Argument Structure-Driven Parsing in Tagalog

Previous literature on sentence processing has identified several structure building biases. One is a preferencetoward satisfying a verb’s argument structure as soon as possible, the argument preference [1,2,3]. Another is apreference toward attaching an incoming phrase at the right edge of the existing structure, the local attachment bias([4,5]). The present study provides further support for the importance of argument structure in a novel empiricaldomain: argument/possessor ambiguity in Tagalog.

Because Tagalog is verb-initial, argument structure information is available from the beginning of a sentence.In the Recent Perfective (RP) form, the case morphology on the arguments of some verbs is neutralized withthe genitive form (glossed GEN here in all occurrences). Thus by including a genitive-marked possessor NPmodifying one of the sentential arguments, it is possible to produce sentences containing a string of threeidentically marked NPs. In such sentences the parser must determine the grammatical function and thematicroles of the NPs without case cues. When a transitive RP verb is followed by more than two NPs, Tagalog wordorder is such that one NP must be a genitive on the preceding NP, but the grammar does not determine whetherthe constituency is [. . . [NP1 NP2] NP3] or [. . . NP1 [NP2 NP3]]. Here, the two parsing biases give differentpredictions, detailed below. In this way, how Tagalog speakers parse transitive RP verbs followed by more thantwo NPs can distinguish the argument and local attachment preferences.

In a word-by-word moving-window self-paced reading experiment, 38 native speakers of Tagalog in thePhilippines read sentences (N=32, interspersed with over a hundred fillers) like (1), with three NPs following theverb. In (1), Case Form of NPs (Neutralized vs. Non-Neutralized), Plausibility (Possessed-Possessor sequenceis plausible vs. implausible) and placement of complex NP (first vs. second argument) are manipulated asindependent factors.

(1) Example Stimuli (simplified)a. Pumili

chosengGEN

rantserocowboy

angNOM

narsesnurse

ngGEN

pasyente.patient

The patient’s nurse chose the cowboy.b. Pumili

chosengGEN

narsesnurse

angNOM

pasyentepatient

ngGEN

rantsero.cowboy

The cowboy’s patient chose the nurse.c. Pumili

chosengGEN

rantserocowboy

ngGEN

narsesnurse

angNOM

pasyente.patient

The patient chose the nurse’s cowboy.d. Pumili

chosengGEN

narsesnurse

ngGEN

pasyentepatient

angNOM

rantsero.cowboy

The cowboy chose the patient’s nurse.e. Kapipili

RP.choselangjust

ngGEN

narsesnurse

ngGEN

pasyentepatient

ngGEN

rantsero.cowboy

∼ The cowboy just chose the patient’s nurse.f. Kapipili

RP.choselangjust

ngGEN

rantserocowboy

ngGEN

narsesnurse

ngGEN

pasyente.patient

∼ The patient’s nurse just chose the cowboy.

In the non-case-neutralizing normal perfective, 3-NP sequences are unambiguous provided the argumentmarked with ”ang” is dislocated after the argument marked with “ng”. So in (1a-d) only a single unambiguousparse (equivalent to the translation provided) is possible, because the arguments of the verb are distinctlycase-marked. In (1a,b), the complex NP comes second, while in (1c,d), it come first. The additional factorof Plausibility (Possessed-Possessor sequence is plausible vs. implausible) was also varied in the non-case-neutralizing conditions. In (1a,d), the complex NP (possessed and possessor) is semantically plausible, whilein (1b,c) it is implausible. Due to the absence of case cues, plausibility is indistinguishable from position of thecomplex NP for the RP conditions. In RP conditions, NP2 can be parsed either as the possessor of NP1 or anargument of the verb and the possessed of NP3, with Plausibility serving to bias the parse to one constituency orthe other. In (1e), the former parse is more plausible, while in (1f), the latter is.

Michael Frazier (Northwestern University) Poster Session I Friday 2:45 - 4:45

In the absence of case-cues to distinguish arguments, the parser may behave in two ways. If the parser isdriven by verbal argument structure (Hypothesis A) and by default assigns incoming NPs to theta-positions untilthe available theta-roles are exhausted, NP1 and NP2 will be initially parsed as arguments of the verb. Thusthe parser will experience difficulty when this analysis turns out to be incorrect and NP2 must be reanalyzed asa genitive on the first. This, then, predicts that when NP3 is not a plausible possessor for NP2, reading timeslowdown will be observed at NP3 (e.g. 1e).

In contrast, if the parser preferentially attaches incoming material as locally as possible (Hypothesis B), itshould exhibit difficulty when the second NP in a sequence of two ‘ng’-marked NPs is not a plausible genitive onthe first. This is because a sequence of two ‘ng’-marked NPs has a parse as a Possessed-Possessor sequence, andattaching an incoming NP as a genitive inside an NP just built should be preferred over attaching an argument bythe principles of low attachment and minimal attachment. Reading time slowdown should therefore be observedat NP2 when it is not a plausible possessor for NP1 (e.g. 1f).

In the present study, it is found that reading times at NP3 are significantly slower in the recent perfectivewhen NP3 is not a plausible possessor for NP2 (p<0.05; linear mixed model, t=-3.1186, df=7). No significanteffects are found at NP2 in any condition. (Reading times for the critical regions do not differ significantlyamong the four non-neutralized conditions (a-d).) This supports the hypothesis (Hypothesis A) that argumentstructure information from the verb drives the parser’s integration of case-ambiguous elements, overriding thelocal attachment preference. This finding is not consistent with the local attachment hypothesis (Hypothesis B)according to which the parse should default to building maximally locally in the absence of case cues to thecontrary. Implications for the role of case-marking in language processing more generally are discussed.

References: [1] Pritchett (1988). Garden-path phenomena and the grammatical basis of language processing.Language, 64, 539-576. [2] Speer & Clifton (1998). Plausibility and argument structure in sentencecomprehension. Memory & Cognition 26(5). [3] Gilboy et al. (1995). Argument structure and associationpreferences in Spanish and English complex NPs. Cognition 13t-167. [4] Kimball, J. (1973). Seven principles insurface structure parsing in natural language. Cognition 2:1547. [5] Phillips & Gibson (1997). On the Strength ofthe Local Attachment Preference. J Psycholinguistic Research 26.3.

Michael Frazier (Northwestern University) Poster Session I Friday 2:45 - 4:45