Papers reviewed and determined not to be word norm studies. Use the flag icon to report errors or suggest re-inclusion.
18625 papers
Abstract. In this paper we explore an unsupervised approach to classify video content by analyzing the corresponding subtitles. The proposed method is based on the WordNet lexical database and the WordNet domains and applies natural language processing techniques on video subtitles. The method is divided into several steps. The first step includes subtitle text preprocessing. During the next steps, a keyword extraction method and a word sense disambiguation technique are applied. Subsequently, the WordNet domains that correspond to the correct word senses are identified. The final step assigns category labels to the video content based on the extracted domains. Experimental results with documentary videos show that the proposed method is quite effective in discovering the correct category for each video.
This study compared two technical sign movie formats, Quicktime (2D) and threedimentional animations (3D) and examined recall, comprehension, and rehearsal time of signs. There was no significant difference found between the formats for comprehension and recall. However, for rehearsal the high school students had significantly higher sign production scores after viewing 2D signs as comparaed to 3D signs. In addition, there was a significant difference found relating to gender. Girls scored significantly higher on all three measures, regardless of the movie format. In addition, students with more than two years experience with ASL scored significantly higher than those with less than two years on both recall and comprehension using both movie formats. Implications of these findings will be discussed in this paper.
<h3>Introduction</h3> Natural language applications like machine translation, question answering, and summarization currently are forced to depend on impoverished text models like bags of words or n-grams, while the decisions that they are making ought to be based on the meanings of those words in context. That lack of semantics causes problems throughout the applications. Misinterpreting the meaning of an ambiguous word results in failing to extract data, incorrect alignments for translation, and ambiguous language models. Incorrect coreference resolution results in missed information (because a connection is not made) or incorrectly conflated information (due to false connections). Some richer semantic representation is badly needed. The OntoNotes project is a collaborative effort between BBN Technologies, the University of Colorado, the University of Pennsylvania, and the University of Southern California's Information Sciences Institute to produce such a resource. It aims to annotate a large corpus comprising various genres of text (news, conversational telephone speech, weblogs, use net, broadcast, talk shows) in three languages (English, Chinese, and Arabic) with structural information (syntax and predicate argument structure) and shallow semantics (word sense linked to an ontology and coreference). OntoNotes builds on two time-tested resources, following the Penn Treebank for syntax and the Penn PropBank for predicate-argument structure. Its semantic representation will include word sense disambiguation for nouns and verbs, with each word sense connected to an ontology, and coreference. The current goals call for annotation of over a million words each of English and Chinese, and half a million words of Arabic over five years. The authors wish to make this resource available to the natural language research community so that decoders for these phenomena can be trained to generate the same structure in new documents. Lessons learned over the years have shown that the quality of annotation is crucial if it is going to be used for training machine learning algorithms. Taking this cue, we ensure that each layer of annotation in OntoNotes will have at least 90% inter- annotator agreement. Our pilot studies have shown that predicate structure, word sense, ontology linking, and coreference can all be annotated rapidly and with better than 90% consistency. <h3>Samples</h3> The following screen captures provide examples of the data contained in this corpus. <ul> <li> <a href="./desc/addenda/LDC2007T21_eng_tbk.jpg" rel="nofollow">English tree</a>. </li> <li> <a href="./desc/addenda/LDC2007T21_sense_pred.jpg" rel="nofollow">English sense predicate structure</a>. </li> <li> <a href="./desc/addenda/LDC2007T21_chi_comp.jpg" rel="nofollow">Chinese tree and sense predicate structure</a>. </li> </ul><h3>Sponsorship</h3> This work was suppported in part by the Defense Research Advanced Projects Agency, GALE Program Grant No. HR0011-06-C-0022. The content of this publication does not necessarily reflect the position or policy of the Government, and no official endorsement should be inferred. </br> Portions © 1989 Dow Jones & Company, Inc., © 1996-2001 Sinorama Magazine, © 1994-1998 Xinhua News Agency, © 1995, 2005, 2006, 2007 Trustees of the University of Pennsylvania
One may need to build a statistical parser for a new language, using only a very small labeled treebank together with raw text. We argue that bootstrapping a parser is most promising when the model uses a rich set of redundant features, as in recent models for scoring dependency parses (McDonald et al., 2005). Drawing on Abney’s (2004) analysis of the Yarowsky algorithm, we perform bootstrapping by entropy regularization: we maximize a linear combination of conditional likelihood on labeled data and confidence (negative Rényi entropy) on unlabeled data. In initial experiments, this surpassed EM for training a simple feature-poor generative model, and also improved the performance of a feature-rich, conditionally estimated model where EM could not easily have been applied. For our models and training sets, more peaked measures of confidence, measured by Rényi entropy, outperformed smoother ones. We discuss how our feature set could be extended with cross-lingual or cross-domain features, to incorporate knowledge from parallel or comparable corpora during bootstrapping. 1
We aim to improve the performance of a syntactic parser that uses a part-of-speech (POS) tagger as a preprocessor. Pipelined parsers consisting of POS taggers and syntactic parsers have several advantages, such as the capability of domain adaptation. However the performance of such systems on raw texts tends to be disappointing as they are affected by the errors of automatic POS tagging. We attempt to compensate for the decrease in accuracy caused by automatic taggers by allowing the taggers to output multiple answers when the tags cannot be determined reliably enough. We empirically verify the effectiveness of the method using an HPSG parser trained on the Penn Treebank. Our results show that ambiguous POS tagging improves parsing if outputs of taggers are weighted by probability values, and the results support previous studies with similar intentions. We also examine the effectiveness of our method for adapting the parser to the GENIA corpus and show that the use of ambiguous POS taggers can help development of portable parsers while keeping accuracy high. 1
Due to the data sparseness problem, the lexical information from a treebank for a lexicalized parser could be insufficient. This paper proposes an approach to learn head-modifier pairs from a raw corpus, and to integrate them into a lexicalized dependency parser to parse a Chinese Treebank. Experimental re-sults show that this approach not only enlarged the coverage of bi-lexical de-pendency, but also improved the accuracy of dependency parsing significantly.
