Papers reviewed and determined not to be word norm studies. Use the flag icon to report errors or suggest re-inclusion.
18625 papers
We present discourse-level annotation of newspaper texts in German and English, as part of an ongoing project aimed at investigating information structure from a cross-linguistic perspective. Rather than annotating some specific notion of information structure, we propose a theory-neutral annotation of basic features at the levels of syntax, prosody and discourse, using treebank data as a starting point. Our discourse-level annotation scheme covers properties of discourse referents (e.g., semantic sort, delimitation, quantification, familiarity status) and anaphoric links (coreference and bridging). We illustrate what investigations this data serves and discuss some integration issues involved in combining different levels of stand-off annotations, created by using different tools.
We present an automatic semantic roles labeling system for structured trees of Chinese sentences. It adopts dependency decision making and example-based approaches. The training data and extracted examples are from the Sinica Treebank, which is a Chinese Treebank with semantic role semantic roles including thematic roles, such as ‘agent’; ‘theme’, ‘instrument’, and secondary roles of ‘location’, ‘time’, ‘manner ’ and roles for nominal modifiers. The design of role assignment algorithm is based on the different decision features, such as head-argument/modifier, case makers, sentence structures etc. It labels semantic roles of parsed sentences. Therefore the practical performance of the system depends on a good parser which labels the right structures of sentences. The system achieves 92.71 % accuracy in labeling the semantic roles for pre-structure- bracketed texts which is considerably higher than the simple method using probabilistic model of head-modifier relations. 1.
Predicting the location of phrase breaks within an utterance is an important task in text-to-speech synthesis, and can be done with reasonable accuracy using part-of-speech (POS) tags as features. However, it seems unlikely that the 40 or more different tags used by most taggers all contribute to this task, and in fact many may contribute noise. In this paper, we present an algorithm for reducing the standard Penn Treebank POS tag set for use in predicting phrase breaks. Using the best first search approach, the algorithm considers possible groupings of tags, searching the groupings that yield the highest overall performance. The reduced tag sets were evaluated by an n-gram model trained on POS sequences along with their associated juncture (break/non-break), the reduced tag set raised the model's performance on junctures correct from 90.38% to 92.43%, and reduced insertions from 2.89% to 1.83%.
1. Preface 2. Words, numbers and all that: The lexicon in sentence understanding (by Stevenson, Suzanne) 3. The lexicon in Optimality Theory (by Bresnan, Joan) 4. Optimality-theoretic Lexical Functional Grammar (by Johnson, Mark) 5. The lexicon and the laundromat (by Fodor, Jerry) 6. Semantics in the spin cycle: Competence and performance criteria for the creation of lexical entries (by Weinberg, Amy) 7. Connectionist and symbolist sentence processing (by Steedman, Mark) 8. A computational model of the grammatical aspects of word recognition as supertagging (by Kim, Albert E.) 9. Incrementality and lexicalism: A treebank study (by Lombardo, Vincenzo) 10. Modular architectures and statistical mechanisms: The case from lexical category disambiguation (by Crocker, Matthew) 11. Encoding and storage in working memory during sentence comprehension (by Stowe, Laurie A.) 12. The time course of information integration in sentence processing (by Spivey, Michael J.) 13. The lexical source of unexpressed participants and their role in sentence and discourse understanding (by Mauner, Gail) 14. Reduced relatives judged hard require constraint-based analyses (by Filip, Hana) 15. Predicting thematic role assignments in context (by Altmann, Gerry T.M.) 16. Lexical semantics as a basis for argument structure frequency biases (by Argamann, Vera) 17. Verb sense and verb subcategorization probabilities (by Roland, Doug) 18. Author index 19. Item index
To assign semantic roles in building Treebanks, there is a need for annotators having a guideline in determining semantic relations between phrasal head and its modifiers or arguments. Semantic roles are hard to have clear-cut definitions. It is not always easy to determine thematic relations between two concepts. This paper aims to introduce an integrated nominal modifier system. Basically we adopt other scholars ' incisive idea in analyzing semantic roles that modify general nouns. We use the approach of building a fine-grain taxonomy of role system. The taxonomy of fine-grain thematic roles makes the role determination easier for human annotators, since the meaning of a fine-grain semantic role is self explanatory and a higher-level semantic role is described by its hyponyms. The proposed taxonomy has been attested during construction of Sinica TreeBank and HowNet definitions of nominal concepts and proven to be more applicable than conventional flat structures. 1
The current research explored the processes that predominate during the anticipation of an emotionally salient event. Experiment 1 (N536), employed three different conditional stimuli followed by pictorial pleasant, unpleasant or neutral unconditioned stimuli. Half the participants were trained with visual CSs, the other half with tactile CSs. In the group trained with visual CSs, startle eyeblinks were larger and faster during CSs that were paired with unpleasant pictures than CSs paired with neutral or pleasant pictures respectively, indicating an affect startle pattern. This linear trend was not found in the group trained with tactile CSs. Experiment 2 (N564) aimed to investigate whether the affective pattern found in the startle data in Experiment 1 could also be found using a behavioural measure of emotion. This time participants’ reaction time during a post-experimental affective priming taskwas used as dependantmeasure to assess the presence of emotional learning. Instead of a simple differential conditioning task, an occasion setting paradigm was employed and participants were trained using either a feature positive or feature negative design with pleasant or unpleasant picture USs. For participants trained with unpleasant USs, valence ratings collected before and after conditioning training suggested the presence of emotional learning, whereas no such pattern was found for participants trained with pleasant USs. These findings were not confirmed in the priming data.
