Verbnet Representations: Subevent Semantics for Transfer Verbs

Total Page:16

File Type:pdf, Size:1020Kb

Verbnet Representations: Subevent Semantics for Transfer Verbs VerbNet Representations: Subevent Semantics for Transfer Verbs Susan Windisch Brown Julia Bonn James Gung University of Colorado University of Colorado University of Colorado [email protected] [email protected] [email protected] Annie Zaenen James Pustejovsky Martha Palmer Stanford University Brandeis University University of Colorado [email protected]@[email protected] Abstract verb senses and introduce a means for automati- cally translating text to these representations. We This paper announces the release of a new ver- explore the format of these representations and the sion of the English lexical resource VerbNet types of information they track by thoroughly ex- with substantially revised semantic represen- amining the representations for transfer of posses- tations designed to facilitate computer plan- sions and information. These event types are ex- ning and reasoning based on human language. We use the transfer of possession and trans- cellent examples of complex events with multi- fer of information event representations to il- ple participants and relations between them that lustrate both the general framework of the change across the time frame of the event. By representations and the types of nuances the aligning our new representations more closely new representations can capture. These repre- with the dynamic event structure encapsulated by sentations use a Generative Lexicon-inspired the Generative Lexicon, we can provide a more subevent structure to track attributes of event precise subevent structure that makes the changes participants across time, highlighting opposi- tions and temporal and causal relations among over time explicit (Pustejovsky, 1995; Pustejovsky the subevents. et al., 2016). Who has what when and who knows what when are exactly the sorts of things that we 1 Introduction want to extract from text, but this extraction is dif- ficult without explicit, computationally-tractable Many natural language processing tasks have seen representations. These event types also make up a rapid advancement in recent years using deep substantial portion of VerbNet: 37 classes of verbs learning methods; however, those tasks that re- deal with change of possession and transfer of in- quire precise tracking of event sequences and par- formation out of VerbNet’s 300+ classes, covering ticipants across a discourse still perform better 810 verbs. using explicit representations of the meanings of each sentence or utterance. To be most useful 2 Background for automatic language understanding and genera- tion, such representations need to be both automat- The language resource VerbNet (Kipper et al., ically derivable from text and reasonably format- 2006) is a hierarchical, wide-coverage verb lexi- ted for computer analysis and planning systems. con that groups verbs into classes based on sim- For applications like robotics or interactions with ilarities in their syntactic and semantic behavior avatars, commonsense inferences needed to under- (Schuler, 2005). Each class in VerbNet includes stand human language directions or interactions a set of member verbs, the thematic roles used are often not derivable directly from the utterance. in the predicate-argument structure of these mem- Tracking intrinsic and extrinsic states of entities, bers (Bonial et al., 2011), and the class-specific such as their existence, location or functionality, selectional preferences for those roles. The class currently requires explicit statements with precise also provides a set of typical syntactic patterns and temporal sequencing. corresponding semantic representations. A verb In this paper, we describe new semantic rep- can be a member of multiple classes; for exam- resentations for the lexical resource VerbNet that ple, run is a member of 8 VerbNet classes, in- provide this sort of information for thousands of cluding the run-51.3.2 class (he ran to the store) 154 Proceedings of the First International Workshop on Designing Meaning Representations, pages 154–163 Florence, Italy, August 1st, 2019 c 2019 Association for Computational Linguistics and the function-105.2.1 class (the car isn’t run- these caused-motion frames. ning). These memberships usually correspond to Each frame also includes a semantic represen- coarse-grained senses of the verb. The resource