Theses and Dissertations from UMD
Permanent URI for this communityhttp://hdl.handle.net/1903/2
New submissions to the thesis/dissertation collections are added automatically as they are received from the Graduate School. Currently, the Graduate School deposits all theses and dissertations from a given semester after the official graduation date. This means that there may be up to a 4 month delay in the appearance of a give thesis/dissertation in DRUM
More information is available at Theses and Dissertations at University of Maryland Libraries.
Browse
64 results
Search Results
Item A Computational Theory of the Use-Mention Distinction in Natural Language(2011) Wilson, Shomir; Perlis, Donald R.; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)To understand the language we use, we sometimes must turn language on itself, and we do this through an understanding of the use-mention distinction. In particular, we are able to recognize mentioned language: that is, tokens (e.g., words, phrases, sentences, letters, symbols, sounds) produced to draw attention to linguistic properties that they possess. Evidence suggests that humans frequently employ the use-mention distinction, and we would be severely handicapped without it; mentioned language frequently occurs for the introduction of new words, attribution of statements, explanation of meaning, and assignment of names. Moreover, just as we benefit from mutual recognition of the use-mention distinction, the potential exists for us to benefit from language technologies that recognize it as well. With a better understanding of the use-mention distinction, applications can be built to extract valuable information from mentioned language, leading to better language learning materials, precise dictionary building tools, and highly adaptive computer dialogue systems. This dissertation presents the first computational study of how the use-mention distinction occurs in natural language, with a focus on occurrences of mentioned language. Three specific contributions are made. The first is a framework for identifying and analyzing instances of mentioned language, in an effort to reconcile elements of previous theoretical work for practical use. Definitions for mentioned language, metalanguage, and quotation have been formulated, and a procedural rubric has been constructed for labeling instances of mentioned language. The second is a sequence of three labeled corpora of mentioned language, containing delineated instances of the phenomenon. The corpora illustrate the variety of mentioned language, and they enable analysis of how the phenomenon relates to sentence structure. Using these corpora, inter-annotator agreement studies have quantified the concurrence of human readers in labeling the phenomenon. The third contribution is a method for identifying common forms of mentioned language in text, using patterns in metalanguage and sentence structure. Although the full breadth of the phenomenon is likely to elude computational tools for the foreseeable future, some specific, common rules for detecting and delineating mentioned language have been shown to perform well.Item A Formal Model of Ambiguity and its Applications in Machine Translation(2010) Dyer, Christopher; Resnik, Philip; Linguistics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Systems that process natural language must cope with and resolve ambiguity. In this dissertation, a model of language processing is advocated in which multiple inputs and multiple analyses of inputs are considered concurrently and a single analysis is only a last resort. Compared to conventional models, this approach can be understood as replacing single-element inputs and outputs with weighted sets of inputs and outputs. Although processing components must deal with sets (rather than individual elements), constraints are imposed on the elements of these sets, and the representations from existing models may be reused. However, to deal efficiently with large (or infinite) sets, compact representations of sets that share structure between elements, such as weighted finite-state transducers and synchronous context-free grammars, are necessary. These representations and algorithms for manipulating them are discussed in depth in depth. To establish the effectiveness and tractability of the proposed processing model, it is applied to several problems in machine translation. Starting with spoken language translation, it is shown that translating a set of transcription hypotheses yields better translations compared to a baseline in which a single (1-best) transcription hypothesis is selected and then translated, independent of the translation model formalism used. More subtle forms of ambiguity that arise even in text-only translation (such as decisions conventionally made during system development about how to preprocess text) are then discussed, and it is shown that the ambiguity-preserving paradigm can be employed in these cases as well, again leading to improved translation quality. A model for supervised learning that learns from training data where sets (rather than single elements) of correct labels are provided for each training instance and use it to learn a model of compound word segmentation is also introduced, which is used as a preprocessing step in machine translation.Item COMMITMENT AND