This paper describes practical issues in the framework-independent evaluation of deep and shallow parsers. We focus on the use of two dependencybased syntactic representation formats in parser evaluation, namely, Carroll et al. (1998)’s Grammatical Relations and de Marneffe et al. (2006)’s Stanford Dependency scheme. Our approach is to convert the output of parsers into these two formats, and measure the accuracy of the resulting converted output. Through the evaluation of an HPSG parser and Penn Treebank phrase structure parsers, we found that mapping between different representation schemes is a non-trivial task that results in lossy conversions that may obscure important differences between different parsing approaches. We discuss sources of disagreements in the representation of syntactic structures in the two dependency-based formats, indicating possible directions for improved framework-independent parser evaluation.
The authors examine personality variables and interview format as potential antecedents of impression management (IM) behaviors in simulated selection interviews. The means by which these variables affect ratings of interview performance is also investigated. The altruism facet of agreeableness predicted defensive IM behaviors, the vulnerability facet of emotional stability predicted self- and other-focused behaviors, and interview format (behavior description vs. situational questions) predicted self-focused and defensive behaviors. Consistent with theory and research on situational strength, antecedent—IM relations were consistently weaker in a strong situation in which interviewees had an incentive to manage their impressions. There was also evidence that IM partially mediated the effects of personality and interview format on interview performance in the weak situation.
This paper explores a parsimonious approach to Data-Oriented Parsing. While allowing, in principle, all possible subtrees of trees in the treebank to be productive elements, our approach aims at nding a manageable subset of these trees that can accurately describe empirical distributions over phrase-structure trees. The proposed algorithm leads to computationally much more tracktable parsers, as well as linguistically more informative grammars. The parser is evaluated on the OVIS and WSJ corpora, and shows improvements on efciency, parse accuracy and testset likelihood. 1 Data-Oriented Parsing Data-Oriented Parsing (DOP) is a framework for statistical parsing and language modeling originally proposed by Scha (1990). Some of its innovations, although radical at the time, are now widely accepted: the use of fragments from the trees in an annotated corpus as the symbolic grammar (now known as itreebank grammarsi, Charniak, 1996) and inclusion of all statistical dependencies between nodes in the trees for disambiguation (the iallsubtrees approachi, Collins & Duffy, 2002). The best known instantiations of the DOPframework are due to Bod (1998; 2001; 2003), using the Probabilistic Tree Substitution Grammar (PTSG) formalism. Bod has advocated a maximalist approach to DOP, inducing grammars that contain all subtrees of all parse trees in the treebank, and using them to parse unknown sentences where all of these subtrees can potentially contribute to the most probable parse. Although Bod’s empirical results have been excellent, his maximalism poses important computational challenges that, although not necessarily unsolvable, threaten both the scalability to larger treebanks and the cognitive plausibility of the models. In this paper I explore a different approach to DOP, that I will call iParsimonious Data-Oriented Parsingi (P-DOP). This approach remains true to Scha’s original program, by allowing, in principle, all possible subtrees of trees in the treebank to be the productive elements. But unlike Bod’s approach, P-DOP aims at nding a succinct subset of such elementary trees, chosen such that it can still accurately describe observed distributions over phrasestructure trees. I will demonstrate that P-DOP leads to computationally more tracktable parsers, as well as linguistically more informative grammars. Moreover, as P-DOP is formulated as an enrichment of the treebank Probabilistic Context-free Grammar (PCFG), it allows for much easier comparison to alternative approaches to statistical parsing (Collins, 1997; Charniak, 1997; Johnson, 1998; Klein and Manning, 2003; Petrov et al., 2006).