tation that uses basic predicates to show the rela- was originally based on Levin’s (1993) analysis of tionships between the thematic role arguments and English verbs but has since been expanded to in- to track any changes over the time course of the clude dozens of additional classes and hundreds of event. Thematic roles that appear in the ”syntax” additional verbs and verb senses. should always appear somewhere in the semantic VerbNet representations previously formed the representation. Overall, this linking in each frame basis for Parameterized Action Representation of the syntactic pattern to a semantic representa- (PAR) providing a conceptual representation of tion is a unique feature of VerbNet that emphasizes different types of actions (Badler et al., 1999). the close interplay of syntax and semantics. These actions involve changes of state, changes of location, and exertion of force and can be used 2.2 Revision of the Semantic Representations to animate human avatars in a virtual 3D environ- VerbNet’s old representations included an event ment (R. Bindiganavale and Palmer, 2000). They variable E as an argument to the predicates. are particularly well suited for motion and con- Representations of states were indicated with tact verb classes, providing an abstract, language- either a bare E, as for the own-100 class: independent representation (Kipper and Palmer, has possession(E, Pivot, Theme), or During(E), 2000). The more precise temporal sequencing de- as for the contiguous location-47.8 class (Italy scribed here is even more suitable as a foundation borders France): contact(During(E), Theme, Co- for natural language instructions and human-robot Theme). Most classes having to do with change, or human-avatar interactions. such as changes in location, changes in state and changes in possession, used a path rel predicate 2.1 VerbNet in combination with Start(E), During(E), and End/Result(E) to show the transition from one VerbNet has long been used in NLP for seman- location or state to another (1). tic role labeling and other inference-enabling tasks (Shi and Mihalcea, 2005; Giuglea and Moschitti, 2006; Loper et al., 2007; Bos, 2008). In addi- (1) The rabbit hopped across the lawn. tion, automatic disambiguation of a verb’s Verb- Theme V Trajectory Net class has been used as a stand-in for verb sense motion(during(E), Theme) disambiguation (Abend et al., 2008; Brown et al., path rel(start(E), Theme, ?Initial location1, 2014; Croce et al., 2012; Kawahara and Palmer, CH OF LOC, prep) 2014). path rel(during(E), Theme, Trajectory, VerbNet’s semantic representations use a CH OF LOC, prep) Davidsonian first-order-logic formulation that path rel(end(E), Theme, ?Destination, incorporates the thematic roles of the class. Each CH OF LOC, prep) frame in a class is labeled with a flat syntactic pattern (e.g., NP V NP). The ”syntax” that follows Efforts to use VerbNet’s semantic representa- shows how the thematic roles for that class appear tions (Zaenen et al., 2008; Narayan-Chen et al., in that pattern (e.g., Agent V Patient), much 2017), however, indicated a need for greater con- like the argument role constructions of Goldberg sistency and expressiveness. We have addressed (2006). A previous revision of the VerbNet se- consistency on several fronts. First, all neces- mantic representations made the correspondence sary participants are accounted for in the repre- of these patterns to constructions more explicit sentations, whether they are instantiated in the by using a common predicate (i.e., path rel) for syntax, incorporated in the verb itself (e.g., to all caused-motion construction frames(Hwang, drill), or simply logically necessary (e.g., all en- 2014). At the request of some users, we are sub- tities that change location begin in an initial loca- stituting more specific predicates for the general tion, whether it is commonly mentioned or not). path rel predicate, such as has location, has state 1A question mark in front of a thematic role indicates a and has possession, although the subevent pat- role that appears in the syntax in some frames for the class terns continue to show the commonality across but not in this particular frame. 