Bhuvana Narasimhan

A Multi-Representational and Multi-Layered Treebank for Hindi/Urdu

This paper describes the simultaneous development of dependency structure and phrase structure treebanks for Hindi and Urdu, as well as a PropBank. The dependency structure and the PropBank are manually annotated, and then the phrase structure treebank is produced automatically. To ensure successful conversion the development of the guidelines for all three representations are carefully coordinated.

Differential Case-Marking in Hindi

Publisher Summary This chapter describes how a model of case marking based on the two functions of case should be augmented with the notion of argument strength and shows how this extended approach derives split-case marking in the subject and object positions in Hindi. Hindi presents a serious problem for an analysis along the lines of the distinguishing function of case when it comes to differential subject marking. Differential subject marking in transitive clauses is restricted by the verb class and aspect in the sense that agents of perfective highly transitive predicates are assigned the ergative case. For the class of highly transitive predicates, there seems to be a general principle of preserving the relative distance between the subject and the object such that case marking appears when the object becomes more agent-like.

Schwa-Deletion in Hindi Text-to-Speech Synthesis

We describe the phenomenon of schwa-deletion in Hindi and how it is handled in the pronunciation component of a multilingual concatenative text-to-speech system. Each of the consonants in written Hindi is associated with an “inherent” schwa vowel which is not represented in the orthography. For instance, the Hindi word pronounced as [namak] (‘salt’) is represented in the orthography using the consonantal characters for [n], [m], and [k]. Two main factors complicate the issue of schwa pronunciation in Hindi. First, not every schwa following a consonant is pronounced within the word. Second, in multimorphemic words, the presence of a morpheme boundary can block schwa deletion where it might otherwise occur. We propose a model for schwa-deletion which combines a general purpose schwa-deletion rule proposed in the linguistics literature (Ohala, 1983), with additional morphological analysis necessitated by the high frequency of compounds in our database. The system is implemented in the framework of finite-state transducer technology.

Splitting the notion of ‘agent’: case-marking in early child Hindi

Two construals of agency are evaluated as possible innate biases guiding case-marking in children. A BROAD construal treats agentive arguments of multi-participant and single-participant events as being similar. A NARROWER construal is restricted to agents of multi-participant events. In Hindi, ergative case-marking is associated with agentive participants of multi-participant, perfective actions. Children relying on a broad or narrow construal of agent are predicted to overextend ergative case-marking to agentive participants of transitive imperfective actions and/or intransitive actions. Longitudinal data from three children acquiring Hindi (1;7 to 3;9) reveal no overextension errors, suggesting early sensitivity to distributional patterns in the input.

Cutting, breaking, and tearing verbs in Hindi and Tamil*

Abstract Tamil and Hindi verbs of cutting, breaking, and tearing are shown to have a high degree of overlap in their extensions. However, there are also differences in the lexicalization patterns of these verbs in the two languages with regard to their category boundaries, and the number of verb types that are available to make finer-grained distinctions. Moreover, differences in the extensional ranges of corresponding verbs in the two languages can be motivated in terms of the properties of the instrument and the theme object.

The Hindi/Urdu Treebank Project

The goal of Hindi/Urdu treebanking project is to build multi-layered treebanks that will provide both syntactic and semantic annotations. In the past two decades, dozens of treebanks have been created for languages such as Arabic, Chinese, Czech, English, French, German, and many more. Our treebanks differ from the previous treebanks in two important aspects: they are multi-representational, i.e., they include several layers of representation from the initial design; and they cover two standardized registers that are often considered separate languages: Hindi and Urdu.

