* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Full Paper
Ojibwe grammar wikipedia , lookup
Modern Greek grammar wikipedia , lookup
Lithuanian grammar wikipedia , lookup
Japanese grammar wikipedia , lookup
Udmurt grammar wikipedia , lookup
Malay grammar wikipedia , lookup
Esperanto grammar wikipedia , lookup
Navajo grammar wikipedia , lookup
Portuguese grammar wikipedia , lookup
Scottish Gaelic grammar wikipedia , lookup
Macedonian grammar wikipedia , lookup
Kannada grammar wikipedia , lookup
English clause syntax wikipedia , lookup
Proto-Indo-European verbs wikipedia , lookup
Modern Hebrew grammar wikipedia , lookup
Old Norse morphology wikipedia , lookup
Latin syntax wikipedia , lookup
Spanish grammar wikipedia , lookup
Old Irish grammar wikipedia , lookup
Ukrainian grammar wikipedia , lookup
Germanic weak verb wikipedia , lookup
Polish grammar wikipedia , lookup
Ancient Greek grammar wikipedia , lookup
Ancient Greek verbs wikipedia , lookup
Swedish grammar wikipedia , lookup
Germanic strong verb wikipedia , lookup
Italian grammar wikipedia , lookup
Georgian grammar wikipedia , lookup
Sotho verbs wikipedia , lookup
Hungarian verbs wikipedia , lookup
Russian grammar wikipedia , lookup
Kagoshima verb conjugations wikipedia , lookup
Old English grammar wikipedia , lookup
Lexical semantics wikipedia , lookup
Icelandic grammar wikipedia , lookup
Yiddish grammar wikipedia , lookup
Interpreting Voice in W. Austronesian Edward L. Keenan 3/31/06 abstract The voice system in Malagasy is typically W. Austronesian1. Nuclear Ss present DPs with the same theta role in different syntactic configurations using verbs with different voice morphology. Such Ss are theta equivalent in the way that actives and their passives are in English. In this article we provide an explicit means of deriving and interpreting such Ss proving theta equivalence. Novel here is: (1) Theta equivalence derives from semantic interpretation, not UTAH, which fails. So we come to understand better the empirical and notational commitments of theories which adopt UTAH. (2) Expressions are derived bottom up using only Merge and Voice functions with no A or A= movement. Voice and case morphology are autonomously structural [K&S 2003], not reflexes of movement or add-ons to phrase structure. Support for our analysis is that relative clauses are built just using constituents of nuclear Ss. No movement, empty categories or variable binding are needed. This yields a novel account of the ASubjects Only@ constraint in W. Austronesian. 1. Introduction We first overview Malagasy clause structure, illustrating its voice morphology. Then Part 2 derives and interprets nuclear clauses built from verbs in different voices. Part 3 compares our approach with mainstream ones (GHT 1992, Paul 1999, Richards 2000, R&T 2000, Pearson 2005). Typical of W. Austronesian languages, verbs in Malagasy are clause initial, presenting different morphologies built by affixing roots. For example from the root2 tolotra >offer= we construct the four verbs in (1). (1a-d) have basically the same meaning, so the translation of (1a) suffices for each. The primary constituent break, indicated by square brackets, receives massive support, not repeated here as generative grammarians working on Malagasy (Keenan 1976, 1995; GHT 1992; Paul 1999; R&T 2000; Pearson 2005; Sabel 2005) agree on it, though they differ concerning the category and internal structure of the bracketed constituent. We refer to it theory neutrally as a one place Predicate Phrase (P1). It combines with one DP to form a S (P0), the type of expression which is true or false. Semantically, P1s denote functions, called properties here, mapping entities into {True, False}. (1) a. [P1 m+aN+tolotra (manolotra) vary ny vahiny amin'ny lovia vaovao] aho Pres+AN+offer rice the guest on'the dishes new 1sg.nom I offer rice to the guests on the new dishes b. [P1 a+tolotra+ko (atolotro) ny vahiny amin'ny lovia vaovao] ny vary A-+offer+1sg.gen the guests with'the dishes new the rice c. [P1 tolotra+ina+ko (tolorako) varyt+amin'ny lovia vaovao] ny vahiny offer+INA+1sg.gen rice past+with'the dishes new the guests d. [P1 aN+tolotra+ana+ko (anolorako) vary ny vahiny] ny lovia vaovao [[AN + offer]+-ANA]+1sg.gen rice the guests the dishes new The DP sister3 of the P1 is replaceable by a distinctive shape of pronoun B aho >1 sg=, izy >3sg/pl=, ... which we gloss nom. So nom means replaceable by a pronoun from that series. Pearson (2005) justifies calling them Adefault@ forms. In (1b,c,d) the Agent is glossed gen. It has exactly the same (complex) forms as possessors in nominals. Compare (2) with Agents in the a- prefix verbs in (3). See K&P for details: (2) a. trano house b. ny tranoko the house+my c. ny tranon-dRabe4 the house+of+Rabe (3) a. atao do/done b. ataoko I do/@done my@ c. ataon-dRabe Rabe does/@done of Rabe@ Lastly, pronominal replacements for Themes and Goals in (1a) also have distinctive pronominal forms: ahy >me=, azy >him, her, them=. We call these acc forms. 2. The core voice system in Malagasy We assume a lexicon of roots (Abinal & Malzac, 1963; Rabenilaina 1993). We form nuclear clauses and DPs bottom up by affixing and concatenating lexical items, not by inserting them in prefabricated trees. (This renders difficult detailed comparison of our proposals with more usual P&P or Minimalist ones). 2.1 Primary Affixes apply to roots yielding verbs B syntactically simple Pns (n-place predicates) which carry selection features for the categories and cases of their arguments. We consider only 7 such affixes here. Their diversity supports our approach over fixed tree models. Tense affixes (present = m- /-, past = n(o)-, future = h(o)-) combine with Pns to form tensed Pns, preserving argument structure. 2.1.1 (m)aN- (4b) records the derivational history of (4a) identifying its constituents. (4) a. n+aN+enjika (nanenjika) ny jiolahy Rabe past+AN+chase the thief Rabe Rabe chased the thief b. Spst Spst[DPnom:AG] Spst[DPacc:TH,DPnom:AG] Tpst DPacc DPnom ny jiolahy Rabe S[DPacc:TH,DPnom:AG] RT{AG,TH} n (m)aN- + enjika enjika is a root with two theta roles, Agent (AG) and Theme (TH). We note its category as RT{AG,TH}. (m)aN- is prefixed to enjika, with which the tense marker n- combines to form a tensed P2 of category Spst[DPacc:TH,DPnom:AG]. It selects a DPacc interpreted as Theme to form a tensed P1, which merges with a DPnom interpreted as Agent to form a past tense P0 (Spst ). Prefixing (m)aN is done by a morphological function AN, as in (5)5. (5) enjika: RT{AG,TH} AN ===> -anenjika: S[DPacc:TH, DPnom:AG] We treat roots, and expressions in general, as ordered pairs whose first coordinate is a phonological string, and whose second is a grammatical category. (In English the noun honor and the verb honor are different expressions with the same first coordinate but different category coordinates). We often omit brackets around ordered pairs. Syntax A function Merge concatenates a string s of category Pn = Stns[C1α1:θ1,...,Cnαn:θn] with a t of category C1α1 to form st of category Stns[C2α2:θ2,...,Cnαn:θn], satisfying a category (C1) and a case (α1) requirement. In a variant vocabulary we may say that merging a C1α1 with the Pn checks the C1α1 feature. When all case and category features are checked we have a P0 (0 = no features to check). In (4a), Merge concatenates nanenjika with ny jiolahy >the thief= a DPacc to yield nanenjika ny jiolahy, a Spst[DPnom:AG]. Then that P1 merges with Rabe, a DPnom, to form a P0. Theta role diacritics only appear in the selection feature list of Pn=s. They are used in semantic interpretation. Semantics Given a model with domain E, (enjika, RT{AG,TH}) denotes a binary relation over E (a set of pairs of elements of E). A P1 denotes a property, a function from E into truth values5. In general a Pn+1, an n+1-ary predicate, maps an entity to an n-ary predicate denotation. For simplicity here we use DPs interpretable as elements of E. Extending our treatment to the full class of generalized quantifiers is straightforward. Theta roles are relations between entities in E and n-ary predicate denotations, n > 0. Writing d for the semantic interpretation of an expression d, we define: (6) AN(enjika)(y)(x) = True iff (x,y) enjika THEME(y enjika)) AGENT(x, enjika) We often omit the category coordinate of an expression, writing just enjika instead of <enjika, RT{AG,TH}>. In (6) enjika denotes a binary relation enjika. Some might want to read the right hand side above as Ax and y are participants in a chasing event enjika, y is Theme of enjika and x is Agent@. All we need is compositionality: that the derived expression AN(enjika) be interpreted as a function of what it is derived from, enjika. From (6) it is the P1 AN(enjika), not just the root enjika, which associates argument and theta role. (4a) is True (in E) if and only if the pair <j,r> is in the enjika relation, j = the thief and r = Rabe, and j bears the Theme relation to enjika and r the Agent relation to it. So ny jiolahy >the thief= denotes the Theme of AN(enjika) >chase= and Rabe its Agent in (4a). Thus theta role assignment derives from semantics. And we define: Def 1 A verb is Actor Voice (AV) iff it does not select a DPgen. Pn=s built from an AV verb are also called AV. A non-AV verb is called θ voice if it selects a DPnom:θ. Thus AN(enjika) in (4a) is AV, as is (4a) itself. A verb recall, is a syntactically simple Pn. It consists of a root with zero or more affixes. The result of coordinating a verb with a Pn, modifying it by an adverb or PP, or merging it with an argument expression is not syntactically simple, and so is not a verb. Fact 1 below motivates our explicit (given a lexicon) but narrowly language specific definition of Actor Voice (Schachter 1976). Fact 1 A verb (and its Pn=s) is Actor Voice iff it selects a DPnom whose theta role outranks those of any other DP it selects on the partial order in (7): (7) Agent/Cause/Perceiver > Theme/Goal > Instrument/Benefactive/Locative > Other Fact 1 holds for the range of verbs we consider, but complete support would require treating predicates like mendrika >merits=, as in mendrika azy izy >Henom merits itacc= with He = Goal and it = Theme. Refining Fact 1 leaves Def 1 unchanged. Being Actor Voice is likely a syntactic invariant of Malagasy grammar [K&S]. This property is syntactically significant in that: (1) Only AV verbs are m-initial in present tense; (2) AV verbs form imperatives (K&P, Koopman 2005) by suffixing -a, non-AV verbs suffix -o (= /u/), or -y (= /i/) if the root contains an o; (3) Causative and reciprocal affixes only apply to AV verbs, forming AV verbs, and (4) Acircumstantial@ morphology (below) applies only to AV verbs, deriving non-AV verbs. To complete our analysis of (4a) we stipulate that the domain of the function AN, the set of expressions it applies to, isdef the set of roots marked with man- in Abinal & Malzac 1963. These roots r often have category RT{AG,TH} (but not all such roots are in the domain of AN; see I below). AN maps them to AN(r) of category S[DPacc:TH,DPnom:AG]. Some present tense examples are: AN(didy) = mandidy >cuts=, AN(vaky) = mamaky >reads=, AN(vono) = mamono >kills=, AN(fonos-) = mamono >envelopes=. AN also applies to roots r of category RT{AG,TH,GOAL}, mapping them to AN(r) of category S[DPacc:TH,DPacc:GL,DPnom:AG], as in: AN(roso) = mandroso >serves=, AN(tolotra) = manolotra >offers=, AN(ome) = manome >gives=. AN applies to a few roots of category RT{AG} yielding P1s: AN(leha) = mandeha >goes=, AN(dihy) = mandihy >dances=, AN(lainga) = mandainga >lies=, AN(lohalika) = mandohalika >kneels=. Also AN has some roots of category RT{TH} in its domain: AN(hetaheta) = mangetaheta >is thirsty=, AN(hatsiaka) = mangatsiaka >is cold=, AN(firatra+firatra) = mamirapiratra >scintillates=. (8) is a generalization that aids in language acquisition by providing default interpretations for novel verbs. (8) AN always builds Actor Voice predicates (and so will also be called Actor Voice). To complete defining AN we should say what phonological value it assigns to the string coordinate of a root r. We often just write AN(r), but as in other Austronesian languages (m)aN- prefixation is phonologically complex: vowel initial roots just prefix an-. Root initial voiceless stops usually drop: tolotra >offer= manolotra, voiced continuants usually mutate to the closest prenasalized non-continuant: roso >serve= = mandroso, /ma.ndru.su/, etc. See K&P and Paul (1996). Many voice affixes are determined by functions whose domains overlap but are not identical. That is, the affixes select their roots, no two select exactly the same set, but they often overlap. Thus it is natural to represent (m)aN- affixation by a function AN since to define the domain of AN is to state what roots (m)aN- selects. Similarly for the other verbal affixes. To derive a P1 from a root we must see the affix, not simply know that there is one, as many different affixes combine with the same root. Even given the theta role of the DPnom of the P1 we still cannot uniquely predict the affix (below). 2.1.2 I is a second AV function. It prefixes i- to the string coordinate of a root. It applies to many roots of category RT{AG,TH} building P2s of category S[DPacc:TH, DPnom:AG], like AN: I(kapoka) = mikapoka >beats=, I(vidy) = mividy >buys=, I(orina) = miorina >builds=, I(varotra) = mivarotra >sells=, I(laza) = milaza >says=, I(tady) = mitady >seeks=, I(fidy) = mifidy >chooses=, I(saotra) = misaotra >thanks=, I(karakara) = mikarakara >takes care of=, I(taona) = mitaona >carries=, I(jery) = mijery >watches=, I(antso) = miantso >calls=. None of the roots listed here are also in the domain of AN. Thus if we know of a P1 that it is built from an Agent/Theme root and interprets its DPnom as Agent we cannot uniquely predict the AV prefix: it is aN- with enjika >chase=, but i- with tady >seek=. I also often builds P1s, and this in two ways. It applies to some roots of category RT{AG}: asa ==> miasa >works=, teny ==> miteny >speaks=, homehy ==> mihomehy >laughs=. And it applies to some two place roots which are in the domain of AN: enjika ==> mienjika >runs away=, (manenjika >chases=); sasa ==> misasa >washes (intr)= (manasa >washes (tr)=); hidy ==> mihidy >is locked=, (manidy >locks (tr)=). Another language internal generalization helpful in learning novel verbs is: (9) When i- and aN- combine with the same root the i-verb usually has lesser valence. I- is the most productive AV affix. In distinction to AN, it enables direct recovery of the root. Hilda Koopman (pc) points out nonce creations on the web like miparticiper >participates= built on French with a (m)i- prefix. We turn now to some non-AV affixes. 2.1.3 INA builds non-AV predicates by suffixing roots with -ina (-ena, -(a)na) supporting root final consonants (perhaps modified)2. Suffixing in general shifts stress (phonemic) rightward. We mark it here, but not generally. Contrast (11a) with (4a): énjika: RT{AG,TH} (10) INA ==> enjéhina: S[DPgen:AG, DPnom:TH] (11) a. n+enjika + ina +Rabe (nenjehin-dRabe) ny jiolahy past+chase+INA+Rabe the thief Rabe chased the thief (The thief was chased by Rabe) b. Spst Spst[DPnom:TH] Spst[DPgen:AG,DPnom:TH] DPgen DPnom S[DPgen:AG,DPnom:TH] Tpst RT{AG,TH} n enjika + ina Rabe ny jiolahy Past chase INA Rabe the thief Rabe chased the thief (The thief was chased by Rabe) (12) INA(enjika)(u)(v) = True iff (u,v) enjika THEME(v,enjika) AGENT(u,enjika) So enjehina is a Theme Voice (TV) verb. INA always builds non-AV verbs. Ss like (11a) are often translated as passives in English or French (Rahajarizafy 1962, Keenan 1976, Paul 1999) since, in both, the Theme is an argument of the P1 and the Agent is part of a constituent with the verb excluding the Theme. But in fact TV verbs built from -ina and a- (below) are far more like English actives than passives7. We now define: Def 2 Nuclear Ss are theta equivalent iff their arguments can be bijectively matched such that each argument in one and its match in the other have the same theta roles. Notice now that (13a) follows immediately from (6) and (12), and in general (13b) holds when root is in the domain of both AN and INA. So the derived verbs are logical converses, from which the two propositions below follow. (13) a. AN(enjika)(y)(x) = INA(enjika)(x)(y) and in general, b. AN(root)(y)(x) = INA(root)(x)(y) Proposition 1 For r a two theta role root in the domain of both AN and INA, A. nuclear Ss built from AN(r) and INA(r) are theta equivalent: the P1 argument of each bears the same theta role as the non-P1 argument of the other. B. I(r) and INA(r) build theta equivalent nuclear Ss when I(r) is transitive. Proposition 2 Nuclear Ss built from AN(r) and INA(r) are logically equivalent when their Agents are the same individual denoting DPs and their Themes are the same individual denoting DPs. Comparable claims hold for I(r) and INA(r) in B. Prop-2 provides support not based on theta roles for the correctness of our analysis. Prop-1, extended below, is why UTAH (Uniformity of Theta Role Assignment Hypothesis; Baker 1988) fails on our analysis. UTAH implies that DPs with the same theta role in different expressions are derived from isomorphic initial structures. And our treatment of the voice affixes such as AN/INA and I/INA is that they combine with roots to build syntactically distinct verbs which are semantic converses of each other, whence we can match their argument DPs preserving theta role but violating UTAH. The AN/INA alternation is very productive in Malagasy, many roots lie in the domain of both functions: haja manaja / hajaina >respects=, vono mamono / vonoina >hits, kills=, la manda / lavina >refuses=. Equally the I/INA alternation is productive: bata mibata / bataina >raises up=, kapoka mikapoka / kapohina >beats=, antso miantso / antsoina >calls=. Imagine English with hundreds of pairs of verbs of the form <x+blik, y+blik> which were semantic converses, like please / like. We could then, as in Malagasy, derive nuclear Ss such as Aw likes z@ and Az pleases w@ holding of the same participants but presenting them in different syntactic configurations without ever having been in the same one, violating UTAH, to which we return in 3.2. 2.1.4 A is a second non-AV function. It prefixes a-, and like I and AN, it applies only to roots, so its domain can be listed. The DPnom of a P1 built by A is an AIntermediary@: Theme with roots of movement or transmission, Instrument otherwise. See Paul (1999). Here we just emphasize that A, INA, I, and AN, all select the roots they apply to. For example A applies to tosika below yielding a TV verb, whose first argument is a genitive Agent, always the case with non-AV verbs. (14) a. [n+(m)aN+tosika (nanosika) ny fiara] Rabe past+AN+push the car Rabe Rabe pushed the car (AV) b. [n+a+tosika+Rabe (natosi-dRabe)] ny fiara past+A+push+Rabe.gen the car Rabe pushed the car (TV) tosika is not in the domain of INA: *tosihina. Thus we cannot uniquely predict the voice affix of a P1 built from a Agent/Theme root which interprets its DPnom as Theme. If the root is enjika >chase= we use -ina, if it is tosika >push= we use a-. Other roots in the domain of A but not INA are taov- >do, make= which yields AV manao and TV atao but *taovina; voaka >exit= which yields AV mivoaka >go out= and mamoaka >expell=, as well as TV avoaka, but *voahina. Also lahatra mandahatra and alahatra >put in a line= but *laharina, and tono mitono and atono >grills= but *tonoina. There are many more roots in the domain of INA but not of A: lalao >play= lalaovina but *alalao; la >refuse= lavina but *ala; haja >respect= hajaina but *ahaja. But usually both A and INA apply to ditransitive roots of transmission, as in (1). A assigns Theme to DPnom and INA assigns Goal. Similar roots are roso >serve= and solo >replace, substitute=. This paradigm is not limited to verbs of transmission: (15) a. n+(m)aN+tondro (nanondro) ny tranony tamin=ny tehiny Rabe past+AN+point-out the house-his past+with=the cane-his Rabe Rabe indicated his house with his cane b. n+a+tondro+Rabe (natondron-dRabe) ny tranony ny tehiny past+A+point-out Rabe the house-his the cane-his c. no+tondro+ina+Rabe (notondroin-dRabe) tamin=ny tehiny ny tranony past+point-out+INA+Rabe past+with=the cane-his the house-his Here are the semantic values of AN, A, and INA at ditransitive roots r. (16a) says that the P3 AN(r) interprets its argument z as THEME, the resulting function interprets its argument y as GOAL, and the final P1 function interprets its argument x as AGENT. In contrast the A verb in (16b) interprets its first argument as AGENT, its second as GOAL and its last as THEME. The INA verb in (16c) interprets its first argument as AGENT, its second as THEME, and its last as GOAL. (16) a. AN(p, RT{AG,TH,GOAL})(z)(y)(x) = True iff <z,y,x> P & THEME(z,P) & GOAL(y,P) & AG(x,P) b. A(p, RT{AG,TH,GOAL})(x)(y)(z) = True iff <z,y,x> P & AG(x,P) & GOAL(y,P) & TH(z,P) c. INA(p, RT{AG,TH,GOAL})(x)(z)(y) = True iff <z,y,x> P & AG(x,P) & TH(z,P) & GOAL(y,P) (16b,c), as well as (1b,c) are examples of two non-AV theta equivalent Ss. So theta equivalence doesn=t just apply to pairs we might analogize to active vs passive in English. And we now extend Proposition 1 C. For r in the domain of AN and A, nuclear Ss built from AN(r) and A(r) are theta equivalent, the P1 argument of one bearing the same theta role as that of the interior argument of the other. D. For r a ditransitive root of transmission in the domains of AN, A, and INA the three nuclear Ss they build are theta equivalent, per (16) above. 