[PDF] Proceedings of the Third Joint Conference on Lexical and

View - Download Proceedings of the Third Joint Conference on Lexical and

Previous PDF

Next PDF

Proceedings of the Third Joint Conference on

Lexical and Computational Semantics (*SEM 2014)

August 23-24, 2014

Dublin, Ireland

c ?2014 The *SEM 2014 Organizing Committee.

All papers

c?2014 their respective authors. This proceedings volume and all papers therein are licensed under a Creative Commons

Attribution 4.0 International License.

License details:http://creativecommons.org/licenses/by/4.0/

ISBN 978-1-941643-25-9

ii SEM 2014: Joint Conference on Lexical and Computational Semantics

Semantics (

?SEM)in Montreal 2012 and Atlanta 2013,?SEM provides a forum of exchange for the growing number of NLP researchers working on different aspects of semantic processing, which have been scattered over a large array of small workshops and conferences. The 2014 edition of ?SEM takes place in Dublin on August 23 and 24 and is collocated with SemEval and COLING. On this occasion, ?SEM and SemEval chose to coordinate their programs by featuring a joint invited talk.

In this way,

?SEM aims to bring together the ACL SIGLEX and ACL SIGSEM communities that in

present their top-tier research in computational semantics on this occasion. As in the previous editions

of ?SEM, the acceptance rate was very competitive. We accepted 22 papers (14 long and 8 short

papers) for publication at the conference, out of 49 long and 25 short paper submissions (resulting in

an overall acceptance rate of 29.7%). This is on par with some of the most competitive conferences in

computational linguistics. The papers cover a wide range of topics including formal and distributional

semantics, lexical semantics, discourse semantics, as well as application-oriented themes. We are confident these various contributions will set the stage for an inspiring conference. The ?SEM 2014 schedule consists of oral presentations for long papers and a poster session for short papers. Next to the accepted papers the ?SEM 2014 programme features the following highlights:

Day One, August 23th:

In the morning, a joint?SEM SemEval keynote address byMark Steedman;

In the afternoon, the poster session.

Day Two, August 24th:

In the morning, a keynote address byTimothy Baldwin; Finally, at the end of the day, a ceremony for the?SEM Best Paper Award.

As always,

?SEM 2014 would not have been possible without the considerable efforts of our area chairs and an impressive assortment of reviewers, drawn from the ranks of SIGLEX and SIGSEM, and the computational semantics community at large.

We hope you will enjoy

?SEM 2014, and look forward to engaging with all of you,

Johan Bos, University of Groningen, General Chair

Anette Frank, Heidelberg University, Program Co-Chair Roberto Navigli, Sapienza University of Rome, Program Co-Chairiii *SEM 2014 Chairs and Reviewers

