Linguistics Calendar Archive

A challenge for L1 acquisition research is the formulation of sound, yet unobtrusive methodologies that reliably address questions concerning mental representations of linguistic knowledge. This challenge is most evident in the context of child language development, particularly given the level of experimentation permissible on infants. This talk will present an Agent-based computer model that explores the notion of Phonological Bootstrapping (Christophe et al. 1994, Christophe and Dupoux 1996, etc.). Specifically, Phonological Bootstrapping consists of a two-part analysis of primary phonological and acoustical input which allows infants to start acquiring mental representations of the lexicon and syntax of their native language(s).

Crucially, this computer model frames the dynamic processes of L1 bootstrapping in terms of Complex Adaptive Systems (CAS) approaches, utilizing specific CAS components to organize emergent linguistic knowledge over time. The core architecture in the model is driven by an adaptable analogical search for structures across primary linguistic inputs, based closely on Mitchell's (1993) and Hofstadter's (1995) program of Copycat, a cognitive computer model of high-perception. While Copycat has successfully modeled various complex sequence domains (e.g., music, numbers, and letters), to date it has not been applied in developmental learning tasks. We are currently constructing the Babycat model in order to begin to test L1 acquisition questions.

The main loop of the Babycat program is simple: an input sentence is received and an attempt is made to process the information by assigning descriptions to elements of the input, in order to build lower-level objects into higher-level objects (e.g., bootstrapping sound segments into syllables and words) and to assign descriptions to those newly constructed objects. Babycat eventually provides a mental representation for the input string as a sentence reflecting the appropriate components. Unlike Copycat, Babycat includes a learning module such that once a given input is processed, Babycat will then be able to modify its long term knowledge structures based on that experience.

A series of pilot experiments demonstrating Babycat's applications are highlighted, including a preliminary comparison between monolingual and bilingual bootstrapping. Finally, the talk concludes with implications for computer models such as Babycat in informing and shaping the general L1 acquisition research paradigm.

Probabilistic models of corpus data can be used both to predict higher-level grammatical choices (as discussed, for example, by Gries 2005 and Bresnan, Cueni, Nikitina, and Baayen 2007) and to quantify changes in such choices across different speaker groups in geographic or social space and in historical time (as shown, for example, by Hinrichs and Szmrecsanyi 2007 and Tagliamonte and Jarasz 2008). In recent work we have used probabilistic models in a novel way, to measure and compare the syntactic predictive capacities of speakers of different varieties of the same language in parallel psycholinguistic tasks. We have found a new kind of evidence that speakers of English have detailed probabilistic knowledge of higher-level grammatical structures in their language, which can be tapped in multiple tasks.

In the case of the dative alternation in American English, recent studies have found effects of syntactic probabilities on sentence ratings (Bresnan 2007), phonetic production (Tily, Gahl, Arnon, Snider, Kothari, and Bresnan to appear), and effects of verb bias on eye movements (Tily, Hemforth, Arnon, Shuval, Snider, and Wasow 2008), and earlier work has shown that there are important parallels between the comprehension and production of such constructions in the use of distributional information (MacDonald 1999: 189; Stallings, MacDonald, and O'Seaghdha 1998). But it has not yet been shown that speakers' knowledge of probabilistic grammatical choices can vary across different varieties of the same language and can be detected psycholinguistically in the individual.

Given evidence of probabilistic changes in the English dative alternation across varieties of English (Rohdenburg 2007, Mukherjee and Hoffman 2006, Bresnan and Hay 2008, Collins 1995 vs. Bresnan, Cueni, Nikitina, and Baayen 2007), we examined responses to the verb-argument dependencies in the English dative alternation by six different groups of American and Australian subjects in three parallel psycholinguistic experiments involving sentence ratings, decision latencies during reading, and sentence completion. The experimental items, together with their contexts, were all sampled from a database of 2349 spoken corpus datives stratified by corpus model probabilities.

The findings show that both the Australian and the American subjects can make reliable probabilistic predictions of the syntactic choices of others, that in both groups lexical decision latencies during reading vary inversely with syntactic probabilities, and that there is indeed subtle covariation in these psycholinguistic tasks, which can be explained by differences in patterns of usage in language production between the Australian and American subjects.

Electoral campaigns are the foundation of democratic governance; yet, scholarship on the content of campaign communications remains underdeveloped. In this paper, we advance research on U.S. congressional campaigns by integrating and extending extant theories of campaign communication. We test the resulting predictions with a novel dataset based on candidate websites, over three election cycles. Unlike television advertisements or newspaper coverage, websites provide an unmediated, holistic, and representative portrait of campaigns. We find that incumbents and challengers differ across a broad range of behaviors that reflect varying attitudes toward risk, that incumbents' strategies depend on the competitiveness of the race, and that candidates link negative campaigning to other aspects of their rhetorical strategies. Our efforts provide researchers with a basis for moving towards a more complete understanding of congressional campaigns.