We describe some challenges of adaptation in the 2007 CoNLL Shared Task on Domain Adaptation. Our error analysis for this task suggests that a primary source of error is differences in annotation guidelines between treebanks. Our suspicions are supported by the observation that no team was able to improve target domain performance substantially over a state of the art baseline. 1
We study the correlations in the connectivity patterns of large scale syntactic dependency networks. These networks are induced from treebanks: their vertices denote word forms which occur as nuclei of dependency trees. Their edges connect pairs of vertices if at least two instance nuclei of these vertices are linked in the dependency structure of a sentence. We examine the syntactic dependency networks of seven languages. In all these cases, we consistently obtain three findings. Firstly, clustering, i.e., the probability that two vertices which are linked to a common vertex are linked on their part, is much higher than expected by chance. Secondly, the mean clustering of vertices decreases with their degree — this finding suggests the presence of a hierarchical network organization. Thirdly, the mean degree of the nearest neighbors of a vertex x tends to decrease as the degree of x grows—this finding indicates disassortative mixing in the sense that links tend to connect vertices of dissimilar degrees. Our results indicate the existence of common patterns in the large scale organization of syntactic dependency networks.
We compare the accuracy of a statistical parse ranking model trained from a fully-annotated portion of the Susanne treebank with one trained from unlabeled partially-bracketed sentences derived from this treebank and from the Penn Treebank. We demonstrate that confidence-based semi-supervised techniques similar to self-training outperform expectation maximization when both are constrained by partial bracketing. Both methods based on partially-bracketed training data outperform the fully supervised technique, and both can, in principle, be applied to any statistical parser whose output is consistent with such partial-bracketing. We also explore tuning the model to a different domain and the effect of in-domain data in the semi-supervised training processes.
Visual stimuli are judged for their emotional significance based on two fundamental dimensions, valence and arousal, and may lead to changes in neural and body functions like attention, affect, memory and heart rate. Alterations in behaviour and mood have been encountered in patients with Parkinson's disease (PD) undergoing functional neurosurgery, suggesting that electrical high-frequency stimulation of the subthalamic nucleus (STN) may interfere with emotional information processing. Here, we use the opportunity to directly record neuronal activity from the STN macroelectrodes in patients with PD during presentation of emotionally laden and neutral pictures taken from the International Affective Picture System (IAPS) to further elucidate the role of the STN in emotional processing. We found a significant event-related desynchronization of STN alpha activity with pleasant stimuli that correlated with the individual valence rating of the pictures. Our findings suggest involvement of the human STN in valence-related emotional information processing that can potentially be altered during high-frequency stimulation of the STN in PD leading to behavioural complications.
Application of egalitarian and prioritarian accounts of health resource allocation in low-income countries have both been criticized for implying distribution outcomes that allow decreasing/undermining health gains and for tolerating unacceptable standards of health care and health status that result from such allocation schemes. Insufficient health care and severe deprivation of health resources are difficult to accept even when justified by aggregative efficiency or legitimized by fair deliberative process in pursuing equality and priority oriented outcomes. I affirm the sufficientarian argument that, given extreme scarcity of public health resources in low-income countries, neither health status equality between populations nor priority for the worse off is normatively adequate. Nevertheless, the threshold norm alone need not be the sole consideration when a country's total health budget is extremely scarce. Threshold considerations are necessary in developing a theory of fair distribution of health resources that is sensitive to the lexically prior norm of sufficiency. Based on the intuition that shares must not be taken away from those who barely achieve a minimal level of health, I argue that assessments based on standards of minimal physical/mental health must be developed to evaluate the sufficiency of the total resources of health systems in low-income countries prior to pursuing equality, priority, and efficiency based resource allocation. I also begin to examine how threshold sensitive health resource assessment could be used in the Philippines.
The task we are investigating is unsupervised learning of natural language morphology for inflectional languages. The target morphological grammar consists of a lexicon of morphological base forms and transforms. A base form represents all inflections of a lexeme, and all base forms of the same POS category share the same fine-grained morphosyntactic type. Transforms are morphophonemic rewrite rules that convert base forms to derived forms, and whose context of application is limited to a specific set of base forms. We have developed a greedy algorithm to induce such a grammar. At each iteration, suffixal transforms to convert between base and derived forms of lexemes are hypothesized. The algorithm chooses the transform that maximizes vocabulary coverage, while minimizing the number of conflicts resulting from proposing as base forms words previously found to be derived forms. After base forms and transforms have been learned, a distributional clustering step assigns the base forms to POS classes. In future work, the transforms will be converted to generalized rewrite rules by inducing phonological characteristics common to the base forms. We have tested this algorithm on a version of the Penn Treebank annotated for inflectional morphology. The algorithm achieves 71.7% recall and 92.9% precision on inflectional relations, where both a base and derived form occur in the corpus. We are currently testing the algorithm on other languages, and will present results on the Morphochallenge gold standards.