General Chair

Johan Bos, University of Groningen

Program Co-Chairs

Anette Frank, Heidelberg University

Roberto Navigli, Sapienza University of Rome

Area Chairs

Anna Korhonen, University of Cambridge

Bernardo Magnini, Fondazione Bruno Kessler Trento

Katja Markert, University of Leeds

Gerard de Melo, Tsinghua University

Stephen Pulman, University of Oxford

Yannick Versley, Heidelberg University

Sabine Schulte im Walde, Stuttgart University

Publication Chairs

Valerio Basile, University of Groningen

Kilian Evang, University of Groningen

Program Committee

OmriAbend, EnekoAgirre, MariannaApidianaki, MarcoBaroni, PierpaoloBasile, BeataBeigman Klebanov, Charley Beller, Sabine Bergler, Chris Biemann, Gemma Boleda, Paul Buitelaar, Md. Faisal Mahbub Chowdhury, Philipp Cimiano, Paul Cook, Ann Copestake, Walter Daelemans, Ger- ard de Melo, Mona Diab, Greg Durrett, Katrin Erk, Stefan Evert, Ingrid Falk, Richárd Farkas, Afsaneh Fazly, Vanessa Wei Feng, Anette Frank, Matthew Gerber, Claudio Giuliano, Edward Grefenstette, Sanda Harabagiu, Karl Moritz Hermann, Graeme Hirst, Nancy Ide, Diana Inkpen, RaduIon, DaisukeKawahara, RomanKern, ManfredKlenner, OleksandrKolomiyets, MariaKout- sombogera, Alessandro Lenci, Omer Levy, Annie Louis, Bernardo Magnini, Suresh Manandhar, Katja Markert, Diana McCarthy, Pablo Mendes, Rada Mihalcea, Roser Morante, Preslav Nakov, Vivi Nastase, Roberto Navigli, Guenter Neumann, Hwee Tou Ng, Vincent Ng, Malvina Nissim, Tae-GilNoh, DiarmuidÓSéaghdha, SebastianPadó, MarthaPalmer, RebeccaJ.Passonneau, Mas- simo Poesio, Allan Ramsay, Aarne Ranta, Jonathon Read, Josef Ruppenhofer, Felix Sasaki, Yves Scherrer, Aitor Soroa, Caroline Sporleder, Mark Steedman, Mark Stevenson, Michael Strube, Lin Sun, Stefan Thater, Peter Turney, Tim Van de Cruys, Lonneke van der Plas, Marieke van Erp, Eva Maria Vecchi, Aline Villavicencio, Vinod Vydiswaran, Janyce Wiebe, Fei Xia, Annie Zaenen,

Luke Zettlemoyer, Michael Zockv

Table of Contents

More or less supervised supersense tagging of Twitter Anders Johannsen, Dirk Hovy, Héctor Martínez Alonso, Barbara Plank and Anders Søgaard..... 1 Generating a Word-Emotion Lexicon from #Emotional Tweets Anil Bandhakavi, Nirmalie Wiratunga, Deepak P and Stewart Massie ........................ 12 Improvement of a Naive Bayes Sentiment Classifier Using MRS-Based Features Jared Kramer and Clara Gordon.......................................................... 22
Sense and Similarity: A Study of Sense-level Similarity Measures Nicolai Erbs, Iryna Gurevych and Torsten Zesch ........................................... 30
An Iterative 'Sudoku Style" Approach to Subgraph-based Word Sense Disambiguation Steve L. Manion and Raazesh Sainudiin................................................... 40

Exploring ESA to Improve Word Relatedness

Nitish Aggarwal, Kartik Asooja and Paul Buitelaar......................................... 51

Identifying semantic relations in a specialized corpus through distributional analysis of a cooccurrence

tensor Gabriel Bernier-Colborne................................................................ 57

Learning the Peculiar Value of Actions

Daniel Dahlmeier....................................................................... 63
An analysis of textual inference in German customer emails Kathrin Eichler, Aleksandra Gabryszak and Günter Neumann ............................... 69
Text Summarization through Entailment-based Minimum Vertex Cover Anand Gupta, Manpreet Kaur, Shachar Mirkin, Adarsh Singh and Aseem Goyal .............. 75

Semantic Roles in Grammar Engineering

Wojciech Jaworski and Adam Przepiórkowski ............................................. 81
Semantic Role Labelling with minimal resources: Experiments with French Rasoul Kaljahi, Jennifer Foster and Johann Roturier........................................ 87
Compositional Distributional Semantics Models in Chunk-based Smoothed Tree Kernels Nghia The Pham, Lorenzo Ferrone and Fabio Massimo Zanzotto ............................ 93
Generating Simulations of Motion Events from Verbal Descriptions James Pustejovsky and Nikhil Krishnaswamy.............................................. 99
See No Evil, Say No Evil: Description Generation from Densely Labeled Images Mark Yatskar, Michel Galley, Lucy Vanderwende and Luke Zettlemoyer .................... 110
Extracting Latent Attributes from Video Scenes Using Text as Background Knowledge Anh Tran, Mihai Surdeanu and Paul Cohen............................................... 121
Using Text Segmentation Algorithms for the Automatic Generation of E-Learning Courses Can Özmen, Alexander Streicher and Andrea Zielinski .................................... 132
vii Cognitive Compositional Semantics using Continuation Dependencies William Schuler and Adam Wheeler..................................................... 141

