No description available.

This thesis is about how our syntactic choice changes with linguistic experience. Studies on syntactic priming show that our decisions are influenced by sentences that we have recently heard or recently spoken. They also show that not all sentences have an equal amount of influence; that repetition of verbs increases priming (the lexical-boost effect) and that some verbs are more susceptible to priming than others. This thesis explores how and why syntactic decisions change with time and what these observations tell us about the cognitive mechanism of speaking. Specifically, we set out to develop a theoretical account of syntactic priming. Theoretical accounts require mathematical models and this thesis develops a sequence of mathematical models for understanding various aspects of syntactic priming. Cognitive processes are modelled as dynamical systems that can change their behaviour when they process information. We use these dynamical systems to investigate how each episode of language comprehension or production affects syntactic decisions. We also use these systems to investigate how long priming persists, how groups of consecutive sentences affect structural decisions, why repeating words leads to greater syntactic priming and what this tells us about how words, concepts and syntax are cognitively represented. We obtain two kinds of results by simulating these mathematical models. The first kind of results reveal how syntactic priming evolves over time. We find that structural priming itself shows a gradual decay with time but the lexical enhancement of priming decays catastrophically – a result consistent with experimental observations. We also find that consecutive episodes of language processing add up nonlinearly in memory, which challenges the design of some existing psycholinguistic experiments. The second kind of results reveal how our syntax module might be connected to other cognitive modules. We find that the lexical enhancement of syntactic priming might be a consequence of how the modules of attention and working memory influence syntactic decisions. These models suggest a mechanism of priming that is in contrast to a previous prediction-based account. This prediction-based account proposes that we actively predict what we hear and structural priming is due to error-correction whenever our predictions do not match the stimuli. In contrast, our account embodies syntactic priming in cognitive processes of attention, working memory and long-term memory. It asserts that our linguistic decisions are not based solely on abstract rules but also depend on the cognitive implementation of each module. Our investigations also contribute a novel theoretical framework for studying syntactic priming. Previous studies analyse priming using error-correction or Hebbian learning algorithms. We introduce the formalism of dynamical systems. This formalism allows us to trace the effect of information processing through time. It explains how residual activation from a previous episode might play a role in structural decisions, thereby enriching our understanding of syntactic priming. Since these dynamical systems are also used to model neural processes, this theoretical framework brings our understanding of priming one step closer to its biological implementation, bridging the gap between neural processes and abstract thoughts.

Grammar development over the last decades has seen a shift away from large inventories of grammar rules to richer lexical structures. Many modern grammar theories are highly lexicalised. But simply listing lexical entries typically results in an undesirable amount of redundancy. Lexical inheritance hierarchies, on the other hand, make it possible to capture linguistic generalisations and thereby reduce redundancy. Inheritance hierarchies are usually constructed by hand but this is time-consuming and often impractical if a lexicon is very large. Constructing hierarchies automatically or semiautomatically facilitates a more systematic analysis of the lexical data. In addition, lexical data is often extracted automatically from corpora and this is likely to increase over the coming years. Therefore it makes sense to go a step further and automate the hierarchical organisation of lexical data too. Previous approaches to automatic lexical inheritance hierarchy construction tended to focus on minimality criteria, aiming for hierarchies that minimised one or more criteria such as the number of path-value pairs, the number of nodes or the number of inheritance links (Petersen 2001, Barg 1996a, and in a slightly different context: Light 1994). Aiming for minimality is motivated by the fact that the conciseness of inheritance hierarchies is a main reason for their use. However, I will argue that there are several problems with minimality-based approaches. First, minimality is not well defined in the context of lexical inheritance hierarchies as there is a tension between different minimality criteria. Second, minimality-based approaches tend to underestimate the importance of linguistic plausibility. While such approaches start with a definition of minimal redundancy and then try to prove that this leads to plausible hierarchies, the approach suggested here takes the opposite direction. It starts with a manually built hierarchy to which a supervised machine learning algorithm is applied with the aim of finding a set of formal criteria that can guide the construction of plausible hierarchies. Taking this direction means that it is more likely that the selected criteria do in fact lead to plausible hierarchies. Using a machine learning technique also has the advantage that the set of criteria can be much larger than in hand-crafted definitions. Consequently, one can define conciseness in very broad terms, taking into account interdependencies in the data as well as simple minimality criteria. This leads to a more fine-grained model of hierarchy quality. In practice, the method proposed here consists of two components: Galois lattices are used to define the search space as the set of all generalisations over the input lexicon. Maximum entropy models which have been trained on a manually built hierarchy are then applied to the lattice of the input lexicon to distinguish between plausible and implausible generalisations based on the formal criteria that were found in the training step. An inheritance hierarchy is then derived by pruning implausible generalisations. The hierarchy is automatically evaluated by matching it to a manually built hierarchy for the input lexicon. Automatically constructing lexical hierarchies is a hard task, partly because what is considered the best hierarchy for a lexicon is to some extent subjective. Supervised learning methods also suffer from a lack of suitable training data. Hence, a semi-automatic architecture may be best suited for the task. Therefore, the performance of the system has been tested using a semi-automatic as well as an automatic architecture and it has also been compared to the performance achieved by the pruning algorithm suggested by Petersen (2001). The findings show that the method proposed here is well suited for semi-automatic hierarchy construction.

