with
attributes and values: {flagSemantic Æ [ăn],
flagArg1 Æ [], flagArg2 Æ [], flagContent
Æ
[ăn],
flagCategory
Æ
[verb],
flagMorphology
Æ
[ăn],
flagClass
Æ
[intransitive]}.
x
COMPLETE THE NEW REDUCING VIETNAMESE SENTENCE
ACCORDING TO EACH TYPE
4
Consider adjective “ngăn nắp” described in section
II. We define object aNganNap belonging to class
Adjective
with
attributes
and
values:
The new reducing Vietnamese sentence
Î The syntactic structure: [con chim] + [tới] + “và” + [ăn] +
[bánh mì]
Î The complete Vietnamese sentence: “con chim tới và ăn
bánh mì”
(English: the bird arrives and eats bread)
Î The syntactic structure: [Tín] + [có] + [căn hộ] + [ngăn nắp]
Î The complete Vietnamese sentence: “Tín có một căn hộ
ngăn nắp”
(English: Tín has an neat apartment)
Î The syntactic structure: [Nhân] + [có] + “và” + [thuê] +
[căn hộ]
Î The complete Vietnamese sentence: “Nhân có và thuê một
căn hộ”
(English: Nhân has and hires an apartment)
Î The syntactic structure: [Nhân] + “là” + [đứa bé] + “và” +
[tới] + [trường]
Î The complete Vietnamese sentence: “Nhân là một đứa bé và
tới trường”
(English: Nhân is the boy and goes to school)
IV.
EXPERIMENT AND ANALYSIS
The tested data in this research are 54 pairs of
Vietnamese sentences satisfying characteristics according to
types in section II. Applying the method and techniques in
[17], we identified the antecedence for anaphoric pronouns
and built DRS structures for all these 54 pairs.
In this research, with the improvement presented in
section II, the system built DRS structures for all these 54
pairs of sentences, with predicates containing information
suitable for the research aim. Then, based on built DRS
structures, the system generated 54 syntactic structures and
complete new Vietnamese sentences. With the cognition of
native speaker, these 54 new Vietnamese sentences satisfy
the requirement is understandable and summarize the
meaning of input pair of Vietnamese sentences.
Thus, with this approach, the system can generate the
new reducing Vietnamese sentence if can build the DRS
structure for input pair of Vietnamese sentences. However,
this lead to some limitations:
x For pairs of Vietnamese sentences which the system
cannot build DRS structure, then will not generate
new reducing Vietnamese sentences.
x For paragraphs composing more than two sentences:
in this research, we did not propose the solution for
generating new reducing Vietnamese sentences for
summarizing the meaning of these paragraphs.
{flagSemantic Æ [ngăn_nắp], flagArg Æ [],
flagContent Æ [ngăn,nắp], flagMorphology Æ
[ngăn nắp], flagCategory Æ [adjective]}.
One notice here is at the time of definition, attributes
of lexical objects
have not been assigned values. These attributes will be
assigned appropriate values when analyzing semantic
predicates of noun, verb, adjective.
flagIndex, flagArg1, flagArg2, flagArg
C. Completing the Summarizing Vietnamese Sentence
With the generated syntactic structure and the described
set of lexicons, we generate the complete reducing
Vietnamese sentence with the following algorithm:
Browse each component in structure
If is object or predicate then
Choose the lexical object having attributes
taking corresponding values;
Add the morphology of this lexicon into
sentence;
If is linking word [“và” / “là”] then
Add this word into sentence;
Figure 10. The algorithm for generating the complete new reducing
Vietnamese sentence.
344
These limitations are challenges which we will find the
solution to overcome in next researches.
V.
[6]
DISCUSSION AND CONCLUSION
[7]
In this paper, we presented in detail techniques for
summarizing the meaning of some types of pairs of simple
Vietnamese sentences through representing the meaning by
DRS structure and generating the new reducing Vietnamese
sentence based on this DRS structure. With the current
approach, we built a system to combine two tasks are
understanding the pair of Vietnamese sentences (in field of
Natural Language Understanding) and generating the new
reducing Vietnamese sentence (in field of Natural Language
Generation).