2.1.5 Roots. Malagasy has about 25 roots which function directly as TV P2s. So in the lexicon they have category S[DPgen:AG,DPnom:TH] and take without affixation a genitive Agent and a nominative Theme. They have AV forms in (m)aN- or i- and often (not always) have INA forms. We illustrate with the root résy marking stress: (17) a. Résin=ny mpanjaka izy ireo defeat+INA=the ruler.gen 3nom dem.pl They are/were defeated by the king b. N+aN+résy (nandrésy) azy ireo ny mpanjaka past+AN+defeat 3acc dem.pl the ruler The king defeated them c. Resy+ina+Imp (= reséna+Imp = Reséo) izy! defeat+INA+Imperative 3nom dem.pl Defeat them! Root predicates in general do not take past tense marking, n(o)-. A lexical adjective like adala >crazy= builds Adala Rabe >Rabe is/was crazy=, but *Nadala Rabe >Rabe was crazy=. Similarly resi- in (17a) cannot prefix no-. These TV roots are stative, whereas their -ina counterparts are dynamic (Paul 1999). The imperative in (17c) is the INA form, as we see from the stress shift right and the change in vowel from i to e. The addressee is missed out as is usual (but not obligatory) in imperatives B it is genitive when present, and the Theme is nominative, as expected. In (17a) we have the stative root, in which the n following resi is a purely orthographic reflex of the genitive construction, Malagasy has no geminates. Thus TV roots like résy and their INA forms, reséna, may both take genitive Agents and nominative Themes yielding theta equivalent Ss, both of which are non-AV. For each line below the three indicative forms build nuclear Ss that are theta equivalent, the first two being TV with genitive Agent/Perceiver and nominative Theme, the last being AV with accusative Theme and nominative Agent/Perceiver. The last two imperative forms likewise build theta equivalent imperatives. (The AV imperative theta equivalent to (17c) is Mandresea azy! >Defeat them (3acc)!=). (18) ROOT résy TV reséna AV mandrésy TV+Imp reséo AV+Imp mandreséa >defeat= héno váky tápaka hadíno henóina vakína tapáhina hadinóina mihéno mamáky manápaka manadíno henóy vakío tapáho hadinóy mihenóa mamakía manapáha manadinóa >hear= >read= >cut= >forget= 2.1.6 VOA is a TV function which prefixes voa- to roots forming Ss theta equivalent to INA ones but which differ in aspect. Compare: (19) a. Voa+kapoka (voakapoka) ilay alika VOA+ beat that (aforementioned) dog That dog was beaten b. no+laza+ina+ko (nolazaiko) izany c. voa+laza+ko (voalazako) izany past+say+INA+1sg.gen that VOA+say+1sg.gen that I said that I said that VOA verbs are perfective, do not form imperatives and, like roots, are not marked for pas tense. They are often agentless, (19a), and occur less frequently than INA or A verbs (Keenan and Manorohanta 2001). So they do resemble English passives. Still, as (19c) shows they can take genitive Agents building Ss theta equivalent to ones built from INA verbs. And it is common, as with both roots in (19), for VOA and INA to apply to the same roots: fitaka voafitaka, fitahina >deceived=, vonjy vaovonjy, vonjena >saved=, soratra vaosoratra, soratana >written=. 2.1.7 - prefixes to about 20 roots, all vowel initial, forming present tense AV verbs in m-. They take normal past and future tense marking, n(o)- and h(o)- replacing m-. They have AV imperative forms and circumstantial forms (below), so we treat them as having a zero prefix (m is the present tense marker for all AV derived verbs). Examples are -ila >need=, -ino >believe=, -indrana >borrow=, -anana >possess= and -iditra >enter=. The first four have -ina forms and thus form theta equivalent Ss. (20) a. m++íla (míla) azy aho Pres+AV+need 3acc 1sg.nom I need her b. Ila+ina+ko (iláiko) izy need+ina+1sg.gen 3nom I need her There are yet other primary affixes B the AV ma-, mana- and maha- (Phillips 2000) and the (usually) TV tafa-, but our partial inventory will suffice here. 2.2 Secondary Affixes. Causative AMP (Randriamasimanana 1986; Andrianierenana 1996) and reciprocal IF (Keenan & Razafimamonjy 2004) prefix (m)amp- and (m)ifrespectively to AV verbs (less the initial m) forming AV verbs that lie in the domain of other voice affixes, hence our interest in them. (21) and (22b) both illustrate both affixes. (21) n+if+amp+i+anatra zavatra betsaka isika past+Rec+Cause+I+learn thing many we.nom.incl We have taught each other many things (22) a. n+if+an+enjika izy ireo past+Rec+AN+chase 3nom dem.pl They chased each other b. n+amp+if+an+enjika azy ireo Rabe past+Cause+Rec+AN+chase 3acc dem.pl Rabe Rabe made them chase each other AMP verbs (but not IF ones) lie in the domain of INA (whose DPnom can be either Theme in the case of causatives of transitive verbs). (23) a. n+amp+aN+taov (nampanao) farafara azy ireo Rabe past+Cause+AN+make bed 3.acc dem.pl Rabe Rabe made them make beds b. n+[amp+aN+taov]+ina+Rabe (nampanaovin-dRabe) farafara izy ireo past+Cause+AN+make+INA+Rabe bed 3.nom dem.pl They were made by Rabe to make beds So the primary voice functions AN, I, and A only apply to roots. But INA applies to roots and to AV verbs derived by AMP. INA also applies to a closed class of six AV verbs derived by AN: they suffix -ina directly to form a TV verb: hataka mangataka (AV) angatahina (TV) >asks=; halatra mangalatra (AV) angalarina (TV) >steals=; voatra mamboatra (AV) amboarina (TV) >arranges=. See RR (1971; 102). (24) n+[[aN+halatra]+ina]+Rabe (nangalarin-dRabe) ny omby past[[AV+steal]]+TV+Rabe.gen the cow The cow was stolen by Rabe / Rabe stole the cow 2.3 Circumstantial Voice (CV) is the last and most problematic voice we consider. CV verbs are built by ANA which suffixes -ana (sometimes -anana) shifting stress right, preserving root final consonants2. ANA applies to all prefixed AV verbs and to no bare roots or non-AV verbs. The derived verb is not AV: it first selects a DPgen, with the theta role of the DPnom of the AV verb it is derived from. Its DPnom usually bears an oblique theta role B Benefactive, Instrument, Location, Time, Manner, Price, ..., typically not selected by the AV verb it applies to (see K&P, and Paul 1999)8. Present tense is - and imperatives are formed with -o (or -y), as expected of non-AV verbs. Compare: (25) a. [n+aN+enjika (nanenjika) ny jiolahy t+amin=ny fiara] Rabe past+AV+chase the thief past+with=the car Rabe Rabe chased the thief by means of the car AV b. [n+enjika+ina+Rabe (nenjehin-dRabe) t+amin=ny fiara] ny jiolahy past+chase+TV+Rabe.gen past+with=the car the thief Rabe chased the thief by means of the car TV c. [n+[aN+enjika]+ana+Rabe (nanenjehan-dRabe) ny jiolahy] ny fiara past+[[AV+chase]+CF]+Rabe.gen the thief the car Rabe chased the thief in/with/...the car CV (26) a. [n+i+vidy (nividy) akanjo ho an=i Bosy] Rasoa past+I+buy clothes for=art Bosy Rasoa Rasoa bought clothes for Bosy AV b. [no+[vidy+ina]+Rasoa (novidin-dRasoa) ho an=i Bosy] ny akanjo past+buy+INA+Rasoa.gen for=art Bosy the clothes Rasoa bought the clothes for Bosy TV c.[n+i+vidy+anana+Rasoa (nividianan-dRasoa) akanjo] i Bosy past+[[I+buy]+ANA]+Rasoa.gen clothes art Bosy Rasoa bought clothes for/because of...Bosy CV (27) is a sample application of ANA to a transitive AV verb On the right in (28) are some present tense CV verbs built from enjika >chase=. ANA (27) verbAV:S[DPacc:TH, DPnom:AG] ==> verb+ana:S[DPgen:AG, DPacc:TH, DPnom] (28) miénjika manénjika mifanénjika mampifanénjika runs away chases chase each other makes chase e.o. ienjéhana anenjéhana ifanenjéhana ampifanenjéhana circumstance of fleeing circ. of chasing circ. of chasing each other circ. of causing to chase e.o. In (27) DPnom is not marked for theta role. The CV form tells us that its DPnom stands in some oblique relation to the action or state expressed by the verb, but just which one is decided pragmatically, not coded in the syntax or morphology. For example, in (29) Rasoa might be Benefactee or might be Cause (she nagged Rabe so much he finally built a house). Other theta roles, such Instrument or Locative are grammatically possible but pragmatically implausible in this case. (29) +[[aN+taov]+ana]+Rabe (anaovan-dRabe) ny trano Rasoa pres+AN+make+ANA+Rabe.gen the house Rasoa Rabe is making the house for / because of Rasoa So a CV sentence is not fully theta equivalent to a corresponding AV or TV one with an overt PP. But the arguments which are selected by the underlying AV verb do bear the same theta roles in both Ss. Here is our analysis: PPs and adverbials denote predicate modifiers B functions F mapping predicate denotations P to predicate denotations F(P). Normally they are restricting: F(P) P, meaning that for all entities x, if F(P)(x) = True then P(x) = True. for Max is restricting in (30a), whence it entails (30b). Replacing for Max by with a screwdriver preserves the entailment since it is also restricting. (30) a. b. Sue opened the box for Max Sue opened the box [(for Max)(open (the box))](Sue) [(open(the box))](Sue) So Prepositions denote functions p mapping entities to restricting functions, and CV verbs quantify over such p as in (31), the circumstantial of a transitive verb. (29) is derived as in (32) and compositionally interpreted as in (33), tense omitted9. (31) ANA(verbAV)(x)(y)(z) = True iff p p(z)(verbAV(y))(x) = True (32) Spres Spres[DPnom] Spres[DPacc, DPnom] Spres[DPgen, DPacc, DPnom] S[DPgen, DPacc, DPnom] S[DPacc,DPnom] DPgen DPacc DPnom Rabe ny trano Rasoa RT{AG,TH} (33) aN taov ana aN taov ana Pres AN TAOV ANA Rabe ny trano Rasoa r t s AN(TAOV) ANA(AN(TAOV)) (ANA(AN(TAOV)))(r) (ANA(AN(TAOV)))(r)(t) (ANA(AN(TAOV)))(r)(t)(s) Crucially the semantic interpretation of the CV verb uses that of the AV verb it takes as argument. The (tenseless) AV verb is a constituent of the CV verb, as made clear in the bracketing in (31). And since p(x) is restricting, a nuclear CV S like (29) entails the AV (34) which lacks a PP corresponding to the P1 argument of the circumstantial. (34) m+aN+taov (manao) ny trano Rabe pres+aN+make the house Rabe Rabe is making a house The last line is True iff for some p, p(s)(AN(TAOV)(t))(r) = True. And since p(s) is restricting, the last line entails AN(TAOV)(t)(r), the interpretation of Rabe is making the house. Thus our semantics for the circumstantial S entails that Rabe is Agent of make and the house is Theme. Consider now the predictions of theta equivalence between a circumstantial S and a corresponding AV one with an overt oblique. (35) a. b. +aN+taov+ana+Rabe (anaovan-dRabe) ny trano Rasoa pres+aN+make+ana+Rabe.gen the house Rasoa Rabe is making the house for / because of Rasoa m+aN+taov (manao) ny trano ho an-dRasoa Rabe pres+aN+make the house for Rasoa Rabe Rabe is making the house for Rasoa (CV) (AV) On our semantics (35b) entails (35a), but the converse fails. (35a) is true in a model in which Rasoa=s nagging prompted Rabe to build a house for sale (not for her). Our analysis also shows that the genitive complement Rabe in (35a), has the same theta role, Agent, as the P1 argument in (35b), and the accusative ny trano >the house= is Theme in both Ss. But while Rasoa is a Benefactive in (35b), forced by the choice of Prep ho an>for=, Rasoa is not assigned a theta role in (35a). So (35a,b) are not fully theta (or logically) equivalent. Our semantics just quantifies over functions from entities to restricting functions, different theta roles for Rasoa correspond to different choices of function. Thus on our analysis a CV S is vague, not ambiguous, regarding the theta role of its DPnom. We just know that it bears some oblique relation to the predicate denotation. The case is analogous to the bridge where the spies meet, which is unspecific but not ambiguous regarding whether they meet under, on, or next to the bridge. In contrast purely syntactic approaches which derive CV forms by incorporating a Prep into the verbal root (GHT 1992, Paul 1999) would seem to predict ambiguity not vagueness for CV Ss according to the Prep incorporated.. 