This article discusses irony, parody, and satire in the oral-literary tradition of Finnish working-class youth during the first decades of the twentieth century. The most important research material is Valistaja ( The Enlightener, 1914–1925), a handwritten newspaper produced by young working-class people in the industrial town of Karkkila in southern Finland. This research material provides examples of different kinds of parody: ideological parody is directed against both political opponents and the texts representing their ideology; generic parody involves playing with linguistic norms and generic conventions. Parody and satire provided means for exposing the cruelty and cowardice of the anti-Bolshevik Whites, the hypocrisy of the Church, and the conservatism of the older generation of workers. The ironic expressions reflect the experiences and tensions among groups of young people in Karkkila.
GeometryNet can be considered as a lexical database for geometric entities and concepts. The idea is borrowed from WordNet, a popular knowledge repository often used for natural language processing tasks in AI applications. The basic objective behind the construction of a GeometryNet is to analyze and understand the geometric problems and draw relevant geometric figures automatically. Initial emphasis is put on machine understanding of problem statements that involve geometric constructions. School level geometry problems practiced by students of age group 13-16 are targeted. This paper explains different aspects of a GeometryNet, issues behind its construction, and its possible use for machine understanding of geometric problem statements.
Students learned teaching principles either with or without (control group) the presentation of a classroom exemplar in video or text format. Across 2 experiments, the video group produced higher transfer scores and affective ratings than the other groups. Four weeks later, the video group recalled more information about the exemplar than the text group, but no treatment effects were found on transfer. Qualitative analyses (Experiment 2) showed that the video group produced a significantly larger number of modeled behaviors in the transfer test than the text (immediate) and control (immediate and delayed) groups. Results encourage using classroom video exemplars to promote students ’ affect and retention, but suggest that additional pedagogies are needed to promote longer term transfer of theory into practice.
Proceedings of the Sixth International Workshop on Treebanks and \nLinguistic Theories. \nEditors: Koenraad De Smedt, Jan Hajič and Sandra Kübler. \nNEALT Proceedings Series, Vol. 1 (2007), 61-72. \n© 2007 The editors and contributors. \nPublished by \nNorthern European Association for Language \nTechnology (NEALT) \nhttp://omilia.uio.no/nealt. \nElectronically published at \nTartu University Library (Estonia) \nhttp://hdl.handle.net/10062/4476.
In 1993 the Ministers of Education in the Netherlands and Flanders decided to install a binational committee of experts in order to co-ordinate, streamline, improve and stimulate the production of bilingual dictionaries and lexical databases with Dutch as a source or target language. This committee, called Commissie voor Lexicografische Vertaalvoorzieningen (Committee for Interlingual Lexicographical Resources) or CLVV, has, under the presidency of W. Martin, set up Action Plans involving some twenty dictionary projects which have been finished or are nearly finished by now. In this article the general policy lines of the CLVV are presented next to the criteria for the selection of language pairs, the infrastructure used, the results obtained and the lessons to be drawn from this ‘Dutch’ approach. The article also serves as a framework in which to situate the articles that follow.
One of the goals of natural language processing (NLP) systems is determining the meaning of what is being transmitted. Although much work has been accomplished in traditional written and spoken language domains, little has been performed in the newer computer-mediated communication domain enabled by the Internet, to include text-based chat. This is due in part to the fact that there are no annotated chat corpora available to the broader research community. The purpose of our research is to build a chat corpus, initially tagged with lexical and discourse information. Such a corpus could be used to develop stochastic NLP applications that perform tasks such as conversation thread topic detection, author profiling, entity identification, and social network analysis. During the course of our research, we preserved 477,835 chat posts and associated user profiles in an XML format for future investigation. We privacy-masked 10,567 of those posts and part-of-speech tagged a total of 45,068 tokens. Using the Penn Treebank and annotated chat data, we achieved part-of-speech tagging accuracy of 90.8%. We also annotated each of the privacy-masked corpus's 10,567 posts with a chat dialog act. Using a neural network with 23 input features, we achieved 83.2% dialog act classification accuracy.