Idiolects are person-dependent similarities in language use. They imply that texts by one author show more similarities in language use than texts between authors. Sociolects, on the other hand, are group-dependent similarities in language use. They imply that texts by a group of authors, for instance in terms of gender or time period, share more similarities within a group than between groups. Although idiolects and sociolects are commonly used terms in the humanities, they have not been investigated a great deal from corpus and computational linguistic points of view. To test several idiolect and sociolect hypotheses a factorial combination was used of time period (Modernism, Realism), gender of author (male, female) and author (Eliot, Dickens, Woolf, Joyce) totaling 16 corresponding literary texts. In a series of corpus linguistic studies using Boolean and vector models, no conclusive evidence was found for the selected idiolect and sociolect hypotheses. In final analyses testing the semantics within each literary text, this lack of evidence was explained by the low homogeneity within a literary text.
Schwarz (2001, 2002) proposed the ex-Wald distribution, obtained from the convolution of Wald and exponential random variables, as a model of simple and go/no-go response time. This article provides functions for the S-PLUS package that produce maximum likelihood estimates of the parameters for the ex-Wald, as well as for the shifted Wald and ex-Gaussian, distributions. In a Monte Carlo study, the efficiency and bias of parameter estimates were examined. Results indicated that samples of at least 400 are necessary to obtain adequate estimates of the ex-Wald and that, for some parameter ranges, much larger samples may be required. For shifted Wald estimation, smaller samples of around 100 were adequate, at least when fits identified by the software as having ill-conditioned maximums were excluded. The use of all functions is illustrated using data from Schwarz (2001). The S-PLUS functions and Schwarz’s data may be downloaded from the Psychonomic Society’s Web archive, www. psychonomic.org/archive/.
Abstract This paper surveys work on applying the insights of lexicalized grammars to low‐level discourse, to show the value of positing an autonomous grammar for low‐level discourse in which words (or idiomatic phrases) are associated with discourse‐level predicate–argument structures or modification structures that convey their syntactic‐semantic meaning and scope. It starts by describing a lexicalized Tree Adjoining Grammar for discourse (D‐LTAG). It then reviews an initial experiment in parsing text automatically, using both a lexicalized TAG and D‐LTAG, and then touches upon issues involved in how lexico‐syntactic elements contribute to discourse semantics. The paper concludes with a brief description of the Penn Discourse TreeBank, a resource being developed for the study of discourse structure and semantics.
Finding definitions in huge text collections is a challenging problem, not only because of the many ways in which defini-tions can be conveyed in natural language texts but also be-cause the definiendum (i.e., the thing to be defined) has not, on its own, enough discriminative power to allow selection of definition-bearing passages from the collection. We have de-veloped a method that uses already available external sources to gather knowledge about the “definiendum ” before trying to define it using the given text collection. This knowledge consists of lists of relevant secondary terms that frequently co-occur with the definiendum in definition-bearing passages or “definiens”. External sources used to gather secondary terms are an on-line enyclopedia, a lexical database and the Web. These secondary terms together with the definiendum are used to select passages from the text collection perform-ing information retrieval. Further linguistic analysis is car-ried out on each passage to extract definition strings from the passages using a number of criteria including the presence of main and secondary terms or definition patterns.