Making computers automatically find the appropriate meaning of words in context is an interesting problem that has proven to be one of the most challenging tasks in natural language processing (NLP). Widespread potential applications of a possible solution to the problem could be envisaged in several NLP tasks such as text simplification, language learning, machine translation, query expansion, information retrieval and text summarization. Ambiguity of words has always been a challenge in these applications, and the traditional endeavor to solve the problem of this ambiguity, namely doing word sense disambiguation using resources like WordNet, has been fraught with debate about the feasibility of the granularity that exists in WordNet senses. The recent trend has therefore been to move away from enforcing any given lexical resource upon automated systems from which to pick potential candidate senses,and to instead encourage them to pick and choose their own resources. Given a sentence with a target ambiguous word, an alternative solution consists of picking potential candidate substitutes for the target, filtering the list of the candidates to a much shorter list using various heuristics, and trying to match these system predictions against a human generated gold standard, with a view to ensuring that the meaning of the sentence does not change after the substitutions. This solution has manifested itself in the SemEval 2007 task of lexical substitution and the more recent SemEval 2010 task of cross-lingual lexical substitution (which I helped organize), where given an English context and a target word within that context, the systems are required to provide between one and ten appropriate substitutes (in English) or translations (in Spanish) for the target word. In this dissertation, I present a comprehensive overview of state-of-the-art research and describe new experiments to tackle the tasks of lexical substitution and cross-lingual lexical substitution. In particular I attempt to answer some research questions pertinent to the tasks, mostly focusing on completely unsupervised approaches. I present a new framework for unsupervised lexical substitution using graphs and centrality algorithms. An additional novelty in this approach is the use of directional similarity rather than the traditional, symmetric word similarity. Additionally, the thesis also explores the extension of the monolingual framework into a cross-lingual one, and examines how well this cross-lingual framework can work for the monolingual lexical substitution and cross-lingual lexical substitution tasks. A comprehensive set of comparative investigations are presented amongst supervised and unsupervised methods, several graph based methods, and the use of monolingual and multilingual information.

The aim of this thesis is to investigate lexical replacement processes from several complementary perspectives. It does so through three studies, each with a different scope and time depth. The first study (chapter 3) takes a high time depth perspective and investigates factors that affect the rate (likelihood) of lexical replacement in the core vocabulary of 98 Indo-European language varieties through a multiple linear regression model. The chapter shows that the following factors predict part of the rate of lexical replacement for non-grammatical concepts: frequency, the number of synonyms and senses, and how imageable the concept is in the mind. What looks like a straightforward lexical replacement at a high time depth perspective is better understood as several intertwined gradual processes of lexical change at lower time depths. The second study (chapter 5) narrows the focus to seven closely-related Germanic language varieties (English, German, Bernese, Danish, Swedish, Norwegian, and Icelandic) and a single semantic domain, namely color.  The chapter charts several lexical replacement and change processes in the pink and purple area of color space through experiments with 146 speakers. The third study (chapter 6) narrows the focus even more, to two generations of speakers of a single language, Swedish. It combines experimental data on how the two age groups partition and label the color space in general, and pink and purple in particular, with more detailed data on lexical replacement and change from interviews, color descriptions in historical and contemporary dictionaries, as well as botanical lexicons, and historical fiction corpora. This thesis makes a descriptive, methodological and theoretical contribution to the study of lexical replacement. Taken together, the different perspectives highlight the usefulness of method triangulation in approaching the complex phenomenon of lexical replacement.