155 Second, similar event types are represented with troduced a new event variable, e¨, to distinguish a similar format; for example, all states are rep- a process from other types of subevents, such as resented with E, never with During(E). Finally, states. For example, see the motion predicate in predicates are given formal definitions that apply (2). across classes. (2) The rabbit hopped across the lawn. In order to clarify what is happening at each Theme V Trajectory stage of an event, we turned to the Generative has location(e , Theme, ?Initial Location) Lexicon (Pustejovsky, 1995) for an explicit the- 1 motion(e¨ , Theme, Trajectory) ory of subevent structure. Classic GL character- 2 :has location(e , Theme, ?Initial location) izes the different Aktionsarten in terms of struc- 2 has location(e , Theme, ?Destination) tured subevents, with states represented with a 3 simple e, processes as a sequence of states char- acterizing values of some attribute, e1...en, and Although people infer that an entity
Recommended publications
  • An Artificial Neural Network Approach for Sentence Boundary Disambiguation in Urdu Language Text
    The International Arab Journal of Information Technology, Vol. 12, No. 4, July 2015 395 An Artificial Neural Network Approach for Sentence Boundary Disambiguation in Urdu Language Text Shazia Raj, Zobia Rehman, Sonia Rauf, Rehana Siddique, and Waqas Anwar Department of Computer Science, COMSATS Institute of Information Technology, Pakistan Abstract: Sentence boundary identification is an important step for text processing tasks, e.g., machine translation, POS tagging, text summarization etc., in this paper, we present an approach comprising of Feed Forward Neural Network (FFNN) along with part of speech information of the words in a corpus. Proposed adaptive system has been tested after training it with varying sizes of data and threshold values. The best results, our system produced are 93.05% precision, 99.53% recall and 96.18% f-measure. Keywords: Sentence boundary identification, feed forwardneural network, back propagation learning algorithm. Received April 22, 2013; accepted September 19, 2013; published online August 17, 2014 1. Introduction In such conditions it would be difficult for a machine to differentiate sentence terminators from Sentence boundary disambiguation is a problem in natural language processing that decides about the ambiguous punctuations. beginning and end of a sentence. Almost all language Urdu language processing is in infancy stage and processing applications need their input text split into application development for it is way slower for sentences for certain reasons. Sentence boundary number of reasons. One of these reasons is lack of detection is a difficult task as very often ambiguous Urdu text corpus, either tagged or untagged. However, punctuation marks are used in the text.
    [Show full text]
  • An Arabic Wordnet with Ontologically Clean Content
    Applied Ontology (2021) IOS Press The Arabic Ontology – An Arabic Wordnet with Ontologically Clean Content Mustafa Jarrar Birzeit University, Palestine [email protected] Abstract. We present a formal Arabic wordnet built on the basis of a carefully designed ontology hereby referred to as the Arabic Ontology. The ontology provides a formal representation of the concepts that the Arabic terms convey, and its content was built with ontological analysis in mind, and benchmarked to scientific advances and rigorous knowledge sources as much as this is possible, rather than to only speakers’ beliefs as lexicons typically are. A comprehensive evaluation was conducted thereby demonstrating that the current version of the top-levels of the ontology can top the majority of the Arabic meanings. The ontology consists currently of about 1,300 well-investigated concepts in addition to 11,000 concepts that are partially validated. The ontology is accessible and searchable through a lexicographic search engine (http://ontology.birzeit.edu) that also includes about 150 Arabic-multilingual lexicons, and which are being mapped and enriched using the ontology. The ontology is fully mapped with Princeton WordNet, Wikidata, and other resources. Keywords. Linguistic Ontology, WordNet, Arabic Wordnet, Lexicon, Lexical Semantics, Arabic Natural Language Processing Accepted by: 1. Introduction The importance of linguistic ontologies and wordnets is increasing in many application areas, such as multilingual big data (Oana et al., 2012; Ceravolo, 2018), information retrieval (Abderrahim et al., 2013), question-answering and NLP-based applications (Shinde et al., 2012), data integration (Castanier et al., 2012; Jarrar et al., 2011), multilingual web (McCrae et al., 2011; Jarrar, 2006), among others.