Expert observers discriminate elementary features (e.g., colour, orientation) of salient objects outside their current focus of attention, but not complex features such as T/L shape or colour arrangement. Surprisingly, natural scene category behaves like an elementary feature in this context, in that scenes with vehicles, animals, etc. are successfully classified in a dual-task situation (Li et al., 2002). To confirm and extend this finding, we used a similar dual-task paradigm and presented a natural scene in the near periphery (8 to 14 eccentricity) simultaneously with an attention-demanding task near fixation (seven rotated Ts/Ls). Visual persistence was controlled by a particularly effective form of masking. For scenes of animals or vehicles (black and white with matched luminance scale), categorization performance exhibited little or no attentional cost, i.e., dual- and single-task performance were comparable. Thus, scene categorization under dual-task conditions can be explained neither by inadequate masking nor by trivial colour cues. In a second experiment, observers categorized scenes from the International Affective Image System database (IAPS, Lang et al., 1995) as “pleasant” or “unpleasant”. Although absolute performance was now lower (∼75% of the level reached under ideal viewing conditions), there was no significant attentional cost. As low-level features do not identify affective content, this implies some comprehension of scene gist. A control experiment highlighted the contrast between natural scenes and geometric stimuli in the dual-task situation (and also confirmed the peripheral absence of attention), in that observers failed to discriminate the colour arrangement of a peripherally presented ‘pill’ (half red, half green, inclined ±45 ). Li FF et al. (2002) Rapid natural scene categorization in the near absence of attention. PNAS 99: 9596ff. Lang PJ et al. (1995) IAPS: Technical Manual and Affective Ratings. Gainsville, FL.
UNLABELLED: The aim of this study was to compare PET with O-(2-(18)F-fluoroethyl)-L-tyrosine ((18)F-FET) and SPECT with 3-(123)I-iodo-alpha-methyl- L-tyrosine ((123)I-IMT) in patients with brain tumors. METHODS: Twenty patients with a suspected brain tumor were investigated by (18)F-FET PET, (123)I-IMT SPECT, and MRI within 3 wk. Region-of-interest analyses were performed on coregistered PET/SPECT/MRI images and the tumor-to-brain ratio (TBR), muscle-to-brain ratio (MBR), cerebellum-to-brain ratio (CerBR), and sinus-to-brain ratio (SBR) were calculated. In addition, the presence of tumor and the discrimination of anatomic structures on (18)F-FET PET and (123)I-IMT SPECT images were visually determined by 3 observers who were unaware of clinical data. RESULTS: The TBR of (18)F-FET and (123)I-IMT uptake in cerebral tumors showed a highly significant correlation (r = 0.96; P < 0.001). In the visual analysis for the presence or absence of tumors, no differences for (123)I-IMT SPECT and (18)F-FET PET were found in 19 of 20 patients; in one patient a low-grade glioma was only identified on (18)F-FET PET images but not on (123)I-IMT SPECT images. The contrast between tumor and normal brain was significantly higher in (18)F-FET PET (TBR, 2.0 +/- 0.9) than in (123)I-IMT SPECT (TBR, 1.5 +/- 0.5). The discrimination of anatomic structures yielded a significantly better score on (18)F-FET PET images (rating score, 2.6 +/- 0.9) compared with (123)I-IMT SPECT images (rating score, 1.7 +/- 0.9). The uptake of (18)F-FET in the muscles was significantly higher compared with (123)I-IMT (MBR (18)F-FET, 1.4 +/- 0.3; MBR (123)I-IMT, 0.6 +/- 0.2; P < 0.001) and (18)F-FET demonstrated a significantly higher blood-pool radioactivity than (123)I-IMT (SBR (18)F-FET, 1.3 +/- 0.2; SBR (123)I-IMT, 0.8 +/- 0.2; P < 0.001). CONCLUSION: The significant correlation of the TBRs of (18)F-FET and (123)I-IMT indicates that clinical experiences of brain tumor diagnostics with (123)I-IMT SPECT might be valid for (18)F-FET PET although substantial differences of the physiologic behavior were identified in extracerebral tissue. As (18)F-FET PET allows improved discrimination of anatomic structures and the tumor-to-brain contrast was significantly superior compared with (123)I-IMT SPECT scans, the results are encouraging for further evaluation of (18)F-FET for imaging brain tumors.
In this study, a technique called semantic self-organization is used to scale up the subsymbolic approach by allowing a network to optimally allocate frame representations from a semantic dependency graph. The resulting architecture, INSOMNet, was trained on semantic representations of the newly-released LinGO Redwoods HPSG Treebank of anno-tated sentences from the VerbMobil project. The results show that INSOMNet is able to accurately represent the semantic dependencies while demonstrating expectations and defaults, coactivation of multiple interpretations, and robust process-ing of noisy input. The cognitive plausibility of the model is underscored by the collective modelling of four experiments from the visual worlds paradigm to show the model’s ability to adapt to context.
