Research

The meaning of ‘most’ can be described in many ways. We offer a framework for distinguishing semantic descriptions, interpreted as psychological hypotheses that go beyond claims about sentential truth conditions, and an experiment that tells against an attractive idea: ‘most’ is understood in terms of one-to-one correspondence. Adults evaluated ‘Most of the dots are yellow’, as true or false, on many trials in which yellow dots and blue dots were displayed for 200ms. Displays manipulated the ease of using a “one-to-one with remainder” strategy, and a strategy of using the Approximate Number System to compare of (approximations of) cardinalities. Interpreting such data requires care, in thinking about how meaning is related to verification. But the results suggest that ‘most’ is understood in terms of cardinality comparison, even when counting is impossible.

Measuring event-related potentials (ERPs) has been fundamental to our understanding of how language is encoded in the brain. One particular ERP response, the N400 response, has been especially influential as an index of lexical and semantic processing. However, there remains a lack of consensus on the interpretation of this component. Resolving this issue has important consequences for neural models of language comprehension. Here we show that evidence bearing on where the N400 response is generated provides key insights into what it reflects. A neuroanatomical model of semantic processing is used as a guide to interpret the pattern of activated regions in functional MRI, magnetoencephalography and intracranial recordings that are associated with contextual semantic manipulations that lead to N400 effects.

This dissertation is concerned with whether the sentence processor can compute plausible relations among a cluster of neighboring open class words without taking into account the relationships between these words as dictated by the structure of the sentence. It has been widely assumed that compositional semantics is built on top of syntactic structures (Heim & Kratzer, 1998; Pollard & Sag, 1994). This view has been challenged by recent electrophysiological findings (Kim and Osterhout, 2005; Kuperberg, 2007; van Herten et al., 2005, 2006) that appear to show that semantic composition can proceed independently of syntactic structure. This dissertation investigates whether the evidence for independent semantic composition is as strong and widespread as has been previously claimed. Recent studies have shown that sentences containing a semantically anomalous interpretation but an unambiguous, grammatical structure (e.g., The meal was devouring) elicit a P600 response, the component classically elicited by syntactic anomalies, rather than an N400, the component typically elicited by semantic anomalies (Kim and Osterhout, 2005). This has been interpreted as evidence that the processor analyzed meal as a good theme for devour, even though this interpretation is not supported by the sentential structure. This led to the claim that semantic composition can proceed independently of syntactic structure. Two event-related potentials (ERP) studies investigated whether the processor exploits prior structural biases and commitments to restrict semantic interpretations to those that are compatible with that expected structure. A further ERP study and a review of relevant studies reveal that in the majority of studies the P600 is not modulated by manipulations of thematic fit or semantic association between the open class words. We argue that a large number of studies that have been taken as evidence for an independent semantic processing stream can be explained as violations of the verb's requirement that its subject be agentive. A small number of studies in verb-final languages cannot be explained in this way, and may be evidence of independent semantic composition, although further experimental work is needed. We conclude that the evidence for independent semantic composition is not as extensive as was previously thought.

This dissertation discusses how pragmatic factors contribute to children's behavior in interpreting scopally ambiguous forms. In particular, we look at children's interpretation of negated sentences involving disjunction in the object (NegDisjunction). Languages like English and Chinese allow scope interaction between negation and disjunction of this kind of strings and thus two corresponding interpretations: the narrow scope disjunction interpretation (the NSD, meaning "neither"), thus the wide scope disjunction interpretation (the WSD, meaning "not this or not that"); but languages like Japanese only allow the WSD. Previous studies found out that children of different languages accessed the NSD instead of the WSD given "not this or not that" scenarios (e.g., Crain, Gualmini & Meroni 2000; Goro & Akiba 2004a; Jing, Crain & Hsu 2005) and concluded that preschool children systematically lack the WSD in their grammar. However, given the fact that the WSD is pragmatically more complex than the NSD, and the well documented observations that children's immature capacity in pragmatic computation sometimes masks their linguistic competence (e.g., Gualmini 2004; Musolino & Lidz 2006), the findings in previous studies could reveal children's strong preference toward the NSD rather than their lack of the WSD. Four main experiments, which aim to test the hypothesis that children's grammar can generate the WSD and that children can access this interpretation when the relevant pragmatic computation is facilitated, are reported in this dissertation. During various experimental manipulations designed to facilitate children's pragmatic computation, we observed that children accessed the normally dispreferred WSD more often, when the "not this or not that" meaning was made more directly relevant in the context, when explicit disambiguating information was present in the discourse, after they were trained to be more sensitive to the "not this or not that" aspect of the context, and after they immediately experienced the use of certain alternative form (NegConjunction) to express the "neither" meaning (corresponding to the NSD of NegDisjunction). The findings in these experiments reveal children's hidden grammatical knowledge of the WSD and highlight the role of pragmatic computation in the acquisition of meaning.