    [Show full text]
  • A Comparison of Knowledge Extraction Tools for the Semantic Web
    A Comparison of Knowledge Extraction Tools for the Semantic Web Aldo Gangemi1;2 1 LIPN, Universit´eParis13-CNRS-SorbonneCit´e,France 2 STLab, ISTC-CNR, Rome, Italy. Abstract. In the last years, basic NLP tasks: NER, WSD, relation ex- traction, etc. have been configured for Semantic Web tasks including on- tology learning, linked data population, entity resolution, NL querying to linked data, etc. Some assessment of the state of art of existing Knowl- edge Extraction (KE) tools when applied to the Semantic Web is then desirable. In this paper we describe a landscape analysis of several tools, either conceived specifically for KE on the Semantic Web, or adaptable to it, or even acting as aggregators of extracted data from other tools. Our aim is to assess the currently available capabilities against a rich palette of ontology design constructs, focusing specifically on the actual semantic reusability of KE output. 1 Introduction We present a landscape analysis of the current tools for Knowledge Extraction from text (KE), when applied on the Semantic Web (SW). Knowledge Extraction from text has become a key semantic technology, and has become key to the Semantic Web as well (see. e.g. [31]). Indeed, interest in ontology learning is not new (see e.g. [23], which dates back to 2001, and [10]), and an advanced tool like Text2Onto [11] was set up already in 2005. However, interest in KE was initially limited in the SW community, which preferred to concentrate on manual design of ontologies as a seal of quality. Things started changing after the linked data bootstrapping provided by DB- pedia [22], and the consequent need for substantial population of knowledge bases, schema induction from data, natural language access to structured data, and in general all applications that make joint exploitation of structured and unstructured content.
    [Show full text]
  • ACL 2019 Social Media Mining for Health Applications (#SMM4H)
    ACL 2019 Social Media Mining for Health Applications (#SMM4H) Workshop & Shared Task Proceedings of the Fourth Workshop August 2, 2019 Florence, Italy c 2019 The Association for Computational Linguistics Order copies of this and other ACL proceedings from: Association for Computational Linguistics (ACL) 209 N. Eighth Street Stroudsburg, PA 18360 USA Tel: +1-570-476-8006 Fax: +1-570-476-0860 [email protected] ISBN 978-1-950737-46-8 ii Preface Welcome to the 4th Social Media Mining for Health Applications Workshop and Shared Task - #SMM4H 2019. The total number of users of social media continues to grow worldwide, resulting in the generation of vast amounts of data. Popular social networking sites such as Facebook, Twitter and Instagram dominate this sphere. According to estimates, 500 million tweets and 4.3 billion Facebook messages are posted every day 1. The latest Pew Research Report 2, nearly half of adults worldwide and two- thirds of all American adults (65%) use social networking. The report states that of the total users, 26% have discussed health information, and, of those, 30% changed behavior based on this information and 42% discussed current medical conditions. Advances in automated data processing, machine learning and NLP present the possibility of utilizing this massive data source for biomedical and public health applications, if researchers address the methodological challenges unique to this media. In its fourth iteration, the #SMM4H workshop takes place in Florence, Italy, on August 2, 2019, and is co-located with the
    [Show full text]
  • Verbnet Based Citation Sentiment Class Assignment Using Machine Learning
    (IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 11, No. 9, 2020 VerbNet based Citation Sentiment Class Assignment using Machine Learning Zainab Amjad1, Imran Ihsan2 Department of Creative Technologies Air University, Islamabad, Pakistan Abstract—Citations are used to establish a link between time-consuming and complicated. To resolve this issue there articles. This intent has changed over the years, and citations are exists many researchers [7]–[9] who deal with the sentiment now being used as a criterion for evaluating the research work or analysis of citation sentences to improve bibliometric the author and has become one of the most important criteria for measures. Such applications can help scholars in the period of granting rewards or incentives. As a result, many unethical research to identify the problems with the present approaches, activities related to the use of citations have emerged. That is why unaddressed issues, and the present research gaps [10]. content-based citation sentiment analysis techniques are developed on the hypothesis that all citations are not equal. There are two existing approaches for Citation Sentiment There are several pieces of research to find the sentiment of a Analysis: Qualitative and Quantitative [7]. Quantitative citation, however, only a handful of techniques that have used approaches consider that all citations are equally important citation sentences for this purpose. In this research, we have while qualitative approaches believe that all citations are not proposed a verb-oriented citation sentiment classification for equally important [9]. The quantitative approach uses citation researchers by semantically analyzing verbs within a citation text count to rank a research paper [8] while the qualitative using VerbNet Ontology, natural language processing & four approach analyzes the nature of citation [10].