In this paper, a hybrid language model is defined as a combination of a word-based <i>n</i>-gram, which is used to capture the local relations between words, and a category-based stochastic context-free grammar (SCFG) with a word distribution into categories, which is defined to represent the long-term relations between these categories. The problem of unsupervised learning of a SCFG in General Format and in Chomsky Normal Form by means of estimation algorithms is studied. Moreover, a bracketed version of the classical estimation algorithm based on the Earley algorithm is proposed. This paper also explores the use of SCFGs obtained from a treebank corpus as initial models for the estimation algorithms. Experiments on the UPenn Treebank corpus are reported. These experiments have been carried out in terms of the test set perplexity and the word error rate in a speech recognition experiment.
In this paper we describe on-going work aimed at creating a dependency-based annotated treebank for the BioMedical domain. Our starting point is the GENIA corpus, which is a corpus of 2000 MEDLINE abstracts, which has been manually annotated for various biological entities, according to the GENIA Ontology. There is an exponential growth of published research in this sector, which makes it difficult even for the experts to follow the recent developments. This creates the need for tools that can automatically process the research literature and extract only relevant information, such as interactions between genes and proteins. In order for these tools to be developed, annotated resources, such as corpora and Treebanks are of fundamental importance. Such resources will support the development of practical domain-specific information extraction tools.
In this paper we describe the extraction of thesaurus information from parsed dictionary definition sentences. The main data for our experiments comes from Lexeed, a Japanese semantic dictionary, and the Hinoki treebank built on it. The dictionary is parsed using a head-driven phrase structure grammar of Japanese. Knowledge is extracted from the semantic representation (Minimal Recursion Semantics). This makes the extraction process language independent.
In this paper, we consider the problem of enriching a Thai lexical database by extending the semantic information with selectional preferences. We propose a novel approach for acquiring selectional preferences of verbs, which is motivated by the tree cut model. We apply a model selection technique called the Bayesian Information Criterion (BIC). Given a semantic hierarchy, our goal is to generalize initial noun classes to the most plausible levels on that hierarchy. We present an iterative algorithm for generalization. The algorithm performs agglomerative merging on the semantic hierarchy in a bottomup manner. The BIC is used to measure the improvement of the model both locally and globally. In our experiments, we consider the Web as large corpora. We also propose approaches for extracting examples from the Web. Preliminarily experimental results are given to show the feasibility and effectiveness of our approach.
The development of wide-coverage grammars is at the core of robust NLP systems.This paper addresses the problem of grammar extraction from treebanks with respect to the issue of broad coverage along three dimensions: the grammar formalism (contextfree grammar, dependency grammar, lexicalized tree adjoining grammar), the domain of the annotated corpus (press reports, civil law) and the language of the corpus (English, Korean, Chinese, Italian).We have extracted three grammars from an annotated corpus of Italian and we have comparatively analyzed the coverage of a test set; then, working on two different domain subcorpora we have compared the cross-domain coverage of the extracted grammars; finally, we have compared the grammars for four different languages.The results are that there are relevant differences in coverage among formalisms and domains; a more limited difference appears in the crosslinguistic comparison.
This paper presents a deterministic dependency parser based on memory-based learning, which parses English text in linear time. When trained and evaluated on the Wall Street Journal section of the Penn Treebank, the parser achieves a maximum attachment score of 87.1%. Unlike most previous systems, the parser produces labeled dependency graphs, using as arc labels a combination of bracket labels and grammatical role labels taken from the Penn Treebank II annotation scheme. The best overall accuracy obtained for identifying both the correct head and the correct arc label is 86.0%, when restricted to grammatical role labels (7 labels), and 84.4% for the maximum set (50 labels).
In the design of a Multilingual Lexical Database, one of the biggest problems is constituted by conceptual mismatches between languages, and the resulting matter of lexical gaps. Lexical gaps concern words for which there is no direct translation in a target language, but which nonetheless need to receive a translation within the system. In this article, it will be shown that the various possible ways of dealing with these lexical gaps can be classified in four basic groups. Using the SIM<it>u</it>LLDA system as an example (Janssen 2002), the advantages of the structured interlingua approach over the other possibilities will be explained. With the SIM<it>u</it>LLDA set-up, it is possible to derive correct lexical definitions for lexical gaps from the lexical database. How this process of “lexical gap filling” works will be shown using a concrete example of a lexical gap: the treatment of the English words <it>river</it> and <it>stream </it>in contrast with the French words <it>fleuve</it> and <it>rivière</it>.