The experiment show that, the system generated the new
reducing Vietnamese sentence for pairs of Vietnamese
sentences which the system can built DRS structures.
Besides, we mentioned some limitations in the current
approach when considering many different types of
Vietnamese paragraphs.
In next works, we will continue to further studies under
the current approach to be able to suggest better solutions.
[8]
[9]
[10]
[11]
[12]
[13]
REFERENCES
[1]
[2]
[3]
[4]
[5]
[14]
D. Das and A. F. T. Martins, “A survey on automatic text
summarization”, Language Technologies Institute, Carnegie Mellon
University, 2007.
E. Lloret, “Text summarization: an overview”, paper supported by the
Spanish Government under the project TEXT-MESS (TIN200615265-C06-01), 2008.
E. Reiter and R. Dale, Building Natural Language Generation
System. Cambridge University Press, 1997.
E. Reiter and R. Dale, “Building Applied Natural Language
Generation Systems”, Natural Language Engineering, vol. 3, no. 1,
pp. 57–87. 1997.
H. Kamp, “A theory of truth and semantic representation”. In
Groenendijk, Jeroen, Janssen, Theo M. V and Stokhof, Martin (eds.),
Formal Methods in the Study of Language, Part 1, 277–322.
Mathematical Centre Tracts, 1981.
[15]
[16]
[17]
[18]
345
H. X. Cao, “Tiếng Việt: Sơ thảo ngữ pháp chức năng” [Vietnamese:
Brief of Functional Grammar]. Nhà xuất bản giáo dục [Education
Publisher], 2006.
I. Mani and M. T. Maybury, “Advances in Automatic Text
Summarization”. MIT Press, 1999.
K. Jezek and J. Steinberger, “Automatic Text summarization”, Vaclav
Snasel (Ed.): Znalosti 2008, ISBN 978-80-227-2827-0, FIIT STU
Brarislava, Ustav Informatiky a softveroveho inzinierstva, pp. 1–12,
2008.
K. S. Jones, “Automatic summarizing: factors and directions”, in:
Mani, I. and Marbury, M., editors, Advances in Automatic Text
Summarization, MIT Press, 1999.
K. S. Jones, “Automatic summarising: a review and discussion of the
state of the art”, Technical Report 679, Computer Laboratory,
University of Cambridge, 2007.
M. A. Covington, “GULP 4: An Extension of Prolog for Unification
Based Grammar”, Research Report number: AI-1994-06. USA:
Artificial Intelligence Center, The University of Georgia, 2007.
M. A. Covington and N. Schmitz, “An Implementation of Discourse
Representation Theory”, ACMC Research Report number: 01-0023.
Advanced Computational Methods Center, The University of
Georgia, 1989.
M. A. Covington, D. Nute, N. Schmitz and D. Goodman, “From
English to Prolog via Discourse Representation Theory”, ACMC
Research Report number: 01-0024. Advanced Computational
Methods Center, University of Georgia, 1988.
M. A. K. Halliday and C. M. I. M. Matthiessen, “An Introduction to
Functional Grammar”, Third Edition. Hodder Arnold, 2004.
M. Johnson and E. Klein, “Discourse, anaphora and parsing”, Report
number: CSLI-86-63. USA: Center for the Study of Language and
Information, Stanford University, 1986.
P. Blackburn and J. Bos, “Representation and Inference for Natural
Language – Volume II: Working with Discourse Representation
Structures”. Department of Computational Linguistics, University of
Saarland, 1999.
T. Tran and D. T. Nguyen, “A Solution for Resolving Inter-sentential
Anaphoric Pronouns for Vietnamese Paragraphs Composing Two
Single Sentences”, Proc. of the 5th Int. Conf. of Soft Computing and
Pattern Recognition (SoCPaR 2013), Hanoi, Vietnam, 2013, pp. 172–
177.
T. Tran and D. T. Nguyen, “Merging Two Vietnamese Sentences
Related by Inter-sentential Anaphoric Pronouns for Summarizing”,
Proc. The 1st NAFOSTED Conf. on Information and Computer
Science, Hanoi, Vietnam, 2014, pp. 371–381.