Traditional self-report measures suffer from weaknesses in either the quantitative or qualitative assessment of subjective experience. Researchers interested in the subjective intensity of oral sensation have attempted to reduce these scale limitations by developing rating scales with empirically determined placement of verbal descriptors along a continuous visual analogue scale continuum. In the present research, a similar empirical approach to scale construction was adopted to develop a rating scale of emotional valence. The potential benefits of using an empirically derived valence scale and techniques for validating the scale are discussed.
This paper describes an automatic prediction model of Chinese prepositional phrase boundary location based on HMM.It consists of two stages: automatically identify the phrase boundary using statistics from treebank,then,post-tune the results with dependency grammar knowledge generated by dependency treebank.Experimental results demonstrate a high rate of success for predicting boundary location(86.5% correct rate for close testing and 77.7% for open testing).
By raising the question of folk linguistics in the discussion of linguistic phenomena, it is possible to abandon the typically binary approach to norms (descriptive versus prescriptive), and to propose a third term: perception. In France, scientific truth is still defined along very Cartesian lines, and linguistic norms remain monopolized by purism and academic orthodoxy. As a result, perceptive norms has not been allowed to emerge as a field of research. And yet, data drawn from non-academic linguistic discourse could make a considerable contribution to work on linguistic norms and varieties of language. It proves worthy of study not as a naive, pre-scientific effort, but as one possible theory of language—thus causing us to rethink the way the rules of the system actually function, beyond their representations. Intuition, urbane linguistics, folk linguistics, proper noun, perception, folk sociolinguistics.
Following social identity theory, this research examines the relationship between group identification and intergroup relations by focusing on religion. Religious as well as Dutch national group identification was examined among Turkish-Dutch Muslims. Identification was studied in relation to general affective ratings of multiple religious groups (Muslims, Christians, Hindustanis, Jews and non-believers) and the endorsement of Islamic group rights. The results show that Muslim identification was more like a nominal (high or `total' identification) than a continuous variable, and that many participants showed low identification with the national group. The affective ratings of religious out-groups were quite negative, particularly of the Jews and non-believers. Muslim identification was positively and strongly related to feelings toward the religious in-group and to the endorsement of Islamic group rights. National identification was positively related to feelings toward the religious out-groups, but only for `total' Muslim identifiers, supporting the mutual intergroup differentiation model. The findings are discussed in relation to social psychological thinking about group identification and the importance of religion for intergroup relations.
Automatic syllabification of words is challenging, not least because the syllable is difficult to define precisely. This task is important for word modelling in the composition process of cocatenative synthesis as well as in automatic speech recognition. There are two broad approaches to perform automatic syllabification: rule-based and data-driven. The rule-based method effectively embodies some theoretical position regarding the syllable, whereas the data-driven paradigm infers ‘new’ syllabifications from examples assumed to be correctly-syllabified already. This paper compares the performance of the two basic approaches. However, it is difficult to determine a correct syllabification in all cases and so to establish the quality of the ‘gold standard’ corpus used either to quantitatively evaluate the output of an automatic algorithm or as the example-set on which data-driven methods crucially depend. Thus, three lexical databases of pre-syllabified words were used. Two of these lexicons hold the same 18,016 words with their corresponding syllabifications coming from independent sources, whereas the third corresponds to the 13,594 words that share the same syl-labifications according to these two sources. As well as one rule-based approach (Fisher’s implementation of Kahn’s syl-labification theory), three data-driven techniques are evaluated: a look-up procedure, an exemplar-based generalization tech-nique, and syllabification by analogy (SbA). The results on the three databases show consistent and robust patterns: the data-driven techniques outperform the rule-based system in word and juncture accuracies by a very significant margin and best results are obtained with SbA.
This paper investigates probability distributions of dependency distances in six texts ex- tracted from a Chinese dependency treebank. The fitting results reveal that the investigated distribu- tion can be well captured by the right truncated Zeta distribution. In order to restrict the model only to natural language, two samples with randomly generated governors are investigated. One of them can be described e.g. by the Hyperpoisson distribution, the other satisfies the Zeta distribution. The paper also presents a study on sequential plot and mean dependency distance of six texts with three analyses (syntactic, and two random). Of these three analyses, syntactic analysis has a minimum (mean) dependency distance.
The ability to detect similarity in conjunct heads is potentially a useful tool in helping to disambiguate coordination structures - a difficult task for parsers. We propose a distributional measure of similarity designed for such a task. We then compare several different measures of word similarity by testing whether they can empirically detect similarity in the head nouns of noun phrase conjuncts in the Wall Street Journal (WSJ) treebank. We demonstrate that several measures of word similarity can successfully detect conjunct head similarity and suggest that the measure proposed in this paper is the most appropriate for this task.