This paper describes NomBank, a project that will provide argument structure for instances of common nouns in the Penn Treebank II corpus. NomBank is part of a larger effort to add additional layers of annotation to the Penn Treebank II corpus. The University of Pennsylvania’s PropBank, NomBank and other annotation projects taken together should lead to the creation of better tools for the automatic analysis of text. This paper describes the NomBank project in detail including its specifications and the process involved in creating the resource. 1
Abstract. The use of lexicons and corpora advances both linguistic re-search and performance of current natural language processing (NLP) systems. We present a tool that exploits such resources, specifically En-glish and German lexical databases and the World Wide Web to recognise English inclusions in German newspaper articles. The output of the tool can assist lexical resource developers in monitoring changing patterns of English inclusion usage. The corpus used for the classification covers three different domains. We report the classification results and illustrate their value to linguistic and NLP research. 1
Past approaches to developing an effective lexicon component in a grammar development environment have suffered from a number of usability and efficiency issues. We present a lexical database module currently in use by a number of grammar development projects. The database module presented addresses issues which have caused problems in the past and the power of a database architecture provides a number of practical advantages as well as a solid framework for future extension. 1.
In this paper we describe the construction of a new Japanese lexical resource: the Hinoki treebank. The treebank is built from dictionary definition sentences, and uses an HPSG based Japanese grammar to encode the syntactic and semantic information. We show how this treebank can be used to extract thesaurus information from definition sentences in a language-neutral way using minimal recursion semantics. 1
CDG represents a sentence's grammatical structure as assignments of dependency relations to functional variables associated with each word in the sentence. In this paper, we describe a statistical CDG (SCDG) parser that performs parsing incrementally and evaluate it on the Wall Street Journal Penn Treebank. Using a tight integration of multiple knowledge sources, together with distance modeling and synergistic dependencies, this parser achieves a parsing accuracy comparable to several state-of-the-art context-free grammar (CFG) based statistical parsers using a dependency-based evaluation metric. Factors contributing to the SCDG parser's performance are analyzed.
This paper explores the interaction between conceptual structure and morpho-syntax. In particular, we show that ontology-based conceptual classification can be used to predict internal relations in compounds. We propose an ontology-based approach to predict the semantic relation between the two component words in Mandarin VV compounds. A Mandarin VV compound is classified according to the eventive relation between the two simplex verbs. These relations specify how the eventive meanings of the two simplex verbs combine to form the meaning of the compound. The three types of eventive relations that we deal with in this paper are: coordinate, modificational, and resultative. Since the way in which two events combine with each other depends upon their event types, we hypothesize that the eventive relations can be predicted by the conceptual classified event types of the two simplex verbs. An approach of ontology-based prediction is proposed based on this hypothesis. The assignment of ontology classification for each simplex verb is based on SUMO and Sinica BOW. The correlation between the ontology class of each verb position and each eventive type is trained and scored based on a manually tagged lexical database. We encode the ontology information of each VV compound in a 3-tuple based on these correlation scores. This 3-tuple is represented as a three-dimensional vector and used to predict the eventive type of new VV compounds. Our classification experiment on unknown VV compounds yields good recall and precision. 1.
The goal of the MULI (MUltiLingual Information structure) project is to empirically analyse information structure in German and English newspaper texts. In contrast to other projects in which information structure is annotated and investigated (e.g. in the Prague Dependency Treebank, which mirrors the basic information about the topic-focus articulation of the sentence), we do not annotate theory-biased categories like topic-focus or theme-rheme. Trying to be as theory-independent as possible, we annotate those features which are relevant to information structure and on the basis of which typical patterns, co-occurrences or correlations can be determined. We distinguish between three annotation levels: syntax, discourse and prosody. The data is based on the TIGER Corpus for German and the Penn Treebank for English, since the existing information on part-of-speech and syntactic structure can be re-used for our purposes. The actual annotation of an English example sequence illustrates our choice of categories on each level. Their combination offers the possibility to investigate how information structure is realised and can be interpreted. 1.