3. Comparison with mainstream analyses Schachter (1976) first pointed out that properties associated with Asubjects@ (anteceding reflexives, extracting in relativization, etc.) were split between the DPnom of AV Ss and the DPgen of non-AV Ss (our terminology) in Philippine languages. GHT (1992) was the first account for these facts in contemporary generative theory, treating Avoice@ morphology as case assigners and deriving AV and TV Ss by movement of different DPs into Spec(IP) in structures like (36) in order to get case, similar to the derivation of passives in English. (36) IP I= I DP VP DP agent V= V -ina Rabe enjika DP theme ny jiolahy e In (36) -ina assigns case to Rabe, the verb enjika >chase= moves to I hosting the suffix -ina, and ny jiolahy >the thief= raises to DP = Spec(IP) to get case. Assumed stipulations here are that Spec(IP) is assigned case and that the Theme sister of V is not. The corresponding AV S differs from (36) in that the V node dominates both the lexical verb and the morpheme aN-, which assigns case to the Theme. I is empty and the Agent Rabe is caseless in situ and moves to Spec(IP) getting case. UTAH is roughly satisfied in that Agent and Theme assignment are effected in comparable positions in the trees underlying AV and TV Ss (but the trees are not isomorphic due to differences in voice annotation, -ina vs aN- which force different movement options). This case/voice approach is extended in Paul 1999 to include a- verbs, and is implicit in R&T 2000. A second generative approach was initiated by Richards 2000 for Tagalog and developed more fully in Pearson 2005 for Malagasy using structures like (37). Here ny jiolahy >the thief= is not raised to get case, rather it is a topic as in V2 languages like Icelandic and is base generated in Spec(TopP), an A= position not an A position. It is coindexed with an empty operator which has moved from an argument position. Voice morphology encodes the abstract case features associated with the movement chain determined by that operator. This operator originates in Spec(VP) where it is assigned a theta role (Theme) by V but not case. The Asp head preceding it has a strong (by stipulation) case feature, -Vn (= -ina) so it moves across that node into Spec(AspP), acquiring accusative case. Then it moves directly to Spec(WhP). Rabe is generated in Spec(vP) where it is assigned the Agent role by little v. The E head to its left has a weak (by stipulation) case assigner -ny, so Rabe remains in situ, where it is assigned case. The verb is built from enjika which head-moves to Asp, hosting the -Vn(a) suffix, then the verb+-Vn(a) moves to v (without effect), then to E, hosting the suffixal linker -ny, then the whole complex, -enjehiny, moves to T where it hosts the tense prefix nB. The underlying structure is given in (37). In the AV version of (37), nanenjika ny jiolahy Rabe >Rabe, he chased the thief=, presents aN- in Asp, stipulated weak, whence ny jiolahy in Spec(VP) does not move and is assigned case (accusative) and theta role (Theme) in situ. In E however we now find the AV morpheme m-, stipulated strong. Spec(vP) is filled with an empty operator which moves to Spec(EP), acquiring nominative case, and then moves directly to Spec(WhP) where it is co-indexed with Rabe, generated in Spec(TopP). (37) TopP Top= DP ny jiolahyi Top WhP Wh= DP Opi Wh TP T n- EP DP E= E -ny (weak) vP DP Rabe v= v AspP DP ti AspP= Asp -Vn (strong) VP DP ti V= V Nenjehin-dRabe ny jiolahy enjika >chase= We turn now to the comparison, limiting ourselves to (1) descriptive adequacy and (2) generality: how insightfully the analyses carry over to other languages. 3.1 Selection In treating voice affixes as functions10 we have committed ourselves to defining their domains. In several cases, AN, I, A, they are just lists of lexical items. In other cases, such as INA the domain is a mix B some lexical items, 6 AN verbs listed, and then the range of AMP (Causative). The domain of AMP itself is basically the range of AN, I and (for verbs considered here). Listing is not a matter of theoretical interest, though the lists support some language internal generalizations of help to the learner. Mainstream approaches fail to provide the means to constrain the distribution of affixes and roots, since they (largely) occur independently in the tree structures. What prevents us on either mainstream approach from inserting -ina in the slot provided for the voice affix (strong in Pearson) and inserting tosika >push= in the root/verb node? They are independent nodes, and each item can independently go where we put it. But then verb movement will derive the phonologically natural *tosihina, which does not exist, since as we have seen tosika, like many other roots, forms its Theme Voice form with a-, not -ina. Dually many roots take -ina and not a-. Nor can we redesign the lexical entries for a- and -ina so that they occur in the same node, forcing them to be in complementary distribution. Then we would not generate the many verbs built from roots that take both affixes (such as most ditransitives). We emphasize that the problems here are not just a few lexical items. Mainstream approaches massively misgenerate B they scramble verb roots and affixes, the core of Malagasy grammar. Here, quickly, are some further cases. 1B the i- and aN- AV prefixes distinguish many two place roots. Many such roots combine with i- and not aN- and many combine with aN- and not i-. But again we can not put them in the same node in the tree since then we fail to capture the many cases of roots that take both, with the i-verb usually being intransitive. 2 B many i-verbs are AV transitive and their roots may take TV -ina. But we listed a fair number that don=t take -ina, such as voaka >go out=, lahatra >align=, etc. To derive the first case i- and -ina must occupy independent nodes. But then what blocks forming -ina verbs from the many roots that take i- but not -ina? Similarly what blocks forming i- verbs from roots that take -ina but not i-? 3 B traditional suppletive voice forms are not naturally handled on mainstream approaches. For example mitondra / entina >carries=. Mitondra is an AV transitive verb built from I, it has AV tense and imperatives and a CV form, but tondra does not build a TV verb B not with a-, not with -ina. Similarly entina behaves like a regular TV form B it forms imperatives in -o not -a, but it has no corresponding AV form. For us we just include tondra in the domain list for I, and ent- in that for INA. And we stipulate in the lexical entries that INA(ent-)(y)(x) = I(tondra)(x)(y). This is correctly ad hoc as there is no way to predict that mitrondra and entina are logical converses. Other suppletive pairs are maka / alaina >takes= and mahalala / fantatra >knows=. 4 B we have seen that there are six AN verbs in the domain of INA (angataka >ask=, etc.). So these verbs will carry both aN- and -ina, a situation incompatible with either mainstream approach. On GHT for example both Agent and Theme would be case marked. With no other DP in the S, as in (24), the result is incorrectly blocked (an EPP violation). If there is an oblique DP in the S it would advance to subject with the CV morphology -ana on the verb. But this is ungrammatical, -ina verbs cannot suffix -ana. Again we just list these six aN- verbs in the domain of INA. They are exceptional in traditional grammars and learning that they take -ina must be done ad hocly. 5 B case assigning by voice affixes is not sufficiently general internal to Malagasy. Recall the 25 odd roots that function directly as TV transitive verbs: resy >defeat=, hadino >forget=, etc. So in Ss like (17a) Résin=ny mpanjaka izy ireo >They are defeated by the king= we have two case marked DPs, one genitive, one nominative, assigned Agent and Theme respectively. We list these roots in the lexicon with category S[DPgen:AG,DPnom:TH]. But on mainstream approaches there are no affixes to assign case. Also case assignment by voice affix does not extend naturally to case assigners like Adjectives and Prepositions. Arguably there is a very general mechanism (Ntelitheos 2005), represented by a nasal segment (na in GHT, ny in Pearson) which constructs genitive case complements. In (2) we saw Afree@ genitives on Ns like >house= tranondRabe >Rabe=s house=. We find such genitives on some adjectives, as in (38). (38) a. jamba (adj) >blind= b. marary (adj) >sick= a=. jamban=ny vola >blinded by money= b=. mararin=ny tazomoka >sick from malaria= But a few others select accusative: antonina >sufficient=, feno >full=, sahaza >suitable=. (39) a. Antonina azy io satroka io sufficient him that hat that That hat suits him b. Feno azy ny siny full 3acc the jug The jug is full of it Prepositions normally select genitive complements, as in (40a,b,d), but a few, as in (41a,b) select accusative. And one, noho >compared to, because of= selects nominative: noho izy (nom) >compared to him, because of him=. (40) a. amy (prep) >with,...= a= amiko >with me=, amin-dRabe >with Rabe= b. ambony (prep) >on, above= c. akaiky (prep) >near (to)= (41) a. Lavitra ahy izy io Far 1sg.acc 3nom dem.sg That is far from me b=. ambonin=ny latabatra >on the table= c= akaikin-dRabe >near Rabe= b. Nividy akanjo ho azy aho bought clothes for 3acc 1sg.nom I bought clothes for him 6 B mainstream approaches have difficulty adjusting the number of case assigners to the arity of the verb. For ditransitives such as manolotra >offers= the verb only supplies one case assigner, aN-, and the highest DP gets case by stipulation. But how does the third argument get case? Dually with intransitive AV verbs built with aN- or I- we have only one, nominative, argument but two case assigning options: the one assigned by aN(or i-) and the free nominative. Some of these verbs may be treated as unaccusatives, like mangatsiaka >is cold= or manjombona >is overcast= but then we expect aN- to assign accusative case to the underlying object, yet it only surfaces as nominative. 3.2 UTAH Our discussion of the affixes in 2 and of selection in 3.1 has revealed many pairs or triples of Ss which are theta equivalent without satisfying UTAH on our analysis. Both mainstream approaches essentially assume UTAH, which is in part why they need so much movement of DPs, as theta equivalent Ss with the DPs in different orders forces something to move. Our intent here has not been to show that UTAH is empirically incorrect. Several scholars have argued this for one or another language (Spencer 1995, van Valin 1992, and Anderson 2000). Rather our concern has been to provide a syntactic and semantic analysis of the voice morphology of Malagasy which yields correct judgments of well formedness and entailment. The judgments of theta equivalence that would be yielded by UTAH are corollaries to our semantics, which is more general, yielding judgments of logical equivalence and of proper entailment in cases in which logical equivalence fails (as in our discussion of Ss built from circumstantial verbs). We should want to see what a grammar for simple clauses would look like without UTAH even if just to better understand what it rules out. Our need is stronger if we accept the arguments against UTAH by the scholars noted above. Culicover and Jackendoff (2005:73 B 79) provide an enlightening discussion of the consequences of adopting UTAH: small clauses, as in [I find [John reliable]], the presence of PRO in control structures, and VP internal subjects, to name just a few. As each of these expression types in Malagasy merits an extensive investigation we shall not attempt to cover them here. We note only that (1), lacking UTAH we are less committed initially to their analysis, and (2), in 4[2] we illustrate some Malagasy control structures. Their physiognomy appears rather different, indeed much richer, than in English. 