FLEXIBILITY IN THE DEVELOPING PARSER(2010) Omaki, Akira; Phillips, Colin; Linguistics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This dissertation investigates adults and children's sentence processing mechanisms, with a special focus on how multiple levels of linguistic representation are incrementally computed in real time, and how this process affects the parser's ability to later revise its early commitments. Using cross-methodological and cross-linguistic investigations of long-distance dependency processing, this dissertation demonstrates how paying explicit attention to the procedures by which linguistic representations are computed is vital to understanding both adults' real time linguistic computation and children's reanalysis mechanisms. The first part of the dissertation uses time course evidence from self-paced reading and eye tracking studies (reading and visual world) to show that long-distance dependency processing can be decomposed into a sequence of syntactic and interpretive processes. First, the reading experiments provide evidence that suggests that filler-gap dependencies are constructed before verb information is accessed. Second, visual world experiments show that, in the absence of information that would allow hearers to predict verb content in advance, interpretive processes in filler-gap dependency computation take around 600ms. These results argue for a predictive model of sentence interpretation in which syntactic representations are computed in advance of interpretive processes. The second part of the dissertation capitalizes on this procedural account of filler-gap dependency processing, and reports cross-linguistic studies on children's long-distance dependency processing. Interpretation data from English and Japanese demonstrate that children actively associate a fronted wh-phrase with the first VP in the sentence, and successfully retract such active syntactic commitments when the lack of felicitous interpretation is signaled by verb information, but not when it is signaled by syntactic information. A comparison of the process of anaphor reconstruction in adults and children further suggests that verb-based thematic information is an effective revision cue for children. Finally, distributional analyses of wh-dependencies in child-directed speech are conducted to investigate how parsing constraints impact language acquisition. It is shown that the actual properties of the child parser can skew the input distribution, such that the effective distribution differs drastically from the input distribution seen from a researcher's perspective. This suggests that properties of developing perceptual mechanisms deserve more attention in language acquisition research.Item Relating Movement and Adjunction in Syntax and Semantics(2010) Hunter, Timothy; Weinberg, Amy; Linguistics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)In this thesis I explore the syntactic and semantic properties of movement and adjunction in natural language, and suggest that these two phenomena are related in a novel way. In a precise sense, the basic pieces of grammatical machinery that give rise to movement, also give rise to adjunction. In the system I propose, there is no atomic movement operation and no atomic adjunction operation; the terms "movement" and "adjunction" serve only as convenient labels for certain combinations of other, primitive operations. As a result the system makes non-trivial predictions about how movement and adjunction should interact, since we do not have the freedom to stipulate arbitrary properties of movement while leaving the properties of adjunction unchanged, or vice-versa. I focus first on the distinction between arguments and adjuncts, and propose that the differences between these two kinds of syntactic attachment can be thought of as a transparent reflection of the differing ways in which they contribute to neo-Davidsonian logical forms. The details of this proposal rely crucially on a distinctive treatment of movement, and from it I derive accurate predictions concerning the equivocal status of adjuncts as optionally included in or excluded from a maximal projection, and the possibility of counter-cyclic adjunction. The treatment of movement and adjunction as interrelated phenomena furthermore enables us to introduce a single constraint that subsumes two conditions on extraction, namely adjunct island effects and freezing effects. The novel conceptions of movement and semantic composition that underlie these results raise questions about the system's ability to handle semantic variable-binding. I give an unconventional but descriptively adequate account of basic quantificational phenomena, to demonstrate that this important empirical ground is not given up. More generally, this thesis constitutes a case study in (i) deriving explanations for syntactic patterns from a restrictive, independently motivated theory of compositional semantics, and (ii) using a computationally explicit framework to rigourously investigate the primitives and consequences of our theories. The emerging picture that is suggested is one where some central facts about the syntax and semantics of natural language hang together in a way that they otherwise would not.Item The importance of being a complement: CED effects revisited(2010) Jurka, Johannes; Hornstein, Norbert