Verb representation and thinking-for-speaking effects in Spanish-English bilinguals

Speakers of English habitually encode motion events using manner-of-motion verbs (e.g., spin, roll, slide) whereas Spanish speakers rely on path-of-motion verbs (e.g., enter, exit, approach). Here, we ask whether the language-specific verb representations used in encoding motion events induce different modes of “thinking-for-speaking” in Spanish–English bilinguals. That is, assuming that the verb encodes the most salient information in the clause, do bilinguals find the path of motion to be more salient than manner of motion if they had previously described the motion event using Spanish versus English? In our study, Spanish–English bilinguals described a set of target motion events in either English or Spanish and then participated in a nonlinguistic similarity judgment task in which they viewed the target motion events individually (e.g., a ball rolling into a cave) followed by two variants a “same-path” variant such as a ball sliding into a cave or a “same-manner” variant such as a ball rolling away from a cave). Participants had to select one of the two variants that they judged to be more similar to the target event: The event that shared the same path of motion as the target versus the one that shared the same manner of motion. Our findings show that bilingual speakers were more likely to classify two motion events as being similar if they shared the same path of motion and if they had previously described the target motion events in Spanish versus in English. Our study provides further evidence for the “thinking-for-speaking” hypothesis by demonstrating that bilingual speakers can flexibly shift between language-specific construals of the same event “on-the-fly.”

The influence of discourse context on children’s ordering of “new” and “old” information

Abstract Adult speakers typically order referents that have been previously mentioned in the discourse (“old” referents) before newly introduced referents (“new” referents). But 3–5-year-olds acquiring German exhibit a “new-old” preference in a task involving question-answer sequences (Narasimhan, Bhuvana and Christine Dimroth. 2008. Word order and information status in child language. Cognition 107. 317–329). Here we ask whether we can change 4–5-year-olds’ new-old preference by manipulating the context in order to encourage connected discourse. Findings show that discourse context changes children’s new-old preference. Children produce the new-old order in fluent utterances and the old-new order in non-fluent utterances. Adult controls overwhelmingly prefer the old-new order, even more so when the weight (number of syllables) of the old referent label is greater than that of the new referent label. Our study demonstrates that although cognitive and communicative biases may influence children’s ordering patterns in non-adult-like ways, such patterns are not categorical, but are flexibly influenced by factors such as discourse context.

The Influence of Visual, Auditory, and Linguistic Cues on Children’s Novel Verb Generalization

An influential theory proposes that children learn words by initially relying on perceptual cues to identify referents, only later recruiting social or linguistic cues to acquire new words (Golinkoff and Hirsh-Pasek 2007). Here we ask whether a similar bias also characterizes children’s extensions of words to novel contexts. In order to establish relative preference for cues in different modalities, we focus on how children recruit auditory, visual, and lexical distributional cues in extending verbs to unfamiliar contexts involving missing or conflicting information regarding the referent of the verb. Four-year olds were taught two nonce verbs and subsequently asked to produce one of the two verbs in novel contexts involving conflicting cues (e.g., the linguistic cue associated with one verb is paired with the visual cue associated with the other verb). To assess relative cue preference, children’s responses in the ‘bimodal conflicting cues’ condition were compared with their responses in a ‘baseline unimodal’ condition in which only one of the three cues was presented. Four-year olds overwhelmingly prefer to produce verbs associated with visual cues versus auditory or linguistic cues. But there is no advantage for auditory over linguistic cues when they are placed in conflict. These findings demonstrate that children weight perceptual cues more heavily than linguistic cues during the process of verb generalization, but only if the cues are visual. The cue-weighting strategies that children recruit during verb generalization are similar in some respects to those proposed for the process of mapping word forms onto referents during word learning and additionally may be driven by visual dominance effects that characterize cognitive processing in other kinds of tasks (Colavita 1974).

Satellite- vs. Verb-Framing Underpredicts Nonverbal Motion Categorization: Insights from a Large Language Sample and Simulations

Is motion cognition influenced by the large-scale typological patterns proposed in Talmy’s (2000) two-way distinction between verb-framed (V) and satellite-framed (S) languages? Previous studies investigating this question have been limited to comparing two or three languages at a time and have come to conflicting results. We present the largest cross-linguistic study on this question to date, drawing on data from nineteen genealogically diverse languages, all investigated in the same behavioral paradigm and using the same stimuli. After controlling for the different dependencies in the data by means of multilevel regression models, we find no evidence that S- vs. V-framing affects nonverbal categorization of motion events. At the same time, statistical simulations suggest that our study and previous work within the same behavioral paradigm suffer from insufficient statistical power. We discuss these findings in the light of the great variability between participants, which suggests flexibility in motion representation. Furthermore, we discuss the importance of accounting for language variability, something which can only be achieved with large cross-linguistic samples