3.3 Generality Mainstream approaches have been actively concerned with structural similarities between Malagasy and other languages. And the parallels they have noted are enlightening. There are similarities between the Malagasy voice system and those in other W. Austronesian languages, and there are (weaker) similarities between Malagasy voice and European active/passive distinctions. Equally Richards and Pearson exhibit substantive similarities between our DPnom=s and topic DPs in V2 languages. In contrast our work has solely concerned Malagasy. Have we indeed said anything that generalizes to other languages? To this we have two responses. One, before we can offer generalizations concerning the independent syntactic role of morphology we need thorough grammatical studies on which to base them. Are the syntactic functions of Malagasy voice morphemes natural, or aberrant? Are they dual to nominal case marking patterns? Just how can the two be related by applicative affixes? For historical reasons generative grammar has focused on hierarchical structure. Morphological structure often takes the form of annotations on independently existing phrase structure trees. These trees are largely unchanged if the annotations are erased, as they are largely redundant with phrase structure. In contrast K&S exhibit a model case marking language where case markers are structural invariants independent of hierarchical structure: erase case, lose structure. Second, there are other languages with Malagasy-like patterns. We find similar (but not identical, Pearson 2005) ones in related Austronesian languages1. Outside this family the most obvious case is Semitic, in which active, passive, causative, ... verb forms are derived by applying morphological functions to consonantal roots. The active / passive contrast in (42) from the Hebrew <G,D,L> root is illustrative. (42) a. Ha-saba gidel et ha-yeled the-grandfather.nom bring+up[act, past] acc.def the-child The grandfather brought up the child b. Ha-yeled gudal al yedei ha-saba the-child.nom bring+up [pass, past] on hands the-grandfather The child was brought up by the grandfather So analogous to AN and INA the I-E function and the U-A function in Hebrew would take roots like <G,D,L> to verbs specifying the number, case and theta role of their arguments. The pattern in which different voice forms derive from a common root is more widespread than we might expect. Compare the active/passive forms in Latin: (43) Active amo amamus amas amatis amat amant Passive amor amamur amaris amamini amatur amantur We do not derive passive verbs by affixing active ones or vice versa. Amamini is not an affixed form of amatis or vice versa. Rather we should start with roots, like amV and apply active and passive morphology and agreements independently, as do classical grammars. (Haspelmath 1990 surveys 80 languages, finding passives in somewhat more than half. Just 3 had the Aux + participle with adjectival type feature agreement (number, gender, not person) were all Indo-European.. Another easy instance of the Malagasy pattern concerns case assignment. That the lexical verb with voice morphology assigns case to its complement is unexceptional. In languages with rich case marking such as Latin, active verbs like amare >to love= and vincere >to conquer= take accusative case objects while parere >to obey=, imperare >to order=, invidere >to envy=, respondere >to answer= and persuadere >to persuade= and others take dative complements. Passives of both classes take agent phrases in the ablative (with a/ab); in the first class the passive Theme is nominative, and in the second it stays dative (usually). So choice of case needs both root and voice morphology. Also convincing here are the many deponent verbs in Latin and Greek which are formed with passive morphology but are transitive taking a nominative and a nonnominative argument. Lat: comitari >to accompany=, tueri >to protect=, aggredi >to attack=, comperiri >to find out=. Gildersleeve & Lodge (1913) list 81 examples. Similarly Modern Greek has many such deponents (Dimitris Ntelitheos, pc): dehome, thimame, mahome (I) accept, remember, fight, as in (44). So the case of the complement depends on both the root and the voice morphology. (44) O Giannis dehtike tin protasi mu the Giannis.nom [accept+NonAct.pst.3sg] the proposal.acc my Giannis accepted my proposal Also similar to Malagasy is (as is well known) that Preps select case in Latin (Russian, German): a/ab >away from, by=, e/ex >out of=, cum >with= take ablative complements, ad >to=, contra >against=, per >through= take accusative. Similarly some adjectives govern case: ones with meanings like suitable, similar, friendly take dative: amicus Ciceroni >friendly to Cicero=. Dignus and indignus >(un)worthy= take the ablative >vir patre ... dignissimus >a man most worthy of his father=. 3.4 Constituency Both mainstream approaches do not in my judgment adequately represent the very tight constituency between non-AV verbs and their genitive argument. The Agent is not a sister to the verb (even after root raising) but rather a Spec of a sister to the verb. But Spec(XP) do not project their features to their XP and so would seem inaccessible from outside it. On our analysis however the DPgen is selected by the verb (so this is not a free genitive, as with Ns and some As) and forms a very tight constituent with it (Keenan 2000). In distinction to AV verbs which allow adverbs to separate them from a definite accusative (as R&T note), genitives are inseparable from their non-AV verbs. One slightly surprising case is illustrated in (46b). Observe first that indefinite Themes appear at times incorporated into the AV transitive verb: (45) a. Manana vola izy M--has money 3nom b. Manam-bola izy M--has-money 3nom He has money He has money / is wealthy (46) a. [Hadinon-dRabe] ny anaran=ny mpianatra forget (TV RT)+Rabe.gen the name=the student.gen Rabe forgot the name of the student b. [Hadinon-dRabe anarana] izy forget (TV RT)+Rabe.gen name 3nom Rabe name-forgot him / He had his name forgotten by Rabe But bare Ns like anarana >name= cannot Aincorporate@ over a genitive agent (Keenan & Ralalaoherivony 2000). We=d derive Hadinon=anaran-dRabe moving anarana >name= over Rabe in (46b). This at most means Aforgotten by the name of Rabe@, a nonsense. A second somewhat surprising constituency judgment concerns coordinating transitive AV verbs with agented CV forms of transitive roots. On our analysis the CV nividianan-dRabe >buy+Rabe.gen= has the same category as AV namaky >reads=, as both seek an DPacc and then a DPnom to form an S. And they do coordinate (Keenan 2000): (47) [N+[[i+vidy]+anan]-dRabe (nividianan-dRabe) sy namaky] ilay boky ianao past+AV+buy+CV+Rabe.gen and past+AV+buy that book 2sg.nom You were bought for by Rabe and read that book 3.5 Semantic interpretation Neither mainstream approach provides a compositional interpretation of nuclear Ss. Yet this is the most explicit account of how we understand arbitrary many novel utterances: we know what their lexical items mean and we learn how things built in that way take their meaning as a function of the meanings of what they are built from. But I don=t see how the syntactic derivations in mainstream analyses could be compositionally interpreted. The plethora of empty nodes and constituent changing movements are daunting. Plausibly such approaches assume translation into ALF@, but to the extent that such translation changes structure (ignoring or collapsing empty nodes, changing c-command relations) it is unsatisfying. Why do we speak the way we do, if we need to change it in order to semantically interpret it? Mainstream approaches however do make more of an attempt to capture theta equivalence. GHT annotate their structures with theta role diacritics and explicitly espouse UTAH. And while the tree structures from which AV and TV Ss derive are not quite isomorphic they are close. GHT could just stipulate that independent of diacritics a transitive V assigns Theme to its sister and Agent to its Spec. Similarly Pearson=s more complex structures do not directly satisfy UTAH. The Agent for example in a AV S originates in the TopP superstorey and is coindexed with an empty operator originating in Spec(vP). In the TV S the Agent originates and stays in Spec(vP). So an explicit coindexing mechanism must be provided, and the semantic effect of co-indexing must be given explicitly so that the appropriate judgments of theta properties can be inferred. Again, it seems to me that it could be done without otherwise changing the analysis. It is also worth noting that UTAH is not quite satisfied even in BJR 1989. The position from which -en argument morphology moves is not identical to that in which full DP Agents are present in active Ss. But again the two structures are Aclose@. And if such highly similar structures underlie both actives and passives in English we will have to move things since at observable structure their DPs vary in order. But as we have shown, it is not necessary to assume isomorphic underlying structures to derive theta equivalence. An explicit semantics, which we want independently, suffices, and is simpler, eliminating much artifactual movement. It is worth recalling that deriving semantic equivalence from syntactically distinct sources is unavoidable in standard semantics. Consider sentential logic B some atomic Ss closed under combinations with and, or, not, if-then .... Now each formula is logically equivalent to infinitely many non-isomorphic others. For example not(P or Q) is equivalent to (notP and notQ), etc. So the idea that syntactically non-isomorphic Malagasy Ss might be theta equivalent B a semantic relation B is semantically natural. 