R; Linguistics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This dissertation revisits subject island effects (Ross 1967, Chomsky 1973) cross-linguistically. Controlled acceptability judgment studies in German, English, Japanese and Serbian suggest that extraction out of specifiers is consistently degraded compared to extraction out of complements, indicating that the Condition on Extraction domains (CED, Huang 1982) is still empirically viable, contrary to recent claims (Stepanov 2007). As a consequence, recent treatments of the CED in terms of Multiple Spell-Out (Uriagereka 1999) are still tenable. First, a series of NP-subextraction experiments in German using was für-split is discussed. The results indicate that subject island effects cannot be reduced to freezing effects (Wexler \& Culicover 1981). Extraction out of in-situ subjects is degraded compared to extraction out of in-situ objects. Freezing incurs an additional cost, i.e., extraction out of moved domains is degraded compared to in-situ domains. Further results from German indicate that extraction out of in-situ unaccusative and passive subjects is en par with extraction out of objects, while extraction out of in-situ transitive and intransitive unergative subjects causes a decrease in acceptability. Additionally, extraction out of indirect objects is degraded compared to extraction out of direct objects. It is also observed that a second gap improves the acceptability of otherwise illicit was für-split, a phenomenon dubbed Across-the-Board (ATB)-was für-splitand analysed in terms of Sideward Movement (Hornstein \& Nunes 2002). Furthermore, wh-extraction out of non-finite sentential arguments also shows a significant subject/object asymmetry. Experiments in English indicate that NP-subextraction yields the familiar subject/object asymmetry, while the contrast largely disappears when PPs are fronted. Further results show that ECM and passive predicates do not improve the acceptability of the extraction out of subjects. Finally, subject subextraction patterns in Japanese and Serbian are investigated. Both Long-distance scrambling and clefting out of sentential subjects in Japanese leads to a stronger degradation than out of sentential objects. PP-extraction in Serbian also shows the same subject/object asymmetry, while no such contrast is found for Left Branch Extraction.Item The Circle of Meaning: From Translation to Paraphrasing and Back(2010) Madnani, Nitin; Dorr, Bonnie; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The preservation of meaning between inputs and outputs is perhaps the most ambitious and, often, the most elusive goal of systems that attempt to process natural language. Nowhere is this goal of more obvious importance than for the tasks of machine translation and paraphrase generation. Preserving meaning between the input and the output is paramount for both, the monolingual vs bilingual distinction notwithstanding. In this thesis, I present a novel, symbiotic relationship between these two tasks that I term the "circle of meaning''. Today's statistical machine translation (SMT) systems require high quality human translations for parameter tuning, in addition to large bi-texts for learning the translation units. This parameter tuning usually involves generating translations at different points in the parameter space and obtaining feedback against human-authored reference translations as to how good the translations. This feedback then dictates what point in the parameter space should be explored next. To measure this feedback, it is generally considered wise to have multiple (usually 4) reference translations to avoid unfair penalization of translation hypotheses which could easily happen given the large number of ways in which a sentence can be translated from one language to another. However, this reliance on multiple reference translations creates a problem since they are labor intensive and expensive to obtain. Therefore, most current MT datasets only contain a single reference. This leads to the problem of reference sparsity---the primary open problem that I address in this dissertation---one that has a serious effect on the SMT parameter tuning process. Bannard and Callison-Burch (2005) were the first to provide a practical connection between phrase-based statistical machine translation and paraphrase generation. However, their technique is restricted to generating phrasal paraphrases. I build upon their approach and augment a phrasal paraphrase extractor into a sentential paraphraser with extremely broad coverage. The novelty in this augmentation lies in the further strengthening of the connection between statistical machine translation and paraphrase generation; whereas Bannard and Callison-Burch only relied on SMT machinery to extract phrasal paraphrase rules and stopped there, I take it a few steps further and build a full English-to-English SMT system. This system can, as expected, ``translate'' any English input sentence into a new English sentence with the same degree of meaning preservation that exists in a bilingual SMT system. In fact, being a state-of-the-art SMT system, it is able to generate n-best "translations" for any given input sentence. This sentential paraphraser, built almost entirely from existing SMT machinery, represents the