There are many methods to improve performance of statistical parsers. Resolving structural ambiguities is a major task of these methods. In the proposed approach, the parser produces a set of n-best trees based on a feature-extended PCFG grammar and then selects the best tree structure based on association strengths of dependency word-pairs. However, there is no sufficiently large Treebank producing reliable statistical distributions of all word-pairs. This paper aims to provide a self-learning method to resolve the problems. The word association strengths were automatically extracted and learned by parsing a giga-word corpus. Although the automatically learned word associations were not perfect, the constructed structure evaluation model improved the bracketed f-score from 83.09% to 86.59%. We believe that the above iterative learning processes can improve parsing performances automatically by learning word-dependence information continuously from web.
In this paper we present several use cases for the Stockholm TreeAligner, a software tool originally designed for annotating the alignments in a parallel treebank. The tool has been extended and improved to the point that it can now also serve as a general tool for browsing and searching monolingual and parallel treebanks. Among the use cases presented are: building a parallel treebank, browsing mono- and bilingual treebanks, consistency checking using the search function, comparing PP-attachment in different languages, and viewing different versions of the same treebank. A demonstration of the software will be held during the workshop.
Semantic analysis is a standard tool in the Natural Language Processing (NLP) toolbox with widespread applications. In this article, we look at tagging part of the Penn Chinese Treebank with semantic dependency. Then we take this tagged data to train a maximum entropy classifier to label the semantic relations between headwords and dependents to perform semantic analysis on Chinese sentences. The classifier was able to achieve an accuracy of over 84%. We then analyze the errors in classification to determine the problems and possible solutions for this type of semantic analysis.
We show that phrase structures in Penn Treebank style parses are not optimal for syntaxbased machine translation. We exploit a series of binarization methods to restructure the Penn Treebank style trees such that syntactified phrases smaller than Penn Treebank constituents can be acquired and exploited in translation. We find that by employing the EM algorithm for determining the binarization of a parse tree among a set of alternative binarizations gives us the best translation result. 1
This licentiate thesis deals with automatic syntactic analysis, or parsing, of natural languages. A parser constructs the syntactic analysis, which it learns by looking at correctly analyzed sentences, known as training data. The general topic concerns manipulations of the training data in order to improve the parsing accuracy. Several studies using constituency-based theories for natural languages in such automatic and data-driven syntactic parsing have shown that training data, annotated according to a linguistic theory, often needs to be adapted in various ways in order to achieve an adequate, automatic analysis. A linguistically sound constituent structure is not necessarily well-suited for learning and parsing using existing data-driven methods. Modifications to the constituency-based trees in the training data, and corresponding modifications to the parser output, have successfully been applied to increase the parser accuracy. The topic of this thesis is to investigate whether similar modifications in the form of tree transformations to training data, annotated with dependency-based structures, can improve accuracy for data-driven dependency parsers. In order to do this, two types of tree transformations are in focus in this thesis. The first one concerns non-projectivity. The full potential of dependency parsing can only be realized if non-projective constructions are allowed, which pose a problem for projective dependency parsers. On the other hand, non-projective parsers tend, among other things, to be slower. In order to maintain the benefits of projective parsing, a tree transformation technique to recover non-projectivity while using a projective parser is presented here. The second type of transformation concerns linguistic phenomena that are possible but hard for a parser to learn, given a certain choice of dependency analysis. This study has concentrated on two such phenomena, coordination and verb groups, for which tree transformations are applied in order to improve parsing accuracy, in case the original structure does not coincide with a structure that is easy to learn. Empirical evaluations are performed using treebank data from various languages, and using more than one dependency parser. The results show that the benefit of these tree transformations used in preprocessing and postprocessing to a large extent is language, treebank and parser independent.
In this article, we introduce a new representation based on lattice theory for lexical data from a lexical-database embodying the frame-semantic approach to language description, FrameNet. We present proof of the abundance of Concept Lattices as proposed in Formal Concept Analysis both in the theory of frames and in its present-day incarnation, the FrameNet resource, by constructing several types of these. We further argue for the adequacy of such lattices in representing linguistic data with contributions that range from data-visualization to the fine-tuning of some frame-theoretical concepts. We argue finally that FrameNet is better thought of as being a lexical resource rather than an ontology, but we make the case throughout the article for Concept Lattices being a linguistically adequate, formally effective intermediate representations from which knowledge representation languages may draw knowledge-rich, linguistic facts from FrameNet at their convenience.
We describe how the British National Corpus (BNC), a one hundred million word balanced corpus of British English, was parsed into Lexical Functional Grammar (LFG) c-structures and f-structures, using a treebank-based \nparsing architecture. The parsing architecture uses a state-of-the-art statistical parser and reranker trained on the Penn Treebank to produce context-free phrase structure trees, and an annotation algorithm to automatically annotate \nthese trees into LFG f-structures. We describe the pre-processing steps which were taken to accommodate the differences between the Penn Treebank and the BNC. Some of the issues encountered in applying the parsing \narchitecture on such a large scale are discussed. The process of annotating a gold standard set of 1,000 parse trees is described. We present evaluation results obtained by evaluating the c-structures produced by the statistical parser against the c-structure gold standard. We also present the results obtained by evaluating the f-structures produced by the annotation algorithm against an \nautomatically constructed f-structure gold standard. The c-structures achieve an f-score of 83.7% and the f-structures an f-score of 91.2%.