OBJECTIVES: To compare the effect of a specialized care facility (SCF) on quality of life (QoL) for residents with middle- to late-stage dementia over a 1-year period with residence in traditional institutional facilities. DESIGN: A prospective, matched-group design with assessments of QoL every 3 months for 1 year. SETTING: Twenty-four long-term care centers and four designated assisted living environments in an urban center in western Canada. PARTICIPANTS: One hundred eighty-five residents with Global Deterioration Scores of 5 or greater were enrolled: 62 in the intervention SCF group and 123 in the traditional institutional facilities groups. INTERVENTION: The SCF is a 60-bed purpose-built facility with 10 people living in six bungalows. The facility followed an ecologic model of care that is responsive to the unique interplay of each person and the environment. This model encompasses a vision of long-term care that is more comfortable, more like home, and offers more choice, meaningful activity, and privacy than traditional settings. MEASUREMENTS: QoL outcomes were assessed using the Brief Cognitive Rating Scale, Functional Assessment Staging, Cohen-Mansfield Agitation Inventory, Pleasant Events Scale-Alzheimer's disease, Multidimensional Observation Scale of Elderly Subjects, and Apparent Affect Rating Scale. RESULTS: The intervening SCF group demonstrated less decline in activities of daily living, more sustained interest in the environment, and less negative affect than residents in the traditional institutional facilities. There were no differences between groups in concentration, memory, orientation, depression, or social withdrawal. CONCLUSION: The present study suggests that QoL for adults with middle- to late-stage dementia is the same or better in a purpose-built and staffed SCF than in traditional institutional settings.
Biomimetic design uses ideas from biological phenomena as inspiration in design. To support biomimetic design, biological analogies are identified by finding instances of functional keywords that describe the engineering problem in biological knowledge in natural-language format. Challenges in using this approach include the identification of keywords, and the quantity and quality of results found. WordNet, a lexical database, is used as a language framework to systematically generate alternative keywords to find matches and analyze the results of searches. Troponyms from WordNet were found to provide better and more plentiful keywords than did synonyms. Due to the potentially large number of matches to keywords, matches are analyzed to facilitate extraction of dominant biological phenomena associated with keywords. This analysis found that words that frequently collocated with keywords tend to be objects of the keyword verb or agents that carry out the actions of the keyword. Furthermore, nouns that are inanimate, e.g., substances, tend to be objects, and nouns that are animate e.g., animals, organs, tend to be agents. Distinguishing frequently collocated words and their relationships to keywords can be used to facilitate identification of biological analogies in natural-language format to support design.Copyright © 2004 by ASME
This paper describes a method for ranking a large list of adjectives according to a subjectivity score without resorting to any knowledge-intensive external resources (such as lexical databases, parsers or manual annotation). The method only requires a list of adjectives to be ranked and a small set of &quot;seeds&quot; (manually selected subjective adjectives). The subjectivity score is obtained by computing the mutual information of pairs of adjectives taken from each set, using frequency and cooccurrence frequency counts on the World Wide Web, collected through queries to the AltaVista search engine. The obtained results improve significantly over a comparable low-resource acquisition algorithm.
The purpose of this paper is to describe the TuBa-D/Z treebank of written German and to compare it to the independently developed TIGER treebank (Brants et al., 2002). Both treebanks, TIGER and TuBa-D/Z, use an annotation framework that is based on phrase structure grammar and that is enhanced by a level of predicate-argument structure. The comparison between the annotation schemes of the two treebanks focuses on the different treatments of free word order and discontinuous constituents in German as well as on differences in phrase-internal annotation.
Evolutionary theory predicts that female intrasexual competition will occur when males of high genetic quality are considered to be a resource. It is probable that women compete in terms of attractiveness since this is one of the primary criteria used by men when selecting mates. Furthermore, because hormones influence the mate-selection process, they may also mediate competition. One competitive strategy that women use is derogation--any act intended to decrease a rival's perceived value. To investigate intrasexual competition through derogation, the influence of oestrogen on women's ratings of female facial attractiveness was examined. During periods of high oestrogen, competition, and hence derogation, increased, as evidenced by lower ratings of female facial attractiveness. By contrast, oestrogen levels did not significantly affect ratings of male faces. These findings support the theory of female intrasexual competition with respect to attractiveness.
Especially for English, the number of hand-coded electronic resources available to the Natural Language Processing Community keeps growing: annotated corpora, treebanks, lexicons, wordnets, etc. Unfortunately, initial funding for such projects is much easier to obtain than the additional funding needed to enlarge or improve upon such resources. Thus once one proves the usefulness of a resource, it is difficult to make that resource reach its full potential. We discuss techniques for combining dictionary resources and producing others by semi-automatic means. The resources we created using these techniques have become an integral part of our work on NomBank, a project with the goal of annotating noun arguments in the Penn Treebank II corpus (PTB). 1.
In information retrieval and text mining, information on word senses is usually taken from dictionaries or lexical databases that have been prepared by lexicographers. We propose an automatic method for word sense induction, i.e. for the discovery of a set of sense descriptors to a given ambiguous word. The approach is based on the statistics of word co-occurrence as derived from Web pages. The underlying assumption is that the senses of an ambiguous word are best described by terms that, although bearing a strong association to this word, are mutually exclusive, i.e. whose association strength within the retrieved Web pages is as weak as possible. Measuring association strength is based upon a novel confidence gain approach that relates the observed co-occurrence frequency for two sense descriptor candidates to an average co-occurrence frequency for pairs of arbitrary words. The proposed approach is fully unsupervised and takes into account the contemporary meanings of words, as reflected in texts from the Internet. Our results are evaluated using a list of ambiguous words commonly referred to in the literature.