    [Show full text]
  • Full-Text Processing: Improving a Practical NLP System Based on Surface Information Within the Context
    Full-text processing: improving a practical NLP system based on surface information within the context Tetsuya Nasukawa. IBM Research Tokyo Resem~hLaborat0ry t623-14, Shimotsurum~, Yimmt;0¢sl{i; I<almgawa<kbn 2421,, J aimn • nasukawa@t,rl:, vnet;::ibm icbm Abstract text. Without constructing a i):recige filodel of the eohtext through, deep sema~nfiCamtlys~is, our frmne= Rich information fl)r resolving ambigui- "work-refers .to a set(ff:parsed trees. (.r~sltlt~ 9 f syn- ties m sentence ~malysis~ including vari- : tacti(" miaiysis)ofeach sexitencd in t.li~;~i'ext as (:on- ous context-dependent 1)rol)lems. can be ob- text ilfformation, Thus. our context model consists tained by analyzing a simple set of parsed of parse(f trees that are obtained 1)y using mi exlst- ~rces of each senten('e in a text withom il!g g¢lwral syntactic parser. Excel)t for information constructing a predse model of the contex~ ()It the sequence of senl;,en('es, olIr framework does nol tl(rough deep senmntic.anMysis. Th.us. pro- consider any discourse stru(:~:ure mwh as the discourse cessmg• ,!,' a gloup" of sentem' .., '~,'(s togethel. i ': makes.' • " segmenm, focus space stack, or dominant hierarclty it.,p.(,)ss{t?!e .t.9 !~npl:ovel-t]le ~ccui'a~'Y (?f a :: :.it~.(.fi.ii~idin:.(cfi.0szufid, Sht/!er, dgs6)i.Tli6refbi.e, om< ,,~ ~-. ~t.ehi- - ....t sii~ 1 "~g;. ;, ~-ni~chin¢''ti-mslat{b~t'@~--..: .;- . .? • . : ...... - ~,',". ..........;-.preaches ":........... 'to-context pr0cessmg,,' • and.m" "tier-ran : ' le d at. •tern ; ::'Li:In i. thin.j.;'P'..p) a (r ~;.!iwe : .%d,es(tib~, ..i-!: ....
    [Show full text]
  • Natural Language Processing
    Chowdhury, G. (2003) Natural language processing. Annual Review of Information Science and Technology, 37. pp. 51-89. ISSN 0066-4200 http://eprints.cdlr.strath.ac.uk/2611/ This is an author-produced version of a paper published in The Annual Review of Information Science and Technology ISSN 0066-4200 . This version has been peer-reviewed, but does not include the final publisher proof corrections, published layout, or pagination. Strathprints is designed to allow users to access the research output of the University of Strathclyde. Copyright © and Moral Rights for the papers on this site are retained by the individual authors and/or other copyright owners. Users may download and/or print one copy of any article(s) in Strathprints to facilitate their private study or for non-commercial research. You may not engage in further distribution of the material or use it for any profitmaking activities or any commercial gain. You may freely distribute the url (http://eprints.cdlr.strath.ac.uk) of the Strathprints website. Any correspondence concerning this service should be sent to The Strathprints Administrator: [email protected] Natural Language Processing Gobinda G. Chowdhury Dept. of Computer and Information Sciences University of Strathclyde, Glasgow G1 1XH, UK e-mail: [email protected] Introduction Natural Language Processing (NLP) is an area of research and application that explores how computers can be used to understand and manipulate natural language text or speech to do useful things. NLP researchers aim to gather knowledge on how human beings understand and use language so that appropriate tools and techniques can be developed to make computer systems understand and manipulate natural languages to perform the desired tasks.