Because of the wide variety of contemporary practices used in the automatic syntactic parsing of natural languages, it has become necessary to analyze and evaluate the strengths and weaknesses of different approaches. This research is all the more necessary because there are currently no genre- and domain-independent parsers that are able to analyze unrestricted text with 100% preciseness (I use this term to refer to the correctness of analyses assigned by a parser). All these factors create a need for methods and resources that can be used to evaluate and compare parsing systems. This research describes: (1) A theoretical analysis of current achievements in parsing and parser evaluation. (2) A framework (called FEPa) that can be used to carry out practical parser evaluations and comparisons. (3) A set of new evaluation resources: FiEval is a Finnish treebank under construction, and MGTS and RobSet are parser evaluation resources in English. (4) The results of experiments in which the developed evaluation framework and the two resources for English were used for evaluating a set of selected parsers.
Proceedings of the Sixth International Workshop on Treebanks and \nLinguistic Theories. \nEditors: Koenraad De Smedt, Jan Hajič and Sandra Kübler. \nNEALT Proceedings Series, Vol. 1 (2007), 31-42. \n© 2007 The editors and contributors. \nPublished by \nNorthern European Association for Language \nTechnology (NEALT) \nhttp://omilia.uio.no/nealt. \nElectronically published at \nTartu University Library (Estonia) \nhttp://hdl.handle.net/10062/4476.
This paper describes an annotation guideline for a temporal relation-tagged corpus of Chinese. Our goal is construction of corpora to be used for a corpus-based analysis of temporal relations among events. Since annotating all combinations of events is inefficient, we examine the use of dependency structure to efficiently recognize temporal relations. We annotate a part of Treebank based on our guidelines. Then, we survey a small tagged data set to investigate the coverage of our method. While we find that use of dependency structure drastically reduces manual effort in constructing a tagged corpus with temporal relations, the coverage of the methods achieves about 63%.
The correct attachment of prepositional phrases (PPs) is a central disambiguation problem when parsing natural languages. This paper compares the baseline situation for French as exemplified in the Le Monde treebank with earlier findings for English, German and Swedish. We perform uniform treebank queries and show that the noun attachment rate for French prepositions is strongly influenced by the preposition de which is by far the most frequent preposition and has a strong tendency for noun attachment. We therefore also compute the noun attachment rate for the other prepositions separately as well as for the many complex prepositions that are explicitly marked in this treebank.
Data Oriented Parsing (DOP) is based on the idea of processing new input by com-bining fragments (associated with some probabilities) that are extracted from a treebank. In the simplest case these fragments are subparts of simple phrase structure trees (Tree-DOP). The approach is attractive in many ways but the im-poverished representational basis is a se-rious drawback from a linguistic point of view. This paper describes the the-oretical foundations of a novel version of DOP with a richer representational basis, a form of Typed Feature Struc-ture Grammar, specifically Head-driven Phrase Structure Grammar (HPSG).
Semantic analysis has become a bottleneck of many natural language applications. Machine translation, automatic question answering, dialog management, and others rely on high quality semantic analysis. Verbs are central elements of clauses with strong influence on the realization of whole sentences. Therefore the semantic analysis of verbs plays a key role in the analysis of natural language. We believe that solid disambiguation of verb senses can boost the performance of many real-life applications. In this thesis, we investigate the potential of statistical disambiguation of verb senses. Each verb occurrence can be described by diverse types of information. We investigate which information is worth considering when determining the sense of verbs. Different types of classification methods are tested with regard to the topic. In particular, we compared the Naive Bayes classifier, decision trees, rule-based method, maximum entropy, and support vector machines. The proposed methods are thoroughly evaluated on two different Czech corpora, VALEVAL and the Prague Dependency Treebank. Significant improvement over the baseline is observed.