This thesis considers the automatic acquisition of knowledge about discourse connectives. It focuses in particular on their semantic properties, and on the relationships that hold between them. There is a considerable body of theoretical and empirical work on discourse connectives. For example, Knott (1996) motivates a taxonomy of discourse connectives based on relationships between them, such as HYPONYMY and EXCLUSIVE, which are defined in terms of substitution tests. Such work requires either great theoretical insight or manual analysis of large quantities of data. As a result, to date no manual classification of English discourse connectives has achieved complete coverage. For example, Knott gives relationships between only about 18% of pairs obtained from a list of 350 discourse connectives. This thesis explores the possibility of classifying discourse connectives automatically, based on their distributions in texts. This thesis demonstrates that state-of-the-art techniques in lexical acquisition can successfully be applied to acquiring information about discourse connectives. Central to this thesis is the hypothesis that distributional similarity correlates positively with semantic similarity. Support for this hypothesis has previously been found for word classes such as nouns and verbs (Miller and Charles, 1991; Resnik and Diab, 2000, for example), but there has been little exploration of the degree to which it also holds for discourse connectives. We investigate the hypothesis through a number of machine learning experiments. These experiments all use unsupervised learning techniques, in the sense that they do not require any manually annotated data, although they do make use of an automatic parser. First, we show that a range of semantic properties of discourse connectives, such as polarity and veridicality (whether or not the semantics of a connective involves some underlying negation, and whether the connective implies the truth of its arguments, respectively), can be acquired automatically with a high degree of accuracy. Second, we consider the tasks of predicting the similarity and substitutability of pairs of discourse connectives. To assist in this, we introduce a novel information theoretic function based on variance that, in combination with distributional similarity, is useful for learning such relationships. Third, we attempt to automatically construct taxonomies of discourse connectives capturing substitutability relationships. We introduce a probability model of taxonomies, and show that this can improve accuracy on learning substitutability relationships. Finally, we develop an algorithm for automatically constructing or extending such taxonomies which uses beam search to help find the optimal taxonomy.

This dissertation explores the relationship between English phonotactics – sequential dependencies between adjacent segments – and the metrical parse, which relies on the division of words into syllables. Most current theories of syllabification operate under the assumption that the phonotactic restrictions which co-determine syllable boundaries are constrained by word edges. For example, a syllable can never begin with a consonant sequence that is not also attested as a word onset. This view of phonotactics as categorical is outdated: for several decades now, psycholinguistic research employing monosyllables has shown that phonotactic knowledge is gradient, and that this gradience is projected from the lexicon and possibly also based on differences in sonority among consonants located at word margins. This dissertation is an attempt to reconcile syllabification theory with this modern view of phonotactics. In what follows, I propose and defend a gradient metrical parsing model which assigns English syllable boundaries as a probabilistic function of the well-formedness relations that obtain between potential syllable onsets and offsets. I argue that this well-formedness is subserved by the same sources already established in the phonotactic literature: probabilistic generalizations over the word edges as well as sonority. In support of my proposal, I provide experimental evidence from five sources: (1) a pseudoword hyphenation experiment, (2) a reanalysis of a well-known, large-scale hyphenation study using real English words, (3) a forced-choice preference task employing nonwords presented as minimal stress pairs, (4) an online stress assignment experiment, and (5) a study of the speech errors committed by the participants of (4). The results of all studies converge in support of the gradient parsing model and correlate significantly with each other. Subsequent computer simulations suggest that the gradient model is preferred to the categorical alternative throughout all stages of lexical acquisition. This dissertation contains co-authored material accepted for publication.