Vagueness and Learning: A Type-Theoretic Approach

Raquel Fernandez and Staffan Larsson................................................... 151
Contrasting Syntagmatic and Paradigmatic Relations: Insights from Distributional Semantic Models Gabriella Lapesa, Stefan Evert and Sabine Schulte im Walde............................... 160
Dead parrots make bad pets: Exploring modifier effects in noun phrases Germán Kruszewski and Marco Baroni .................................................. 171
Syntactic Transfer Patterns of German Particle Verbs and their Impact on Lexical Semantics Stefan Bott and Sabine Schulte im Walde ................................................ 182
viii

Conference Program

Saturday, August 23

9:00-9:30 Welcome

9:30-10:30 Joint *SEM and SemEval keynote by Mark Steedman

Robust Semantics for NLP

10:30-11:00 Coffee break

11:00-11:30More or less supervised supersense tagging of Twitter

Anders Johannsen, Dirk Hovy, Héctor Martínez Alonso, Barbara Plank and Anders

Søgaard

11:30-12:00Generating a Word-Emotion Lexicon from #Emotional Tweets

Anil Bandhakavi, Nirmalie Wiratunga, Deepak P and Stewart Massie

12:00-12:30Improvement of a Naive Bayes Sentiment Classifier Using MRS-Based Features

Jared Kramer and Clara Gordon

12:30-14:00 Lunch break

14:00-14:30Sense and Similarity: A Study of Sense-level Similarity Measures

Nicolai Erbs, Iryna Gurevych and Torsten Zesch

14:30-15:00An Iterative 'Sudoku Style" Approach to Subgraph-based Word Sense Disambigua-

tion

Steve L. Manion and Raazesh Sainudiin

15:00-15:30 Coffee break

15:30-17:30 Poster session with lightning talks intro

Exploring ESA to Improve Word Relatedness

Nitish Aggarwal, Kartik Asooja and Paul Buitelaar

Identifying semantic relations in a specialized corpus through distributional analy- sis of a cooccurrence tensor

Gabriel Bernier-Colborne

Learning the Peculiar Value of Actions

Daniel Dahlmeier

ix

Saturday, August 23 (continued)

An analysis of textual inference in German customer emails Kathrin Eichler, Aleksandra Gabryszak and Günter Neumann Text Summarization through Entailment-based Minimum Vertex Cover Anand Gupta, Manpreet Kaur, Shachar Mirkin, Adarsh Singh and Aseem Goyal

Semantic Roles in Grammar Engineering

Wojciech Jaworski and Adam Przepiórkowski

Semantic Role Labelling with minimal resources: Experiments with French Rasoul Kaljahi, Jennifer Foster and Johann Roturier Compositional Distributional Semantics Models in Chunk-based Smoothed Tree Kernels Nghia The Pham, Lorenzo Ferrone and Fabio Massimo Zanzotto

Sunday, August 24

9:00-10:00 Keynote by Timothy Baldwin

Robust Semantics for NLP

10:00-10:30Generating Simulations of Motion Events from Verbal Descriptions

James Pustejovsky and Nikhil Krishnaswamy

10:30-11:00 Coffee break

11:00-11:30See No Evil, Say No Evil: Description Generation from Densely Labeled Images

Mark Yatskar, Michel Galley, Lucy Vanderwende and Luke Zettlemoyer

11:30-12:00Extracting Latent Attributes from Video Scenes Using Text as Background Knowledge

Anh Tran, Mihai Surdeanu and Paul Cohen

12:00-12:30Using Text Segmentation Algorithms for the Automatic Generation of E-Learning Courses

Can Özmen, Alexander Streicher and Andrea Zielinski

12:30-14:00 Lunch break

14:00-14:30Cognitive Compositional Semantics using Continuation Dependencies

William Schuler and Adam Wheeler

x

Sunday, August 24 (continued)