3.6 ASubjects only@ It is well known that only DP sisters of P1s can be relativized and wh-questioned in W.Austronesian. To refer to the thief that Rabe chased, the verb chase must be in the TV form; in the man who chased the thief it must be AV, and in the car in which he chased the thief it must be CV. Both mainstream approaches capture this in a more general way than the AOnly subjects relativize@ constraint of Keenan 1972. R&T suggest a Kaynian ASVO@ underlying structure in which the Malagasy VOS order is derived by fronting the V+O constituent. Then a general constraint blocking extraction from moved constituents leaves the subject as the only constituent available for extraction. R&T refer to an earlier proposal along these lines by Pensalfini; Aldridge (2002) suggests the same idea for Seediq (Formosan). Specifically, R&T assume a structure in which a Speech Act Phrase (headed by question particles for example) dominates an AgrSP whose specifier is NPsubj. The complement of the (undefined) head of AgrSP is TP which includes the verb and all its arguments. After one moves to NPsubj the TP moves left over Speech Act P, and once moved no longer hosts extraction (a stipulation). Appealing to an independent constraint on extraction is a plus, but much remains to be worked out to make this proposal explicit. What motivates movement of TP? To what position does it move? Is the movement obligatory? What blocks extraction without moving? That is, why can=t we extract directly from the original SVO structure? Once we raise a DP out of TP into Spec(AgrSP), does it not violate the intuition of cyclicity to go back down to TP to move it over that DP? Why may the antecedent of an anaphor in an AV S not c-command it? ([Manaja tena] Rabe >Rabe respects himself=has [manaja tena] asymmetrically ccommanding Rabe). Can Rabe bind its original trace without c-commanding it? These questions may well have satisfying answers but R&T do not provide them. As yet I am not persuaded by this movement approach for two reasons. First, the movement of TP (or V+O) is an artifact of having chosen an empirically incorrect order to begin with. On the approach proposed here no movement is involved in deriving simple Ss. And second, Aextraction@ as classically understood introduces complexities motivated in English but not in Malagasy. Below we provide a simpler and more natural analysis of relativization in Malagasy that involves no movement or variable binding. But first consider Asubjects only@ from the V2 topic perspective of Pearson and Richards. Both note similarities in specificity requirements on DPnom and Germanic Topics. Pearson finds similarities in weak crossover. But most striking is that we cannot wh-question out of X in a Topic+V2+X sentence in German, (48c), analogous to (unpermitted) extraction out of P1s in Malagasy. (48) a. Sie hat dem Hans das Buch gegeben She.nom has to.the Hans the book given She gave the book to Hans (Pragmatically neutral order) b. Dem Hans hat sie das Buch gegeben the.dat Hans has she.nom the book given To Hans she gave the book c. *Was dem Hans hat sie gegeben? What to.the Hans has she given What did she give to Hans? (V2, dative DP topicalized) c=. Was hat sie dem Hans gegeben?) What has she to.the Hans given What did she give to Hans? The mechanism Pearson uses to block questioning a non-Topic in Malagasy is that a question word, such as Iza >Who?=, must be co-indexed with an empty operator in Spec(WhP) B review (37). But that is the location of the empty operator co-indexed with the DPnom in Spec(TopP). (Iza is not in Spec(TopP)). The stipulation that no position can host more than one operator blocks wh-questioning anything other than DPnom. (49) a. Iza no nenjehin-dRabe? Who Focus chased+TV+Rabe.gen Who (was it who) Rabe chased? b. Iza no nanenjika azy? Who Focus AV+chased 3sg.acc Who (was it who) chased him? Pearson=s *Operator Cohabitation is an original account of Subjects Only, but it seems arbitrary. Why couldn=t empty operators just pile up, form a set, and get coindexed as needed? Perhaps we could restate Pearson=s idea to say that Spec(WhP) is filled with a variable, bindable by DPnom in declaratives and a wh operator in questions. Then invoke *Double Bind: No two operators can bind the same occurrence of a variable. In logic xxPx is equivalent to xPx. Variable binding operators, x, cannot bind already bound variables. So we have an independent instance of *Double Bind. The parallel between Malagasy DPnoms and German Topics is genuinely intriguing. But there are enough differences to support that the two phenomena are genuinely different, albeit similar. Here are some differences: (1) German fronted Topics retain the case they have in the pre-topicalized structure, there being no analogue of a default case there. (2) V2-Topics with different theta roles do not correlate with verbs in differing morphology. Had we fronted das Buch instead of dem Hans from (48a) the shape of the auxiliary and past participle would not change. (3) At least in German, Topic Fronting as in (48b) is largely a root clause phenomenon (Daniel Büring, pc), such Ss often being ungrammatical in subordinate contexts, (50b), while their Malagasy counterparts, (50c) are fully natural. (50) a. ...weil die Studenten das Buch nicht gelesen haben because the students that book not read have because the students didn=t read that book b. ... *weil das Buch haben die Studenten nicht gelesen because the book have the students not read c. ... satria tsy novakin=ny mpianatra io boky io ... because not read+TF=the students.gen that book that ... because that book wasn=t read by the students Lastly, as Pearson (2000: footnote 28) notes, the core extraction facts are not fully parallel. In (51a) we seem to have wh-questioned into a topicalized clause, retaining V2. This suggests that the star in (48c) results less from competition between topic and extractee than simple violation of V2 B two constituents precede the tensed verb. (51) a. Was hat dem Hans jemand gegeben? What has to.the Hans someone given What did someone give to Hans? b. das Buch das dem Hans jemand gegeben hat the book that to.the Hans someone given has the book that to Hans someone gave (51b) seems to relativize out of a topicalized clause, which forces an Aux final order. Pearson suggests that the fronted dative dem I.Hans in the subordinate clause may be an instance of Scrambling, but even so it has a topicalizing effect, blemishing the parallel with extraction from P1s in Malagasy. We turn now to the natural non-extraction way of forming relative clauses in Malagasy. 3.7 Relative Clauses (RCs) in Malagasy are formed just by concatenating a noun with a P1 in any voice (the result combining freely with Determiners to form DPs): (52) N = N + P1. (53) a. [Nzaza [P1 nanenjika ny jiolahy] ] child (who) [chased the thief] b. [Nzaza [P1 nenjehin-dRabe]] child [chased-by-Rabe] P1s denote properties, functions from individuals to truth values. The property denoted by nanenjika ny jiolahy maps an individual x to True iff x chased the thief. That denoted by nenjehin-dRabe maps x to True iff Rabe chased x. In RCs P1s function as property denoting expressions exactly as they do in nuclear Ss. In (53a) the children referred to are those that nanenjika ny jiolahy >chased the thief= is True of. In (53b) it is those nenjehin-dRabe >was chased by Rabe= maps to True. Formally, (54) a. Noun + P1 = Noun {x| P1(x) = True} b. zaza nenjehin-dRabe = zaza {x|nenjehin-dRabe(x) = True} = CHILD {x|(CHASED BY RABE)(x) = True}. Thus we compositionally interpret (53b)11. No movement or variable binding is needed, we just concatenate Ns and P1s, both used independently in nuclear clauses. We emphasize as well that our analysis captures judgments of selection restrictions. #Nanenjika azy ny trano >The house chased him= is bizarre in Malagasy as is its English translation, since houses can=t autonomously move and so can=t chase things. And we interpret the RC trano nanenjika azy >house (that) chased him= as the set of houses x such that nanenjika azy x >x chased him= is true. So anything in the denotation of the RC must be a house that did some chasing, violating the selectional restrictions. So on (54) RC Formation in Malagasy involves no extraction B no movement, no empty categories and no variable binding. Linguists (myself included, 1972) have ethnocentrically modeled Malagasy RC formation and interpretation on the English pattern, using structures like (55) in which the post-N expression is a wh operator followed by a full S, the whole thing interpreted as a P1 when wh binds its trace in the S. (55) zaza [whi [S[P1 nanenjika ny jiolahy] [ti]]]. child who chased the thief (55) is compositionally interpretable but carries superfluous, hence objectionable, structure: an embedded S, an inaudible P1 argument and a silent variable binding operator (vbo), whi which converts the S (P0) back to a P1. It obscures the fact that to form and interpret RCs, the Malagasy child just uses Ns and P1s already learned in interpreting nuclear Ss. What would motivate a child to posit and interpret additional, unpronounced, material? English, I claim, provides an answer to this question, Malagasy does not. To see this consider the interpretation of the RCs in (56). (56) a. child that chased John zaza nanenjika Rajaona b. child John chased c. rice Rasoa offered the guests d. child Rasoa bought clothes for zaza nenjehen-dRajaona vary natolo-dRasoa ny vahiny zaza nividianan-dRasoa akanjo In each of the Malagasy RCs above the string following the head N, zaza >child= or vary >rice=, is a P1 interpreted as a property, exactly as in a nuclear S. (If it had combined with a full DP on its right the whole thing would have been an S and interpreted as one). So the constituents of RCs in Malagasy are all expressions that occur with the same meaning in simple Ss12 and DPs. There is no need to invoke new objects such as empty categories (ec=s) and vbo=s. We might stress that in the English column in (56) we are not claiming that all strings following the head N are triggers for ec=s and vbo=s. That in (56a) is not. But that in the town where Max was captured it is, since we must force an apparent complete S, Max was captured to be a predicate of places. The English expressions on the left in (56) do yield triggers for these new objects. Namely in all cases except (56a) the learner must interpret strings that do not occur as (interpreted) constituents in simple Ss. For example in Rasoa offered the guests some rice the string Rasoa offered the guests is not assigned a meaning. Invoking an ec (or resumptive pronoun) and a vbo is a simple way to convert these strings to semantic P1s, familiar from simple Ss. We may anticipate an objection to our analysis of RCs on the grounds that internal to Malagasy it is insufficiently general. How will we handle other Aextraction@ structures such as wh-questions and clefts (from which wh-questions are formed in Malagasy)? (57) a. Rabe no nanao izany Rabe Foc didAV that It was Rabe who did that b. Iza no nanao izany? Who Foc didAV that? Who (was it who) did that? There are many differences between wh-Qs/clefts and RCs in Malagasy. For example the interrogative words such as iza >Who?=, inona >What?