In recent years the role of offline consolidation in supporting word learning has attracted great interest and has provided valuable insight into how novel spoken and written words are learned. Relatively little attention has focused on whether offline consolidation supports the learning and generalisation of novel orthographic knowledge. Meanwhile, laboratory-based approaches have proven valuable in overcoming the methodological challenges of studying reading acquisition, i.e. learning letter-sound knowledge. This thesis combines laboratory-based orthographic learning with an overnight consolidation framework to track the effects of sleep on learning novel letters and novel written words in six experiments. Experiment 1 validated the artificial orthography paradigm by using fMRI to show the novel orthography activated similar neural regions to pseudowords written in familiar orthography. Comparing recently learned words and objects additionally highlighted the componential and holistic processes that distinguish reading from object naming. Experiments 2, 3, and 4 investigated whether overnight consolidation had contrasting effects on learning novel letters and learning novel written words. All three studies showed overnight improvements in the ability to use and generalise knowledge of letters. Experiment 3 further assessed whether consolidation supported the formation of bigram representations. While the results did not show bigram consolidation, a recognition memory task indicated participants had consolidated the novel spoken words. Experiment 4 manipulated the internal statistical structure of the novel words finding, in contrast to Experiment 3, participants had consolidated the written forms of the novel words. Experiments 5 and 6 asked whether consolidated and unconsolidated spoken words would support orthographic learning. These studies failed to observe previous findings of spoken word consolidation and did not demonstrate clear effects of lexical knowledge on orthographic learning. The findings of the thesis demonstrate the importance of letter-level learning and consolidation during reading acquisition as well as highlighting the value of laboratory-based studies for understanding the interdependent trajectories of the skills involved in reading.

The present study examines the nature of bilingual lexical access using category fluency across five language combinations using 109 healthy speakers of Hindi-English, Kannada-English, Mandarin-English, Spanish-English, and Turkish-English. Participants completed a category fluency task in each of their languages in three main categories (animals, clothing, food), each with two subcategories, as well as a language use questionnaire assessing their proficiency in each of their languages. Multivariate analyses of variance revealed that the average number of correct items named in the category fluency task across the three main categories varied across the different groups for English items only. A series of repeated-measures analyses of covariance revealed that the exposure component that had been extracted from the language use questionnaire using a principal component analysis significantly affected the average number of items named across the three main categories. When the effect of exposure was controlled, the effect of language combination was no longer significant. A regression analysis showed that the relative amount of exposure participants had to each of their languages predicted participants’ relative performance in each language. Additional multivariate analyses of variance found significant differences in the number of correct items named in each main category and subcategory in both English and participants’ other language based on language combination. Overall, these results demonstrate the effects of particular language combinations on bilingual lexical access and provide important insights into the role of proficiency on access.

Theories of lexical storage differ in how entries are encoded in the lexicon. Exemplar-based accounts posit that lexical items are stored with detailed acoustic information, while abstract accounts argue that fine acoustic detail is removed and an item is stored in more basic phonological units. These separate accounts make distinct predictions about cross-linguistic and bilingual perception. Studies asking participants to compare non-native vowels have shown that people tend to associate multiple non-native phonemes to a single L1 phoneme when the contrast between the two does not exist in the L1. However, several studies have shown that the ability to discriminate sounds is never lost. A line of research has focused on how bilinguals perceive contrasts in their second language. One such study, Pallier et al. (2001) looked at early bilinguals of Spanish and Catalan, testing whether the native Spanish speakers, who were highly proficient in Catalan, perceived certain Catalan minimal pairs as homophones. Importantly, the contrasts of these minimal pairs were exclusive to Catalan. The native Spanish bilinguals heard pairs such as /neta/-/nεta/ in an audio-only lexical decision task (LDT), and showed responses to the second item that were not significantly different from actual item repetitions (i.e., /neta/-/neta/). These results were taken as evidence in favor of abstractionist models of lexical storage. This study was based on Pallier et al, (2001), examining instead the perceptions of heritage speakers of Spanish (HSSs) in the U.S., children of native Spanish speakers who get early and sustained exposure to their second language, English. Unlike the bilinguals studied in Pallier et al., heritage bilinguals receive little linguistic or social support for development of their first language. The L1 proficiency of adult heritage bilinguals varies considerably. In this study, a group of these HSSs participated in an LDT testing their perception of English-exclusive phonemic vowel contrasts (i.e., peak-pick). It was hypothesized that, like Pallier et al.'s highly proficient bilinguals, HSSs would show responses to the second item of these minimal pairs as if it were a repetition of the first. Results of the LDT did not confirm the hypothesis. The heritage Spanish speakers did not perform significantly differently from the native English controls on English-specific contrasts (p = .065), and it was found that the native English speakers showed higher priming on these minimal pairs than HSSs. These results run counter to those of previous studies, and may disfavor an abstract account of lexical storage. At the very least, the construct validity of this methodology is questionable when the control and experimental participants reverse hypothesized behavior.