This paper presents a prosodic phrasing model for Korean to be used in a text-to-speech synthesis (TTS) system. Read text corpora were morpho-syntactically parsed and prosodically labeled following the Penn Korean Treebank (Han, Chunghye, Ko, Eon-Suk, Yi, Heejong, Palmer, M., 2002. Penn Korean Treebank: development and evaluation. In: Proceedings of the 16th Pacific Asian Conference on Language and Computation. Korean Society for Language and Information.) and K-ToBI prosodic labeling conventions (Sun-Ah, J., 2000. K-ToBI (Korean ToBI) labelling conventions. Version 3.1. Available from: URL.), respectively. Decision trees were trained with morpho-syntactic and textual distance features to predict locations of accentual and intonational phrase breaks. Our phrasing model cross-validated on a 300-sentence corpus (6936 words or 21,436 syllables, with an average of 72 syllables or 23 words per sentence) predicted non-breaks with F=92.4% and breaks with F=88.0% (F=72.8% for accentual phrase breaks and F=71.3% for intonational phrase breaks).
We describe the automatic conversion of English Penn Treebank (PTB) annotations into Language Neutral Syntax (LNS) (Campbell and Suzuki, 2002a,b). In this paper, we describe LNS and why it is useful, describe the conversion algorithm, present an evaluation of the conversion, and discuss some uses of the converted annotations and the potential for extending the coverage to other languages. The work described here is in the spirit of other automatic re-annotations of PTB trees (e.g. Frank, 2000 and Meyers, 2001), but differs in the nature of the output.
Large-scale Knowledge Representation Resources for Cognitive Science Research Martha S. Palmer Department of Computer and Information Science, University of Pennsylvania 3330 Walnut Street, Philadelphia, PA 19104-6389 USA George A. Miller Department of Psychology, Princeton University 1-S-5 Green Hall, Princeton, NJ 08544 USA Charles J. Fillmore International Computer Science Institute, University of California at Berkeley 1947 Center St., Suite 600, Berkeley, CA 94704-1198 USA Doug Lenat Cycorp, Inc. Suite 100, 3721 Executive Center Drive, Austin, TX 78731 USA Pat Hayes Institute for Human and Machine Cognition 40 South Alcaniz Street, Pensacola, FL 32502 Treebank syntactic structures: The Proposition Bank, or PropBank. PropBank consists of argument labels for the semantic roles of individual verbs and similar predicating expressions such as participial modifiers and nominalizations. This talk will describe the PropBank verb semantic role annotation being done at Penn for both English and Chinese. The annotation process will be discussed as well as the use of existing lexical resources such as WordNet, Levin classes and VerbNet. Comparisons with similar projects, including the FrameNet Project at Berkeley and the Prague Tectogrammatics project, will be made. Motivation One of the unique features of cognitive science has been its emphasis on understanding how people represent and use knowledge. Unfortunately, knowledge representation can be quite difficult if one is starting from scratch. Having “off the shelf” representation systems that can be used for investigations can make new types of cognitive modeling efforts possible, at a scale that would otherwise be impossible. The participants in this symposium each represent a different approach to building such representational resources. George A. Miller: WordNet is a lexical database that contains 146,900 English nouns, verbs, adjectives, and adverbs that are now organized by semantic relations into 117,500 meanings, where a meaning is represented by a set of synonyms (a synset) that can be used to express that meaning. An entry in WordNet consists of a synset, a definitional gloss, and (sometimes) one or more sentences illustrating usage. The semantic relations used to organize words and entries are synonymy and antonymy, hyponymy, troponymy and hypernymy, meronymy and holonymy. A currently active project, the disambiguation of definitional glosses, will be discussed. Doug Lenat: Cognitive Science research could benefit from large-scale representational building blocks. One such building block is a broad ontology of, well, everything. Another one, resting on that, is a formal axiomatization of most of the meaning of most of those concepts; i.e., millions of axioms about those hundreds of thousands of terms. These two pieces have been worked on for twenty years -- and the better part of a person-century of effort -- as Cyc. It's been highly proprietary so far, with a small tip of its ontology exposed as OpenCyc. Starting this summer, though, Cyc is being made available in its entirety for R&D purposes for free, courtesy of Ron Brachman at DARPA's IPTO. In this presentation I will briefly describe this ResearchCyc ontology and KB, and some initial utilities provided with it (inference engine, English-Cyc lexicon, interfaces), how this might fit in with the other infrastructure elements being described in this panel, and how these might leverage your work. Charles J. Fillmore: The FrameNet project is morphing from a lexicon-building project to a system capable of providing a layer or two of semantic annotation for full sentences. My remarks will summarize the kinds of information the FrameNet database can provide now, and what it should be able to offer in the not too distant future, for researchers in language engineering and cognitive science. FrameNet is moving toward greater coverage of the lexicon, adaptation to specialist vocabularies, more systematic treatment of multiword expressions, and provisions for incorporating a wide variety of (non-core) grammatical constructions. Pat Hayes Pat Hayes, the discussant for this symposium, has the unique distinction of having inventing two of the most widely-used representations for change, the situation calculus (with John McCarthy) and histories. He is a Fellow of the American Association for Artificial Intelligence and the Cognitive Science Society. Martha S. Palmer: Recently, a consensus has been achieved as to a task-oriented level of semantic representation to be layered on top of the existing Penn
