LSD Sem/LSDSem 2015/LSD-Sem
Linking Models of Lexical, Sentential and Discourse-level Semantics

Workshop to be held in conjunction with EMNLP in Lisbon, Portugal

Endorsed by SIGLEX, SIGSEM and SIGdial, the ACL Special Interest Groups on
the Lexicon, Computational Semantics, and Discourse and Dialogue

Improved computational models of semantics hold great promise for applications in language technology, be it semantics at the lexical level, sentence level or discourse level. Large-scale corpora with corresponding annotations (word senses, propositions, attributions and discourse relations) are making it possible to develop statistical models for many tasks and applications. However, developments in lexical and sentence-level semantics remain largely distinct from those in discourse semantics. This workshop aims at bridging this gap by bringing together researchers to discuss how multiple levels of semantics can be integrated and implemented in various applications.


Early linguistic studies considered the interplay between discourse context and meaning at the sentence level, showing inter alia that information structure impacts syntactic realization and that lexical and pragmatics factors affect the realization of verb arguments. Although early computational work involving semantics took some account of discourse, modern approaches to semantic processing are only now beginning to consider it: semantic parsing with context, predicting argument realization and linking implicit arguments, disambiguating words using author and discourse information, and classifying paradigmatic relations between words.

It is clear that interactions between semantic phenomena within sentences and on the discourse level are not unidirectional: Factors such as modality, negation and syntactic patterns have shown to correlate strongly with the sense of discourse relations. Similarly, situational entities mentioned within a sentence can indicate specific discourse modes. Complementing discourse-aware models of semantics are semantically-informed methods for text-level processing, such as discourse parsers that rely on models of rich lexical representations.

Many applications of natural language processing can benefit from bringing together models of semantics on the word, sentence and discourse level. Recent advances can be seen for instance in recognizing textual entailment, sentiment analysis, information extraction, event ordering, and question answering. Besides traditional applications of natural language processing, researchers in cognitive science have started to explore different levels of representation for computing semantic similarity between paragraphs and between documents.


This workshop aims to gather and showcase theoretical and computational approaches to joint models of semantics, and applications that incorporate multi-level semantics. We hope to bring together researchers from various areas: linguists and cognitive scientists working on aspects of representing text with multiple levels of semantics, machine learning researchers interested in joint inference over different types of semantic cues, and also researchers who are interested in applications that require or will benefit from multi-level semantics. A dialog between researchers has great potential to advance work in each of these areas and bring about more powerful and enriched models of text semantics.

Topics of interest include, but are not limited to:

Michael Strube

Heidelberg Institute for Theoretical Studies
The (Non)Utility of Semantics for Coreference Resolution [PDF]

Jacob Eisenstein

Georgia Institute of Technology
From Distributed Semantics to Discourse, and Back [PDF]

Rada Mihalcea

University of Michigan
What Men Say, What Women Hear: Using Semantics To Make Better Sense of Gender Differences in Social Media

9:00 Opening
9:05 Invited talk
From Distributed Semantics to Discourse, and Back
Jacob Eisenstein
9:50 An Exploration of Discourse-Based Sentence Spaces for Compositional Distributional Semantics
Tamara Polajnar, Laura Rimell, Stephen Clark
10:05 Linking discourse modes and situation entity types in a cross-linguistic corpus study
Kleio-Isidora Mavridou, Annemarie Friedrich, Melissa Peate Sørensen, Alexis Palmer, Manfred Pinkal
10:20 Recovering discourse relations: Varying influence of discourse adverbials
Hannah Rohde, Anna Dickinson, Chris Clark, Annie Louis, Bonnie Webber
10:30 Coffee break
11:00 Semantics and Discourse Processing for Expressive TTS
Rodolfo Delmonte and Rocco Tripodi
11:15 Semantically Enriched Models for Modal Sense Classification
Mengfei Zhou, Anette Frank, Annemarie Friedrich, Alexis Palmer
11:30 Identification and Disambiguation of Lexical Cues of Rhetorical Relations across Different Text Genres
Taraneh Khazaei, Lu Xiao, Robert Mercer
11:45 Bridging Sentential and Discourse-level Semantics through Clausal Adjuncts
Rashmi Prasad, Bonnie Webber, Alan Lee, Sameer Pradhan, Aravind Joshi
11:55 Lexical Level Distribution of Metadiscourse in Spoken Language
Rui Correia, Maxine Eskenazi, Nuno Mamede
12:05 Idiom Paraphrases: Seventh Heaven vs Cloud Nine
Maria Pershina, Yifan He, Ralph Grishman
12:15 Where Was Alexander the Great in 325 BC? Toward Understanding History Text with a World Model
Yuki Murakami and Yoshimasa Tsuruoka
12:25 Lunch break
14:00 TextLink: EU COST action on Structuring Discourse in a Multi-Lingual Europe
Bonnie Webber
14:05 Invited talk
What Men Say, What Women Hear: Using Semantics To Make Better Sense of Gender Differences in Social Media
Rada Mihalcea
14:50 Predicting word sense annotation agreement
Héctor Martínez Alonso, Anders Johannsen, Oier Lopez de Lacalle and Eneko Agirre
15:00 Distributional Semantics in Use
Raffaella Bernardi, Gemma Boleda, Raquel Fernandez and Denis Paperno
15:10 Poster session
15:30 (including coffee break)
16:00 Invited talk
The (Non)Utility of Semantics for Coreference Resolution
Michael Strube
16:45 Panel discussion
Is there a future for joint models of semantics or shall we stay separate?
17:30 Closing
All deadlines refer to 11:59pm (UTC/GMT -11 hours).

Program Committee