Reviewed by: Treebanks: Building and using parsed corpora ed. by Anne Abeillé Philip Resnik Treebanks: Building and using parsed corpora. Ed. by Anne Abeillé. Dordrecht: Kluwer, 2003. Pp. 440. ISBN 1402013353. $74.95. Annotated corpora have been the fuel for a number of recent advances in the study of language, notably, although not exclusively, in computational linguistics. At the sentence level, corpus annotations range from shallow levels of linguistic representation, such as part-of-speech categories (Francis & Kučera 1982, Leech et al. 1994) or named entities (Strassel et al. 2003), through intermediate levels such as argument structure (Meyers et al. 2004, Palmer et al. 2005), to deeper levels of semantic representation such as semantic roles (Baker et al. 1998), word senses (Landes et al. 1998), events and temporal relations (Pustejovsky et al. 2003), or language-independent meaning representations (Farwell et al. 2004). When it comes to annotating sentences with linguistic representation, the sky is the limit (Meyers 2005). The ‘sweet spot’ in this range of annotations is occupied by treebanks, which is to say parsed (syntactically annotated) corpora. Unlike shallower annotations, syntactic parses capture hierarchical organization, a fundamental notion in virtually any modern theory of sentence structure. But unlike most deeper representations, syntactic parses can be created with high levels of inter-annotator reliability (though see Hovy et al. 2006 and references therein for recent progress in semantic treebanking). With respect to natural language processing applications, the impact of treebanks during the last decade has been remarkable, more than validating Marcus and colleagues’ (1993) premise that ‘significant, rapid progress can be made … by investigating those phenomena that occur most centrally in naturally occurring unconstrained materials and by attempting to automatically extract information about language from very large corpora’ (313). Beyond applications, syntactically annotated corpora have also begun to play an increasingly productive role in psycholinguistics, theoretical syntax, and language pedagogy (e.g. Corley et al. 2001, Jurafsky 2002, Dillon 2005, Meurers 2005, Resnik et al. 2005). In Treebanks: Building and using parsed corpora, Anne Abeille´ draws together a collection of fifteen short pieces focused primarily on the issues that come up in creating treebanks, demonstrated across an impressive variety of languages, along with six chapters on how treebanks are used. Although twenty-one chapters cannot be covered in detail in a short space, I present a brief walk through the chapters, followed by a discussion of the book as a whole. Abeillé’s introduction offers a very concise but clearly written primer on the main issues that come up in choosing representations, annotating corpora, and using treebanks in applications, folding in the obligatory pointers to the chapters that follow. In ‘The Penn Treebank: An overview’, Anne Taylor, Mitchell Marcus, and Beatrice Santorini provide a short, accessible description of the widely used Penn Treebank, extracting and updating the seminal article by Marcus and colleagues (1993) (which still remains the definitive source for in-depth discussion). In ‘Thoughts on two decades of drawing trees’, Geoffrey Sampson offers an engaging, personal discussion that combines elements of corpus description, linguistic analysis, and position paper. Unlike other chapters in the book, Sampson’s chapter offers a high-level look at corpus linguistics as an engineering and scientific discipline, and, contrary to some treebanking work, suggests that corpus annotation should make detail, accuracy, and explicitness a higher priority than the number of sentences annotated. Among the next thirteen chapters, nine provide detailed discussions of treebanking projects for specific languages, following a pattern that generally includes: (1) the goals and historical context of the project, (2) the selection of data to annotate (usually news text), (3) the annotation process (usually automatic analysis followed by human correction using project-specific tools), (4) details of representation (admitting wide variety, but usually an elaboration on syntactic constituency or grammatical dependency representation, with additional features to address language-specific [End Page 876] issues), (5) tools used for manual creation and correction of annotations (again admitting very wide variety), (6) a set of detailed, language-specific annotation choices that illustrate interesting and challenging aspects of the language under consideration, (7) a description of the project’s status, and (8) a brief evaluation and...
The purpose of the Chinese PropBank (CPB) project is to add a layer of annotation to the hand-parsed sentences in the Chinese Treebank (CTB) (Xue et al., 2005). This layer of annotation assigns predicatespecific argument labels to the constituents in a parse tree. The arguments of each predicate in the sentence, which are limited to verbs and their nominalizations in the work we report here, receive an argument label in the form of ArgN, where N is an integer between 0 and 5. These numbered arguments represent core arguments that are defined in relation to the predicate, which is labeled as Rel. Each core argument plays a unique role with regard to the predicate and generally the total number of core arguments for each predicate does not exceed 6. The core arguments annotated for the verb N (”investigate”) in Example (1) are the NPs (”the police”) and (”accident”) I(”cause”), which are labeled as Arg0 and Arg1 respectively. The semantic role labels added to the parse tree are in bold.
Journal Article Optimizing Procedures for the Making of Bilingual Dictionaries and the Concept of Linking Contrastive Lexical Databases Get access Godelieve Laureys Godelieve Laureys University of Gent (godelieve.laureys@Ugent.be) Search for other works by this author on: Oxford Academic Google Scholar International Journal of Lexicography, Volume 20, Issue 3, September 2007, Pages 295–311, https://doi.org/10.1093/ijl/ecm024 Published: 24 August 2007