=, etc. are not used as relative pronouns. And importantly, the Focus particle no, is specific to cleft constructions. On the other hand (with some systematic exceptions) the string following no is a P1 (any voice). The clefted constituent typically binds the argument of the post-no P1. So (57a) above entails nanao izany Rabe >Rabe did that=, but it differs from it in presupposing that someone did it. So the initial DP can be negated: Tsy Rabe no nanao izany >It wasn=t Rabe who did that=. The core paradigm for clefts is: (58) DP no P1 So, except for the Focus particle no, the constituents of core clefts are used in interpreting nuclear Ss. So clefts don=t force extraction either, though they present complications not present in RCs. Here is one illustration. (59a) is well formed with a CV verb as expected, but (59b) is also fine with an AV verb and its nominative Agent fronted.. (59) a. Aiza no nipetrahan-dRabe? Where Foc past+[[I sit] ANA]+Rabe.gen Where does Rabe live? b. Aiza Rabe no mipetraka? Where Rabe.nom Foc pres+I+sit Where does Rabe live? A full discussion of clefts in Malagasy is (well) beyond the scope of this paper. 4. Conclusion: A question of point of view The most striking properties of basic clause structure in Malagasy concern the construction of complex verbs. Most Ss present these structures, though ones with NP, AP and PP predicates may not. Even here many types of nominals are built from predicates and they inherit without change the case and theta role selection features of that predicate (Ntelitheos 2005). For example, (60) ny tsy f+i+tiav+ana+Rabe (fitiavan-dRabe) azy the not nom+[[AV+love]+CV]+Rabe.gen 3acc Rabe=s not loving her So the perspective I would like to close with is: To discover the grammar of Malagasy seek out the ways it builds complex predicates. Below we offer a few suggestions concerning where to look. But first, contrast this perspective with what we may call the dominant theme of the generative grammar of English: namely, the nature of and constraints on movement rules, specifically movement of nominal expressions. This work has resulted in many empirically non-trivial syntactic constraints on the formation of RCs for example. These constraints hold vacuously in Malagasy. That is, expressions that would be ruled out by these constraints are not generated for independent reasons (not because we have an overt constraint blocking them). For example, No Vacuous Quantification is satisfied in (61) since what follows the first N is not a P1. Analogously for the Coordinate Structure Constraint, (62b), and Subjacency, (63b). (61) *ny trano nenjehin-dRabe ny jiolahy the house chased-by-Rabe the thief The house that Rabe chased the thief (62) a. milalao ao an-trano ny zazalahy sy ny zazavavy play there in-house the boys and the girls The boys and the girls are playing in the house b. *ny zazavavy milalao ao an-trano ny zazalahy sy the girls (that) play there in-house the boys and (63) a. nandositra ny olona nangalatra ny omby ran-away the person stole+AV the cow the person who stole the cow ran away b. *hitako ny omby nandositra ny olona nangalatra I+see the cow (that) ran-away the person (who) stole So Malagasy satisfies these constraints, but stating them in its grammar is redundant. But they are not redundant in a grammar of English. If we removed them the grammar would badly overgenerate. So we support a more general view than usual of Auniversal constraints@. They are properties shared by all highly valued grammars for natural languages. Typically they are merely entailed by the independent definition of the generative mechanisms in of any given language. Usually they are not a functional part of its grammar since to be universal they must pick the lowest common denominator cross linguistically and so, as in Malagasy, be entailed by language specific patterns. We conclude with four types of complex predicates whose formation involves crossverb dependencies (see Polinsky and Potsdam 2005). They are intended both to indicate topics of further research and to enrich the inventory of P1s, the better to see just what the class of relative clauses (wh-questions, etc.) is. [1] Pn = Vmotion+Pn. In (64a) a verb of motion combines with a transitive I-verb >to study= forming, we claim, a P2. The past tense motion verb governs a future/irrealis hon >study=, as that action is future relative to the motion. However the locative complement of >went to study foreign languages= is marked with a past tense deictic tany >past+fairly-far+not visible= (present non-visible = any, present visible = eny, etc.) as the complex verb niakatra hianatra >went up to study= is past. (64b) illustrates that this complex V has a non-AV form, one that is forced when we relativize the Theme >foreign languages=. It also shows the difficulty in reading non-AV verbs as passive. (64) a. b. [niakatra hianatra teny vahiny tany Antananarivo] i Soa past+I+go+up fut+I+study lg foreign past+loc Antananarivo art Soa Soa went up to study foreign languages in Antananarivo ny teny vahiny [niakaran=i Soa hianarana tany Antananarivo] the lg foreign past+[I+go-up]+CV+art Soa.gen fut+[I+study]+CV there Ant. the foreign languages Soa went up to Antananarivo to study *the foreign languages that were gone up by Soa to Antananarivo to be studied [2] Pn = Vaspect/desire/intent + Pn with the verbs both AV or both non-AV. (65) a. b. [n+i+kasa h+aN+vaky (nikasa hamaky) io boky io] Rabe past+I+intend fut+AN+read that book that Rabe Rabe intended to read that book [no+kasa+ina-dRabe (nokasain-dRabe) ho vaky+ina (vakina)] io boky io past+intend+INA+Rabe.gen fut read+INA Rabe intended to read that book *Rabe intended that that book get read c. that book that ny boky nokasain-dRabe ho vakina the book that Rabe intended to read * the book that was intended by Rabe to be read (by someone) Both the AV and TV forms of intend select a future (irrealis) complement verb with the Agent understood to be Rabe in both cases [3] Raising to Object, hosted by over 50 verbs (Paul and Rabaovololona 1998). (66) a. manantena ianao fa n+aN+taov+ana-Rabe (nanaovan-dRabe) trano Rasoa pres+AN+hope 2sg.nom that past+[AV+do]+CV+Rabe.gen house Rasoa You hope that Rabe made a house for Rasoa b. [manantena an-dRasoa ho nanaovan-dRabe trano] ianao (RTO) pres+AN+hope acc-Rasoa as past+AN+do+ANA-Rabe.gen house you.nom You hope Rabe made a house for Rasoa c. [+antena+ina+nao (antenainao) ho nanaovan-dRabe trano] Rasoa pres+hope+INA+2sg.gen as made+ANA-by Rabe.gen house Rasoa You hope Rabe made a house for Rasoa d. ny vehivavy [antenainao ho nanaovan-dRabe trano] the woman hoped+INA+2sg.gen as past+AN+build+ANA-Rabe.gen house the woman you hoped Rabe built a house for ??the woman who is hoped by you to have been built a house for by Rabe The bracketed constituent in (66c) satisfies criteria for being a P1. For example the yes/no question particle ve occurs between it and the following DP, the whole thing meaning AIs Rasoa hoped by you to have been made a house for by Rabe?@. The use of that P1 in (66d) is perhaps slightly more natural than in (66c). And a reviewer points out that if we just put the matrix verb in (66a) in the -ina form, as in (66b), but keep the complement clause as is we obtain (67a) in which the complement clause is the P1 argument. However we can relativize on the final DP position, yielding (67b). (67) a. antenainao fa nanaovan-dRabe trano Rasoa hope+INA+2sg.gen that made+ANA+Rabe.gen house Rasoa You hope that Rabe made a house for Rasoa b. ny vehivavy antenainao fa nanaovan-dRabe trano The woman you hope that Rabe made a house for This differs from (66d) just by fa vs ho, which my consultants generally prefer. However (67b) is still good enough that we should generate it. And if our analysis so far is to apply the substring of (67a) consisting of everything except the last DP Rasoa should be analyzable as a P1. Curiously it satisfies our criterion, hosting ve. (68) antenainao fa nanaovan-dRabe trano ve Rasoa? Do you hope that Rasoa was made a house for by Rabe? Complex predicates like those in (66c,d) are the major mechanism for relativizing (questioning, etc.) arguments of complement clause verbs. Verb voice is used crucially here, as the Araised@ object, an-dRasoa in (66b) must satisfy the theta requirements and selectional restrictions of the tensed P1 following ho. Virtually all verbs that take complement clause objects permit Raising to Object, as in (66b). See Sabel (2005). [4] In (69a) we see that verbs like >help= which take complement verbs don=t impose Agent identity on the two verbs. And it is unnatural to force it with a reflexive object, but it can quite naturally be done using a reciprocal verb, (69b). (69) a. [n+aN+ampy (nanampy) azyj h+i+tsara ny fanadinana] isikai past+AV+help 3acc fut+AV+judge the exams we.incl We helped them grade the exams b [n+if+aN+ampy h+i+tsara ny fanadinana] isika past+Rec+AV+help fut+AV+judge the exams we.incl We helped each other grade the exams c. [nifanampian+tsika hitsarana] ny fanadinana past+[Rec+[AV+help]+CV]+1.gen.pl.incl fut+[[I+judge]+CV] the exams We helped each other grade the exams d. ny fanadinana [nifanampian-tsika hitsarana] the exams past+[Rec+[AV+help]+CV]+1.gen.pl.incl fut+[[I+judge]+CV] the exams we helped each other grade *the exams that were reciprocally helped by us to be graded In sum: What W. Austronesian languages are good at is building predicates. DP positioning is derivative. References Abinal, R.P. et V. Malzac (1963) Dictionnaire Malgache-Français. Mission Catholique de Tananarive. Paris: Editions Maritimes et d=Outre-Mer (reprint from 1888 edition). Aldridge, Edith. 2002. Nominalization and WH-movement in Seediq and Tagalog. Language and Linguistics 3.2 Special Issue: Nominalization in Formosan Languages, Elizabeth Zeitoun (ed). Academia Sinica. Pp. 393 B 427. Anderson, S. 2000. In Lexicon in Focus B. Stiebels and D. Wunderlich (eds) Adad. Verlag. Pp. 123 B 143. Andrianierenana, L. 1996. Morphological causatives in Malagasy. In The Structure of Malagasy M Pearson and I. Paul (eds). UCLA Occasional Papers in Linguistics 17. Pp. 58 B 76. Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. University of Chicago Press. [BJR] Baker, M., K. Johnson, and I. Roberts.1989. Passive arguments raised. Linguistic Inquiry 20:219 B 251. Culicover, P.W. and R. Jackendoff. 2005. Simpler Syntax Oxford Erwin, Sean. 1996. Quantity and moras: An amicable separation. In The Structure of Malagasy vol I. M. Pearson and I. Paul (eds). Dept. of Linguistics, UCLA. pp. 2 S31. Guildersleeve, B.L. and G. Lodge. 1913. Gildersleeve=s Latin Grammar Macmillan & Co. Ltd. London. [GHT] Guilfoyle, Eithne, Henrietta Hung and Lisa Travis. 1992. Spec of IP and Spec of VP: two subjects in Austronesian languages. Natural Language and Linguistic Theory 10:375 S 414. Haspelmath, M. 1990. The grammaticization of passive morphology. In Studies in Language 14.1:25 B 72. Huang, Lillian. 2002. Nominalization in Mayrinax Atayal. Language and Linguistics 3.2 Special Issue: Nominalization in Formosan Languages, Elizabeth Zeitoun (ed). Academia Sinica. pp 197 B 227. Hyams, Nina, Cecile Manorohanta and Dimitris Ntelitheos. 2004. The Acquisition of the Malagasy Voicing System. Dept. of Linguistics, UCLA. Keenan, Edward L. 1972. Relative Clause Formation in Malagasy (and some related some not so related languages. The Chicago Which Hunt University of Chicago. Keenan, Edward L. 1976. Remarkable subjects in Malagasy. In Subject and Topic C. Li (ed). Academic Press. Keenan, Edward L. 1995. Predicate-argument structures in Malagasy. In Grammatical Relations C. Burgess, K. Dziwirek, and D. Gerdts (eds). CSLI. Stanford. pp.171 B 217 Keenan, E.L. 2000. Morphology is structure: A Malagasy test case. In Paul, Phillips, and Travis (eds) Formal Issues in Austronesian Linguistics pp.27 B 49. Kluwer Keenan, Edward L. and Cecile Manorohanta. 