first 180 degrees of the circle of meaning. To complete the circle, I describe a novel connection in the other direction. I claim that the sentential paraphraser, once built in this fashion, can provide a solution to the reference sparsity problem and, hence, be used to improve the performance a bilingual SMT system. I discuss two different instantiations of the sentential paraphraser and show several results that provide empirical validation for this connection.Item COMPUTATIONAL ANALYSIS OF THE CONVERSATIONAL DYNAMICS OF THE UNITED STATES SUPREME COURT(2009) Hawes, Timothy; Lin, Jimmy; Resnik, Philip; Linguistics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The decisions of the United States Supreme Court have far-reaching implications in American life. Using transcripts of Supreme Court oral arguments this work looks at the conversational dynamics of Supreme Court justices and links their conversational interaction with the decisions of the Court and individual justices. While several studies have looked at the relationship between oral arguments and case variables, to our knowledge, none have looked at the relationship between conversational dynamics and case outcomes. Working from this view, we show that the conversation of Supreme Court justices is both predictable and predictive. We aim to show that conversation during Supreme Court cases is patterned, this patterned conversation is associated with case outcomes, and that this association can be used to make predictions about case outcomes. We present three sets of experiments to accomplish this. The first examines the order of speakers during oral arguments as a patterned sequence, showing that cohesive elements in the discourse, along with references to individuals, provide significant improvements over our "bag-of-words" baseline in identifying speakers in sequence within a transcript. The second graphically examines the association between speaker turn-taking and case outcomes. The results presented with this experiment point to interesting and complex relationships between conversational interaction and case variables, such as justices' votes. The third experiment shows that this relationship can be used in the prediction of case outcomes with accuracy ranging from 62.5% to 76.8% for varying conditions. Finally, we offer recommendations for improved tools for legal researchers interested in the relationship between conversation during oral arguments and case outcomes, and suggestions for how these tools may be applied to more general problems.Item Apologies in French: An Analysis of Remedial Discourse Strategies Used by L1 Speakers(2009) Bodapati, Sandhya; Yotsukura, Lindsay; Modern French Studies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Speech act research has contributed much to our understanding of contextual L1 and L2 use in various languages. French, however remains largely ignored. The handful of studies that do exist are confined to a rather small set of speech acts. Although common in everyday discourse, French apologies have been underrepresented in the literature. This exploratory study attempts to observe and quantify apology strategies utilized by the French. Data were collected from L1 speakers in three phases. In Phase 1, 11 respondents provided conflict situations--used to construct a Discourse Completion Task (DCT)--that would require an apology in France. Twenty-two separate speakers completed a rating scale in Phase 2, stating their perceptions regarding sociolinguistic factors underlying the conflict situations. Finally in Phase 3, 85 respondents completed the DCT, which sought their reactions to the apology situations. Five main findings are discussed. First, L1 speakers most commonly used an explicit expression of apology or provided explanations as remedial strategies. This finding differs from previous studies on French L1 apologies in which accepting responsibility for the offense was the second most-used strategy after explicit apologies. Second, it was found that not all apology utterances performed a remedial function in all situations; certain linguistic formulae typically used to offer redress were also used as mitigators to potentially face-threatening acts such as complaining. Third, of several sociolinguistic factors operative within a situation, severity of the offense and the speaker's obligation to apologize seemed to have the most influence on apology strategy selection. Fourth, a survey of L1 speakers revealed that a majority felt it more important for an L2 speaker to be sociopragmatically competent in the target language than to demonstrate grammatical accuracy alone. Finally, the results suggest that the DCT continues to be a highly effective data elicitation instrument. In the present study, it not only facilitated quick access to a large data set, but it also allowed participants to make ancillary comments. Such insights might not have been revealed as readily through data collected in naturalistic settings through participant observation or role-plays--methods that have been deemed more reliable than the DCT.Item Grammatical Gender Representation and Processing in Advanced Second Language Learners of French(2009) Vatz, Karen L.; DeKeyser, Robert; Michael, Erica B; Second Language Acquisition and Application; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)One