This dissertation is concerned with the problem of how structured linguistic representations interact with the architecture of human memory. Much recent work has attempted to unify real-time linguistic memory with a general content-addressable architecture (Lewis & Vasishth, 2005; McElree, 2006). Because grammatical principles and constraints are strongly relational in nature, and linguistic representation hierarchical, this kind of architecture is not well suited to restricting the search of memory to grammatically-licensed constituents alone. This dissertation investigates under what conditions real-time language comprehension is grammatically accurate. Two kinds of grammatical dependencies were examined in reading time and speeded grammaticality experiments: subject-verb agreement licensing in agreement attraction configurations ("The runners who the driver wave to ..."; Kimball & Aissen, 1971, Bock & Miller, 1991), and active completion of wh-dependencies. We develop a simple formal model of agreement attraction in an associative memory that makes accurate predictions across different structures. We conclude that dependencies that can only be licensed exclusively retrospectively, by searching the memory to generate candidate analyses, are the most prone to grammatical infidelity. The exception may be retrospective searches with especially strong contextual restrictions, as in reflexive anaphora. However dependencies that can be licensed principally by a prospective search, like wh-dependencies or backwards anaphora, are highly grammatically accurate.

An argument from resultative compounds in Igbo and Mandarin that direct objects bind a thematic relation introduced, not by either verb, but by their structural context.

Howard Lasnik reviews developments in the history of Case Theory.

This dissertation investigates the contributions of the parser and extra-linguistic information in the selection of a final interpretation of scopally ambiguous strings, integrating data from both children and adults into our understanding of language processing. Previous research has found an advantage for surface scope interpretations in adult sentence processing (Tunstall,1998 & Anderson,2003) and in children's interpretive preferences (Musolino and Lidz, 2006). In light of these findings, we investigate two central questions. One, what is the source of the advantage for surface scope interpretations in adults? Two, what factors contribute to children's ultimate adherence to surface scope interpretations? With respect to the first question, we show that the source of the advantage for surface scope interpretations cannot be described by a parsing preference, but can be described by the ease of the verification strategy utilized for the surface scope interpretation. With respect to the second question, we investigate children's interpretations of scopally ambiguous strings across a range of ages and find that while children appear fixed to surface scope interpretations during a limited window of development, this fixation does not hold at the earliest stage of development, demonstrating a U-shaped curve of development. Additionally, we find evidence that children's interpretations do not vary as a function of task in an adult-like way, and suggest that these findings must be explained by a combination of children's developing parsers and ability to integrate discourse information. We suggest that the non-adult-like interpretations observed in children derive from an initial parser bias for inverse scope interpretations, followed by a period in which children have adult-like parsers, but lack the the ability to integrate discourse information as a means to inform the process of ambiguity resolution.

The focus of this dissertation is syntactic movement and its relationship to surface semantics, morphology, and licensing relations in syntax, with an emphasis on Spanish and English. Chapter 2 argues that Herburger's (2000) Neo-Davidsonian approach to the semantics of focus, as syntactically implemented by Uriagereka (2005), allows for a unified treatment of new information focus and contrastive focus (focus movement to the left periphery and in situ focus) in Spanish. The diverse positions that the focused element can take in the sentence are claimed to be determined by contextual anchoring mechanisms of Raposo and Uriagereka (1995). This entails a remnant movement approach in cases of new information focus in Spanish (Ordóñez 2000). It is suggested that these processes take place covertly in English, contra Kayne (1998). Chapter 3 and Chapter 4 focus on the relationship between syntactic movement and surface semantics by looking at the syntax of preverbal subject in Spanish and English, respectively. According to Chomsky (2001, and subsequent work) and Uriagereka (2008) a.o., movement yields (at least) scopal and discourse-related properties. Movement to Spec,TP in so-called 'flexible word order' languages, like Spanish (contra Alexiadou and Anagnostopoulou 1998, a.o.), and in so-called 'strict' word order languages, like English, provides the testing ground for this hypothesis. It is argued here that both Spanish and English show surface semantics effects correlating with movement into Spec,TP, in keeping with the idea that syntactic movement has an effect on semantics. Chapter 5 explores a number of challenges for the phase-based system dispensing with grammatically significant Spec,H relations. It is proposed here that under a mixed system adopting phases and Long Distance Agreement and, crucially, a Multiple Spell-Out system (Uriagereka 1999), conceptual arguments against Spec,H relations can be circumvented. This is shown to solve a number of problems that the phase-based framework faces.

The best systems for machine translation of natural language are based on statistical models learned from data. Conventional representation of a statistical translation model requires substantial offline computation and representation in main memory. Therefore, the principal bottlenecks to the amount of data we can exploit and the complexity of models we can use are available memory and CPU time, and current state of the art already pushes these limits. With data size and model complexity continually increasing, a scalable solution to this problem is central to future improvement. Callison-Burch et al. (2005) and Zhang and Vogel (2005) proposed a solution that we call "translation by pattern matching", which we bring to fruition in this dissertation. The training data itself serves as a proxy to the model; rules and parameters are computed on demand. It achieves our desiderata of minimal offline computation and compact representation, but is dependent on fast pattern matching algorithms on text. They demonstrated its application to a common model based on the translation of contiguous substrings, but leave some open problems. Among these is a question: can this approach match the performance of conventional methods despite unavoidable differences that it induces in the model? We show how to answer this question affirmatively. The main open problem we address is much harder. Many translation models are based on the translation of discontiguous substrings. The best pattern matching algorithm for these models is much too slow, taking several minutes per sentence. We develop new algorithms that reduce empirical computation time by two orders of magnitude for these models, making translation by pattern matching widely applicable. We use these algorithms to build a model that is two orders of magnitude larger than the current state of the art and substantially outperforms a strong competitor in Chinese-English translation. We show that a conventional representation of this model would be impractical. Our experiments shed light on some interesting properties of the underlying model. The dissertation also includes the most comprehensive contemporary survey of statistical machine translation.