2001. A quantitative study of voice in Malagasy. in Oceanic Linguistics vol 40:67 B 85. [K&P] Keenan, Edward L. and Miriam Polinsky. 1998. Malagasy Morphology. In The Handbook of Morphology. A. Spencer and A. Zwicky (eds) Blackwell. pp. 563 B 624. Keenan, E.L. and Baholisoa Ralalaoherivony. 2000. Raising from NP in Malagasy. Linvisticae Investigationes 23:1:1 B 44. [K&R] Keenan, Edward L. and Jean-Paulin Razafimamonjy. 2004. Reciprocals in Malagasy. in Oceanic Linguistics vol. 43.1:177 B 207. [K&S] Keenan, E.L. and E.P. Stabler. 2003. Bare Grammar CSLI, Stanford. Koopman, H. 2005. Malagasy imperatives. In UCLA Working Papers in Linguistics:12. Kroeger, Paul. 1988. Verbal focus in Kimaragang. In Papers in Western Austronesian Linguistics No. 3:217 B 40. Pacific Linguistics, A-78. Kroeger, Paul. 1993. Phrase Structure and Grammatical Relations in Tagalog. CSLI, Stanford. Ntelitheos, D. 2005. Variation in the distribution of genitive subjects in Malagasy. 35th Michigan Linguistic Society, Lansing. Paul, Ileana. 1996. The Malagasy genitive. In Pearson & Paul, The Structure of Malagasy Vol I, UCLA Occasional Papers in Linguistics 17. Pp 76 B 92. Paul, Ileana. 1999. Malagasy Clause Structure. PhD Diss. Dept. of Linguistics, McGill University Paul, Ileana and Lucie Rabaovololona. 1998. Raising to object in Malagasy. In The Structure of Malagasy vol II I. Paul (ed). Dept. of Linguistics, UCLA. pp.50 B 65. Pearson, Matthew. 2005. The Malagasy subject/topic as an A= element. in Natural Language and Linguistic Theory 23:381 B 457. Phillips, V. 2000. The interaction between prefix and root: the case of maha- in Malagasy. In Formal Issues in Austronesian Linguistics I. Paul, V. Phillips, L. Travis (eds) Kluwer, Dordrecht. Pp. 85 B 105. Polinsky, M. and E. Potsdam. 2005. Malagasy control and its theoretical implications. BLS 30. Prentice, D.J. 1971. The Murut Languages of Sabah Pacific Linguistics Series C B No.18. The Australian National University. Canberra, Australia. Rabenilaina, R-B. 1993. Le verbe malgache B Etude des formes predicatives. Ambozontany: Fianarantsoa, Madagascar. [R&T] Rackowski, Andrea and Lisa Travis. 2000. V-Initial Languages: X or XP movement and adverbial placement. In The Syntax of Verb Initial Languages. Andrew Carnie and Eithne Guilfoyle (eds) OUP Pp. 117 B 143. Rahajarizafy, Antoine. 1960. Essai sur la grammaire malgache Imprimerie Catholique, Antanimena, Tananarive [RR] Rajemisa-Raolison, Regis. 1971. Grammaire malgache. Fianarantsoa. Randriamasimanana, C. 1986. The Causatives of Malagasy Univ. of Hawaii Press. Richards, N. 2000. Another look at Tagalog subjects. In I. Paul, V. Phillips, and L. Travis (eds). Formal Issues in Austronesian Linguistics Pp 105 B 116. Kluwer Ross, M. 2006. The argument structure of undergoer voice clauses in Formosan and other Philippine-type languages. Presented at AFLA 13, Tsing Hua Univ. Taiwan Sabel, J. 2005. Grammatical functions in Malagasy syntax and the >Complements only= restriction. Presented at AFLA XII, UCLA Schachter, Paul. 1976. The subject in Philippine languages: topic, actor, actor- topic, or none of the above. in C. Li (ed) Subject and Topic 1976:491 B 518. Academic Press. Schachter, P. and Fe Otanes. 1972. Tagalog Reference Grammar UC Press. Spencer, A. 1995. Incorporation in Chukchi. Language 71:439 B 489. Van Valin, R.D. 1992. Incorporation in universal grammar: a case study in theoretical reductionism. J. Linguistics 28:199 B 220. Wechsler, Stephen and I.Wayan Arka. 1998. Syntactic Ergativity in Balinese. In Natural Language and Linguistic Theory 16:387 B 441. Footnotes 1 In addition to Malagasy (Madagascar) we include here Tagalog (Schachter and Otanes 1972, Kroeger 1993) and Philippine languages generally, Timugon (Prentice 1971), Kimaragang Dusun (Kroeger 1988), Balinese (Wechsler and Arka 1998), Atayal (Huang, 2002) and Seediq (Aldridge 2002). 2 Erwin (1996) and Pearson (2005) show that many roots are consonant final, in distinction to their standard dictionary form (Abinal & Malzac 1963). If a vowel is supplied via a suffix the final consonant, possibly mutated, is pronounced. With no suffix the consonant drops. For example, from the root taov- >do= we derive manaov manao >does, makes= with loss of -v, but imperative manaovy and circumstantial (later) anaovana. We are not concerned here with the morphophonology of affixation and I tend to give roots in their dictionary form. 3 Linguists often accord a special, and pretheoretically unclear, status to this DP using terms such as Asubject@, Atopic@, Apivot@, or Atrigger@ . We do not do so here, limiting ourselves solely to what we can structurally identify. 4 Examples are given in standard orthography. Possessor formation involves prenasalization of the result of converting the /r/ of Rabe to the post alveolar affricate /dr/. So orthographic n-dR is a single phoneme, a prenasalized affricate /ndr/. 5 Following K&S the set of expressions of a language is the closure of the lexicon under the structure building functions of the language. Morphological functions, like AN, are among those in Malagasy. Another one with a much larger domain is Merge. It combines Pn+1's with argument expressions to form Pn=s. 6 Ultimately we need a richer, intensional, semantics, but for now simplicity rules. 7 Failure to behave like Passive is now well acknowledged (Pearson 2005). We note: Imperatives the non-AV -ina, a-, and -ana verbs (see later) present imperative forms, as do the AV verbs (Koopman 2005). In fact these are the normal forms used when there is a choice. In a recent acquisition study (Hyams, Manorohanta and Ntelitheos 2004) covering three children aged 18 - 30 months over an 8 month period, K&M (Keenan & Manorohanta 2004) report that all 26 of the imperatives in the data were non-AV forms. In general it is the addressee phrase that is missed out in imperatives regardless of voice. So that will be the nominative for AV imperatives, (a), but the genitive complement of non-AV forms in a-, -ina and Bana, (b,c,d): a. manolóra (m+aN+tólotra+a) vary azy! (Stress marked) serve+AV pres+aN+offer+imp rice them serve them rice b. tolóry (tólotra+y) vary izy serve+GV offer+imp rice they serve them rice c. atolóry (a+tólotra+y) azy ny vary serve TV+offer+imp them the rice serve them the rice d. anolóry ([aN+tolotra]+ana+y) vary azy ny lovia vaovao serve+CV rice them the dishes new Use the new dishes to serve them rice As always the P1 argument, when present, must be definite. Imperative suffixes differ for AV verbs and non-AV ones, being the same for a-, -ina and -ana verbs. Selection Both AV and non-AV forms are selected by modal auxiliaries. mahazo >permitted= and mahay >able= select AV verbs, their TV roots azo and hay select nonAV verbs. Replacing AV by non-AV verbs in (i.a) and (ii.a) yields ungrammaticality. i. a. b. ii. a. b. mahazo manao (m+aN+tao(v)) izany aho permitted pres+aN+do that I I can do that Tsy azoko (azo + ko) atao izany Not can + 1.sg.gen a+do that I can=t do that mahay manoratra (m+aN+soratra) amin=io penina io aho pres+can write pres+aN+write with=that pen that I I am able to write with that pen tsy haiko (hay + ko) anoratana (aN+soratra+ana) io penina io not able+1.sg.gen aN+write+ana that pen that I can=t write with that pen Control With verbs of desire, aspect, and intent control is vested in the Agent: a. n+i+kasa (ny) hamaky (h+aN+vaky) io boky io aho past+I+intend (the) fut+AN+read that book that 1.sg.nom I intended to read that book b. nokasaiko (no+kasa+ina+ko) (ny) hovakina (ho+vaky+ina) io boky io past+intend+INA+1.sg.gen (the) fut+read+ina that book that I intended to read that book A second type of Agent Phrase control is seen in coordinate structures, as in (a) and (b), from Keenan 1995 (taken from Malagasy novels): a. araka izao [[ahitana-i sy ahafantara-nareoi] ahy] izao according this [[[see+CV and know+CV]+2.gen.pl] 1.sg.acc] this according to that which you see and (you) understand of me b. ataoko fa [voa+jery-i sy voa+dinikaoi] tsara ireo sary teo I+think that [[TV+see and TV+examine+2.sg.gen] good those pictures there] I think that those pictures there have been seen (by you) and examined well by you K&M show that Agent phrases of TV verbs are present or controlled in more than 2/3 of their occurrences. So in distinction to English passives verbs, TV verbs derived from transitive roots seem not to be intransitive. And third, the other side of this coin, Agents of verbs in all voices can control other missing Agents, as the same examples show. 7 Occasionally it suffixes -anana. 8 In two cases not discussed here the P1 argument of an ANA verb may be selected by the original AV verb. (1) The P1 argument may be the Theme understood partitively, see Paul (1999). Thus APlease buy (me) some paper@ could be rendered Mba ividiano ny taratasy with a CV imperative and definite P1 argument ny taratasy >the paper=. M. Ross (2006) notes a partial affectedness reading deriving from Alocative voice@ in Puyuma (Formosan). Second, occasionally CV forms get pressed into duty when appropriate INA forms lack. Thus reciprocal verbs, with prefix if-, are not in the domain of INA, so to talk about >the many things we taught each other= we would say ny zavatra betsaka nifampianarantsika with a CV form of AV Areciprocally cause to learn@. 9 Adding basic tense interpretation is straightforward: verbs just map intervals to what they currently denote, and the interval gets passed up stepwise to the P0 level. We omit this here because we are not studying tense marking. 10 Indifferently we could put pairs like (aN-,enjika) in the domain of Merge, setting its value to AN(enjika). 11 The P1 of a RC may be preceded by izay (morphologically constant) interpreted as: IZAY(P)(A) = A {x|P(x) = True}. A is a N denotation, P a P1 one. Thus RCs with or without izay are both directly interpreted and yield RCs denoting the same set. 12 As a reader points out, this is only true for a certain (admittedly extensive) range of RCs. A P1 headed by CV verb can take a DP to its right to form a P0 only if it is one denoting a place, instrument, means, matter, benefactee, recipient. According to [RR:112-113] complements of time, cause, purpose, manner, point of view, and price must occur fronted in the cleft construction in otherwise simple Ss: Rahampitso no handehanantsika Tomorrow Focus fut+[AN+go]+ANA+1pl.incl It is tomorrow that we will go If we find a verb that can take a complement in the latter category (say Time) but not in any of the former (place, instrument, etc.) the CV form of that verb couldn=t be learned based on experience with nuclear Ss but would require a cleft, question or relative clause. It is perhaps worth noting that, impressionistically, cleft Ss are much more widely used in spoken Malagasy than the comparable AIt was X who...@ in English.