Several studies have shown that young children can use the linguistic context that a new verb appears in as well as their observation of the environment to make guesses at the verb's meaning. However, successful word learning requires not just finding the referent in the immediate scene, but also determining the word's extension (learning that e.g., "push" first encountered in a door-pushing scene also applies to chair-pushing events).

I will describe three new experiments looking at 2-year-olds' abilities to map a novel verb onto an event given minimal exposure, and to immediately extend the verb beyond the particular scene on which they were taught. I will argue that toddlers' initial representations of verb meaning are sufficiently abstract to allow them to extend new verbs appropriately, but that this ability may hinge on certain informational requirements being met, such as a sufficiently informative linguistic context.

Location: Chambers Hall (600 Foster Street), Lower Level

In this talk I will discuss my ongoing research project investigating the socio-cultural aspects of language acquisition and socialization patterns among Kurds in eastern Turkey. This project includes the collection, analysis, and comparison of cross-sectional and semi-longitudinal naturally-occurring speech and elicited grammatical forms from three groups of young ethnic Kurdish children: Kurdish monolinguals, Turkish monolinguals, and Turkish-Kurdish bilinguals. In addition, on the basis of extensive ethnographic observation supplemented by interviews with family and community members, the linguistic and social contexts of acquisition are examined. These methods attempt to tease out patterns in the process of children's acquisition of communicative competence in the language community in addition to language structures. For instance, community patterns of language use and community and familial attitudes affect how children acquire and use language in these communities, including language choice, code-switching, the rate or success of acquisition, and the use of variable forms.

Due to the lack of research about the structure of Kurdish in this region and the absence of any study on the acquisition of Kurdish, as a first step in this project, I have examined the acquisition of basic patterns in monolingual Kurdish, specifically the acquisition of the split-ergative case system. This analysis has shown a high degree of variability in the use of split-ergative case-marking in the adult community, patterns which are being acquired by the children. These data suggest that children, despite receiving highly variable input, acquire the grammar of the speech community to which they belong, in order to be considered communicatively competent speakers of that of community.

Location: Chambers Hall, Lower Level

Most models of sentence processing focus on how comprehenders establish relationships between words to form a sentence. However, for successful language comprehension, comprehenders must also establish relationships between sentences in order to form larger discourse structures. This talk addresses the question of whether comprehenders generate expectations about upcoming discourse continuations, and furthermore whether those expectations have an impact on sentence-internal phenomena such as coreference and the resolution of syntactic ambiguity. For example, consider (1):

Several different relationships can be inferred to hold between the two sentences in (1), and the different relationships in turn yield different interpretations of the two pronouns. In the case of (1), an 'Explanation relation' supports the reading in which Jane kicked Mary, whereas a narrative 'Occasion relation' supports the reading in which first Mary scolded Jane and then, on top of that, Mary also kicked her. In this talk, I identify several factors that influence comprehenders' expectations about what the operative intersentential coherence relation is likely to be in a given passage. For an example like (1), the presence of a so-called 'implicit-causality' verb (scold) contributes to the likelihood of an Explanation relation, which in turn supports the pattern of pronoun interpretation in which it was Jane who kicked Mary.

The results of a set of off-line and on-line experiments fit within a larger picture of expectation-driven processing in language comprehension. Previous work has shown that comprehenders are sensitive to statistical regularities at the level of sounds, words, and syntactic structures; the experiments presented here provide the first evidence of expectation-driven processing at the discourse level as well.

This dissertation investigates the structure of DP by focusing onnumber marking and determiners across languages. Number is usuallyassumed to come in two distinct forms: singular/plural-type markingand numeral classifiers.  Languages' determiners such as articles varyin their meaning and many languages lack one or more of them. Someearlier theories of both number and determiners are based on thedenotation of nouns, which some claim is parameterized by language(Chierchia 1998). Another theory dispenses with the parameterizationof nouns and posits a common syntax (Borer 2005).

I propose that underlying both number and determiners is a smallset of syntactic features in the functional heads above NP. Theindividual features are universal but because they combine in variousways a wide range of morphological and interpretive variation followsnaturally. Being more fine-grained than earlier theories, thefeature-based system overcomes empirical problems about number andoffers some economy. Similarly for determiners, I claim that featurescan account for cooccurrence restrictions and their variation inmeaning in a principled way, particularly with regard tospecificity. Regardless of whether the morphology is overt, number anddeterminer features are always present to assure both syntactic andsemantic composition of DPs.