14:30-15:00Vagueness and Learning: A Type-Theoretic Approach

Raquel Fernandez and Staffan Larsson

15:00-15:30 Coffee break

15:30-16:00Contrasting Syntagmatic and Paradigmatic Relations: Insights from Distributional Se-

mantic Models Gabriella Lapesa, Stefan Evert and Sabine Schulte im Walde

16:00-16:30Dead parrots make bad pets: Exploring modifier effects in noun phrases

Germán Kruszewski and Marco Baroni

16:30-17:00Syntactic Transfer Patterns of German Particle Verbs and their Impact on Lexical Seman-

tics

Stefan Bott and Sabine Schulte im Walde

17:00-17:30 Best Paper Award and Closing

Invited Talks

Robust Semantics for NLP

Mark Steedman, University of Edinburgh

The paper presents a robust semantics for NLP applications that combines a (fairly) standard treat- ment of logical operators such as negation and quantification (Steedman 2012) with a paraphrase- and entailment-based semantics of relational terms derived from text data (Lewis and Steedman 2013a;

2013b). I"ll consider the extension of the latter component to temporal and causal entailment using

text-based methods, building on Lewis and Steedman 2014.

Lexical Semantic Analysis of Social Media

Timothy Baldwin, University of Melbourne

There has recently been a proliferation of research on Twitter and other social media, but is social media

simply a fad? I argue that it is an inherently different text source to those conventionally targeted in com-

putational linguistics, with unique challenges and opportunities for sub-fields including computational

lexical semantics. In doing so, I draw on recent work on user-level sense distributions and novel senses,

and point to unique features of social media which open up new challenges and opportunities for the field. xi

Proceedings of the Third Joint Conference on Lexical and Computational Semantics (*SEM 2014), pages 1-11,Dublin, Ireland, August 23-24 2014.More or less supervised supersense tagging of Twitter

Anders Johannsen, Dirk Hovy, H

´ector Mart´nez Alonso, Barbara Plank, Anders Søgaard

Center for Language Technology

University of Copenhagen, Denmark

Njalsgade 140

ajohannsen@hum.ku.dk, dirk@cst.dk, alonso@hum.ku.dk plank@cst.dk, soegaard@hum.ku.dk

Abstract

We present two Twitter datasets annotated

with coarse-grained word senses (super- senses), as well as a series of experiments with three learning scenarios for super- sense tagging: weakly supervised learn- ing, as well as unsupervised and super- vised domain adaptation. We show that (a) off-the-shelf tools perform poorly on

Twitter, (b) models augmented with em-

beddings learned from Twitter data per- form much better, and (c) errors can be reduced using type-constrained inference with distant supervision from WordNet.

1 Introduction

Supersense tagging (SST, Ciaramita and Altun,

2006) is the task of assigning high-level ontolog-

ical classes to open-class words (here, nouns and verbs). It is thus a coarse-grained word sense dis- ambiguation task. The labels are based on the lexi- cographer file names for Princeton WordNet (Fell- baum, 1998). They include 15 senses for verbs and 26 for nouns (see Table 1). While WordNet also provides catch-all supersenses for adjectives and adverbs, these are grammatically, not seman- tically motivated, and do not provide any higher- level abstraction (recently, however, Tsvetkov et al. (2014) proposed a semantic taxonomy for ad- jectives). Theywillnotbeconsideredinthispaper.

Coarse-grained categories such as supersenses

are useful for downstream tasks such as question- SST is different from NER in that it has a larger set of labels and in the absence of strong orthographic cues (capitalization, quotation marks, etc.). More- over, supersenses can be applied to any of the lex- ical parts of speech and not only proper names.

Also, while high-coverage gazetteers can be found

for named entity recognition, the lexical resources available for SST are very limited in coverage.Twitter is a popular micro-blogging service, which, among other things, is used for knowledge sharing among friends and peers. Twitter posts (tweets) announce local events, say talks or con- certs, present facts about pop stars or program- ming languages, or simply express the opinions of the author on some subject matter.