of the most difficult challenges of learning French as a foreign language is mastering the gender system. Although there are theories that account for how French native speakers (NSs) master their gender system, it is not fully understood why second language (L2) learners are unable to do the same. The goal of the present study was to investigate this difference in ability between French NSs and non-native speakers (NNSs), specifically, how L2 learners of French store grammatical gender knowledge, and how their storage system relates to processing of grammatical gender in terms of the ability to realize accurate gender agreement throughout a sentence. First, a gender priming task investigated whether advanced L2 learners have developed a gender-nodal system in which gender information is stored as an inherent property of a noun. Second, an online grammaticality judgment task addressed L2 learners' gender agreement ability during processing, while taking into account (a) the role of gender cues available to the participant, and (b) non-linguistic processing constraints such as working memory (WM) through manipulating the distance of an adjective from the noun with which it must agree. In order to investigate the role of a learner's native language (L1) in gender representation and processing, participants included learners of French from three L1 groups: Spanish, whose gender system is congruent to that of French; Dutch, whose gender system is incongruent to that of French; and English, whose gender system is minimal, relative to French. A group of NS controls also participated. Results from the gender priming task indicate that the NNSs in the current study have not developed a native-like gender-nodal system, regardless of L1-L2 gender-system similarity. At-chance accuracy on the grammaticality judgment task indicates L2 gender agreement is far from native-like, even for advanced learners. Whereas the presence of gender cues was beneficial, neither WM nor L1-L2 similarity facilitated performance. The results from this study confirm previous findings on the difficulty of L2 gender agreement, and shed light on the nature of L2 gender representation as a possible explanation for this processing difficulty.Item Fine-Grained Linguistic Soft Constraints on Statistical Natural Language Processing Models(2009) Marton, Yuval Yehezkel; Resnik, Philip; Linguistics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This dissertation focuses on effective combination of data-driven natural language processing (NLP) approaches with linguistic knowledge sources that are based on manual text annotation or word grouping according to semantic commonalities. I gainfully apply fine-grained linguistic soft constraints -- of syntactic or semantic nature -- on statistical NLP models, evaluated in end-to-end state-of-the-art statistical machine translation (SMT) systems. The introduction of semantic soft constraints involves intrinsic evaluation on word-pair similarity ranking tasks, extension from words to phrases, application in a novel distributional paraphrase generation technique, and an introduction of a generalized framework of which these soft semantic and syntactic constraints can be viewed as instances, and in which they can be potentially combined. Fine granularity is key in the successful combination of these soft constraints, in many cases. I show how to softly constrain SMT models by adding fine-grained weighted features, each preferring translation of only a specific syntactic constituent. Previous attempts using coarse-grained features yielded negative results. I also show how to softly constrain corpus-based semantic models of words (“distributional profiles”) to effectively create word-sense-aware models, by using semantic word grouping information found in a manually compiled thesaurus. Previous attempts, using hard constraints and resulting in aggregated, coarse-grained models, yielded lower gains. A novel paraphrase generation technique incorporating these soft semantic constraints is then also evaluated in a SMT system. This paraphrasing technique is based on the Distributional Hypothesis. The main advantage of this novel technique over current “pivoting” techniques for paraphrasing is the independence from parallel texts, which are a limited resource. The evaluation is done by augmenting translation models with paraphrase-based translation rules, where fine-grained scoring of paraphrase-based rules yields significantly higher gains. The model augmentation includes a novel semantic reinforcement component: In many cases there are alternative paths of generating a paraphrase-based translation rule. Each of these paths reinforces a dedicated score for the “goodness” of the new translation rule. This augmented score is then used as a soft constraint, in a weighted log-linear feature, letting the translation model learn how much to “trust” the paraphrase-based translation rules. The work reported here is the first to use distributional semantic similarity measures to improve performance of an end-to-end phrase-based SMT system. The unified framework for statistical NLP models with soft linguistic constraints enables, in principle, the combination of both semantic and syntactic constraints -- and potentially other constraints, too -- in a single SMT model.