    [Show full text]
  • Towards a Cross-Linguistic Verbnet-Style Lexicon for Brazilian Portuguese
    Towards a cross-linguistic VerbNet-style lexicon for Brazilian Portuguese Carolina Scarton, Sandra Alu´ısio Center of Computational Linguistics (NILC), University of Sao˜ Paulo Av. Trabalhador Sao-Carlense,˜ 400. 13560-970 Sao˜ Carlos/SP, Brazil [email protected], [email protected] Abstract This paper presents preliminary results of the Brazilian Portuguese Verbnet (VerbNet.Br). This resource is being built by using other existing Computational Lexical Resources via a semi-automatic method. We identified, automatically, 5688 verbs as candidate members of VerbNet.Br, which are distributed in 257 classes inherited from VerbNet. These preliminary results give us some directions of future work and, since the results were automatically generated, a manual revision of the complete resource is highly desirable. 1. Introduction the verb to load. To fulfill this gap, VerbNet has mappings The task of building Computational Lexical Resources to WordNet, which has deeper semantic relations. (CLRs) and making them publicly available is one of Brazilian Portuguese language lacks CLRs. There are some the most important tasks of Natural Language Processing initiatives like WordNet.Br (Dias da Silva et al., 2008), that (NLP) area. CLRs are used in many other applications is based on and aligned to WordNet. This resource is the in NLP, such as automatic summarization, machine trans- most complete for Brazilian Portuguese language. How- lation and opinion mining. Specially, CLRs that treat the ever, only the verb database is in an advanced stage (it syntactic and semantic behaviour of verbs are very impor- is finished, but without manual validation), currently con- tant to the tasks of information retrieval (Croch and King, sisting of 5,860 verbs in 3,713 synsets.
    [Show full text]
  • Entity Extraction: from Unstructured Text to Dbpedia RDF Triples Exner
    Entity extraction: From unstructured text to DBpedia RDF triples Exner, Peter; Nugues, Pierre Published in: Proceedings of the Web of Linked Entities Workshop in conjuction with the 11th International Semantic Web Conference (ISWC 2012) 2012 Link to publication Citation for published version (APA): Exner, P., & Nugues, P. (2012). Entity extraction: From unstructured text to DBpedia RDF triples. In Proceedings of the Web of Linked Entities Workshop in conjuction with the 11th International Semantic Web Conference (ISWC 2012) (pp. 58-69). CEUR. http://ceur-ws.org/Vol-906/paper7.pdf Total number of authors: 2 General rights Unless other specific re-use rights are stated the following general rights apply: Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal Read more about Creative commons licenses: https://creativecommons.org/licenses/ Take down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. LUND UNIVERSITY PO Box 117 221 00 Lund +46 46-222 00 00 Entity Extraction: From Unstructured Text to DBpedia RDF Triples Peter Exner and Pierre Nugues Department of Computer science Lund University [email protected] [email protected] Abstract.
    [Show full text]
  • Mining Text Data
    Mining Text Data Charu C. Aggarwal • ChengXiang Zhai Editors Mining Text Data Editors Charu C. Aggarwal ChengXiang Zhai IBM T.J. Watson Research Center University of Illinois at Urbana-Champaign Yorktown Heights, NY, USA Urbana, IL, USA [email protected] [email protected] ISBN 978-1-4614-3222-7 e-ISBN 978-1-4614-3223-4 DOI 10.1007/978-1-4614-3223-4 Springer New York Dordrecht Heidelberg London Library of Congress Control Number: 2012930923 © Springer Science+Business Media, LLC 2012 All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com) Contents 1 An Introduction to Text Mining 1 Charu C. Aggarwal and ChengXiang Zhai 1. Introduction 1 2. Algorithms for Text Mining 4 3. Future Directions 8 References 10 2 Information Extraction from Text 11 Jing Jiang 1. Introduction 11 2. Named Entity Recognition 15 2.1 Rule-based Approach 16 2.2 Statistical Learning Approach 17 3.