Abstract We present a computational language learning algorithm that can induce massively probabilistic grammars from treebank data. The paper is based on chapter 6 in my forthcoming thesis “Discontinuous Grammar: A dependency-based model of human parsing and language learning.” We start in section 1 by introducing the prerequisites from probability theory and statistics that are needed in the rest of the paper. In section 2, we define weighted grammars and massively probabilistic grammars, and discuss their relationship to standardly used grammars. In section 3, we address the problem of estimating probability distributions for hierarchically structured categorical data, and present an estimation algorithm that selects a hierarchical partition model by means of local search. We also describe a simulation study that shows that the algorithm performs well unless the distribution that generated the data is highly symmetric. Finally, in section 4, we outline how the algorithm can be used to learn massively probabilistic dependency grammars, exemplified by the task of learning the probabilities of complement structures.
We present a relational learning framework for grammar induction that is able to learn meaning as well as syntax. We introduce a type of constraint-based grammar, lexicalized well-founded grammar (lwfg), and we prove that it can always be learned from a small set of semantically annotated examples, given a set of assumptions. The semantic representation chosen allows us to learn the constraints together with the grammar rules, as well as an ontology-based semantic interpretation. We performed a set of experiments showing that several fragments of natural language can be covered by a lwfg,and that it is possible to choose the representative examples heuristically, based on linguistic knowledge.
The JMdict project has at its aim the compilation of a multilingual lexical database with Japanese as the pivot language. Using an XML structure designed to cater for a mix of languages and a rich set of lexicographic information, it has reached a size of approximately 100,000 entries, with most entries having translations in English, French and German. The compilation involves information re-use, with the French and German translations being drawn from separately maintained lexicons. Material from other languages is also being included. The file is freely available for research purposes and for incorporation in dictionary application software, and is available in several WWW server systems.
Here, we study the design of domain lexicons and address the problems of lexical knowledge representation and linking between different knowledge sources. A domain lexicon contains substantial domain specific vocabularies with associated phonological, morphological, syntactic and semantic/pragmatic information as well as links to general knowledge bases which are rich enough to support knowledge-intensive models for practical NLP systems. We take financial domain as an example to illustrate the representation structures for syntactic and semantic knowledge. In order to suit for both maximal reusability and deep analysis, our domain lexicon is designed with uniform knowledge representation and fine-grain feature encoding. We also address the issues of how to bridge the gaps between coarse-grain general lexicons and fine-grain domain lexicons.
Nowadays language learning systems beyond hypermedia technologies increasingly adopt Computational Linguistics technologies, network technologies and Artificial Intelligence to improve the learning process. However, although such technologies are thought to be very useful to support authentic language learning environments and to simulate real communication situations a certain neglect of such systems is lamented, even within the research community. One reason is that specific knowledge out of different disciplines is necessary to provide an innovative language learning system. This development process can be made more efficient, if methods are found to join independently developed mature products to a single learning environment. In this article we will present two interdisciplinary research projects out of the domain of lexicography, which are both ready to be integrated within other e-learning resources. Word Manager (WM) is a reusable lexical database that provides information about inflection, word formation and orthography. ELDIT is a web-based educational system for learners of the German and Italian languages. Both systems can be integrated into other learning environments as well as extended by new modules by themselves. The collaboration between ELDIT and WM has shown that these advantages can also be realized in practice. Joining the two products efficiently supported the creation of an innovative language learning environment.