Supersense tagging is relevant for Twitter, be-

cause it can aid e.g. QA and open RE. If someone posts a message saying that some LaTeX module now supports "drawing trees", it is important to know whether the post is about drawing natural objects such as oaks or pines, or about drawing tree-shaped data representations.

This paper is, to the best of our knowledge, the

first work to address the problem of SST for Twit- ter. While there exist corpora of newswire and literary texts that are annotated with supersenses, e.g., SEMCOR(Miller et al., 1994), no data is available for microblogs or related domains. This paper introduces two new data sets.

Furthermore, most, if not all, of previous work

on SST has relied on gold standard part-of-speech (POS) tags as input. However, in a domain such as Twitter, which has proven to be challenging for POS tagging (Foster et al., 2011; Ritter et al., 2011), results obtained under the assumption of available perfect POS information are almost meaningless for any real-life application.

In this paper, we instead use predicted POS tags

and investigate experimental settings in which one or more of the following resources are available to us: a large corpus of unlabeled Twitter data;

Princeton WordNet (Fellbaum, 1998);

SEMCOR(Miller et al., 1994); and

a small corpus of Twitter data annotated with supersenses.

We approach SST of Twitter using various de-

grees of supervision for both learning and domain adaptation (here, from newswire to Twitter). In1 weakly supervised learning, onlyunlabeleddata and the lexical resource WordNet are available to us. While the quality of lexical resources varies, this is the scenario for most languages. We present an approach to weakly supervised SST based on type-constrained EM-trained second-order HMMs (HMM2s) with continuous word representations.

In contrast, when usingsupervisedlearning, we

can distinguish between two degrees of supervi- sion for domain adaptation. For some languages, e.g., Basque, English, Swedish, sense-annotated resources exist, but these corpora are all limited to newswire or similar domains. In such lan- guages,unsupervised domain adaptation(DA) techniques can be used to exploit these resources.

The setting does not presume labeled data from

the target domain. We use discriminative mod- els for unsupervised domain adaptation, training on SEMCORand testing on Twitter.

Finally, we annotated data sets for Twitter, mak-

ingsupervised domain adaptation(SU) exper- iments possible. For supervised domain adapta- tion, we use the annotated training data sets from both the newswire and the Twitter domain, as well as WordNet.

For both unsupervised domain adaptation and

supervised domain adaptation, we use structured perceptron (Collins, 2002), i.e., a discriminative

HMM model, and search-based structured predic-

tion (SEARN) (Daume et al., 2009). We aug- ment both the EM-trained HMM2, discrimina- tive HMMs and SEARNwith type constraints and continuous word representations. We also exper- imented with conditional random fields (Lafferty et al., 2001), but obtained worse or similar results than with the other models.quotesdbs_dbs25.pdfusesText_31

[PDF] Barbaramarkt in Bietingen Geschenkidee zu Weihnachten

[PDF] barbara_kruger-3 ( PDF

[PDF] BARBARIAN Invader - Gestion De Données

[PDF] Barbarie n°17 – Hiver 2013/2014 - France

[PDF] BARBATRE Stylisme de mode - 7,12,14,21 jours été 2015 - Gestion De Projet

[PDF] barb?tus barbu. I. 1. Humains : Fr. barbé adj. « barbu, portant la - France

[PDF] barbe a papa stand gonflable 1 300 € ht - Anciens Et Réunions

[PDF] BARBECUE - Cuisine Viking - Conception

[PDF] Barbecue à gaz Koala - Matériel

[PDF] BARBECUE A GAZ NOMADE BQ400E

[PDF] barbecue à gaz portable - Fabrication

[PDF] Barbecue CHARCOAL CLASSIC 2000 - Matériel

[PDF] Barbecue de brousse Feu de camp Plats familiaux Légumes et - Anciens Et Réunions

[PDF] barbecue geant - Francecreationparis

[PDF] BARBECUE GEANT Samedi 11 Juin 2011 À 19h Bulletin réponse