    [Show full text]
  • Leveraging Verbnet to Build Corpus-Specific Verb Clusters
    Leveraging VerbNet to build Corpus-Specific Verb Clusters Daniel W Peterson and Jordan Boyd-Graber and Martha Palmer University of Colorado daniel.w.peterson,jordan.boyd.graber,martha.palmer @colorado.edu { } Daisuke Kawhara Kyoto University, JP [email protected] Abstract which involved dozens of linguists and a decade of work, making careful decisions about the al- In this paper, we aim to close the gap lowable syntactic frames for various verb senses, from extensive, human-built semantic re- informed by text examples. sources and corpus-driven unsupervised models. The particular resource explored VerbNet is useful for semantic role labeling and here is VerbNet, whose organizing princi- related tasks (Giuglea and Moschitti, 2006; Yi, ple is that semantics and syntax are linked. 2007; Yi et al., 2007; Merlo and van der Plas, To capture patterns of usage that can aug- 2009; Kshirsagar et al., 2014), but its widespread ment knowledge resources like VerbNet, use is limited by coverage. Not all verbs have we expand a Dirichlet process mixture a VerbNet class, and some polysemous verbs model to predict a VerbNet class for each have important senses unaccounted for. In addi- sense of each verb, allowing us to incorpo- tion, VerbNet is not easily adaptable to domain- rate annotated VerbNet data to guide the specific corpora, so these omissions may be more clustering process. The resulting clusters prominent outside of the general-purpose corpora align more closely to hand-curated syn- and linguistic intuition used in its construction. tactic/semantic groupings than any previ- Its great strength is also its downfall: adding ous models, and can be adapted to new new verbs, new senses, and new classes requires domains since they require only corpus trained linguists - at least, to preserve the integrity counts.
    [Show full text]
  • Statistical Machine Translation with a Factorized Grammar
    Statistical Machine Translation with a Factorized Grammar Libin Shen and Bing Zhang and Spyros Matsoukas and Jinxi Xu and Ralph Weischedel Raytheon BBN Technologies Cambridge, MA 02138, USA {lshen,bzhang,smatsouk,jxu,weisched}@bbn.com Abstract for which the segments for translation are always fixed. In modern machine translation practice, a sta- However, do we really need such a large rule set tistical phrasal or hierarchical translation sys- to represent information from the training data of tem usually relies on a huge set of trans- much smaller size? Linguists in the grammar con- lation rules extracted from bi-lingual train- struction field already showed us a perfect solution ing data. This approach not only results in space and efficiency issues, but also suffers to a similar problem. The answer is to use a fac- from the sparse data problem. In this paper, torized grammar. Linguists decompose lexicalized we propose to use factorized grammars, an linguistic structures into two parts, (unlexicalized) idea widely accepted in the field of linguis- templates and lexical items. Templates are further tic grammar construction, to generalize trans- organized into families. Each family is associated lation rules, so as to solve these two prob- with a set of lexical items which can be used to lex- lems. We designed a method to take advantage icalize all the templates in this family. For example, of the XTAG English Grammar to facilitate the extraction of factorized rules. We experi- the XTAG English Grammar (XTAG-Group, 2001), mented on various setups of low-resource lan- a hand-crafted grammar based on the Tree Adjoin- guage translation, and showed consistent sig- ing Grammar (TAG) (Joshi and Schabes, 1997) for- nificant improvement in BLEU over state-of- malism, is a grammar of this kind, which employs the-art string-to-dependency baseline systems factorization with LTAG e-tree templates and lexical with 200K words of bi-lingual training data.
    [Show full text]