Ling7800: Advanced Computational Linguistics: Lexical Semantics

Spring 2010

Time and Location: Monday/Wednesday 11-12:15, Hellems 291, begining Jan 20, MUEN D 424, ICS small conference room (4th floor of MUEN, to the right)
Assessment: Presentation of two papers, two homeworks and a term project.
Office Hours: Monday 2:30-3:30 and Wednesday 3-4
Instructor: Martha Palmer
Textbook: Semantic Role Labeling (eBook), Martha Palmer, Daniel Gildea, Nianwen Xue, In Synthesis Lectures on Human Language Technologies , ed., Graeme Hirst, Morgan & Claypool, 2010. ISBN: 9781598298321


One of the great challenges of Natural Language Processing is the multitude of choices that language gives us for expressing the same thing in different ways. This is obviously true when taking other languages into consideration - the same thought can be expressed in English, French, Chinese or Russian, with widely varying results. But it is also true when considering a single language such as English. Light verb constructions, nominalizations, idioms, slang, paraphrases, and synonyms all give us myriads of alternatives for "coining a phrase."

With respect to other languages, one solution that has been often touted is that of an "interlingua:" a universal, language neutral semantic representation that all languages could be mapped onto. This approach has an immediate appeal, since it would obviate the need for specific translation systems for every possible pair of languages. Instead, it would only be necessary to build systems for each individual language that can produce the interlingua representation from an analysis of the sentences in the language, and that can generate fluent sentences from interlingua representations. As desirable as this may seem, and in spite of the tremendous effort that has gone into this quest, the realization of a suitable "interlingua" has proven to be elusive.

The students in this course will be encouraged to form their own opinion of the feasibility of hand-crafted semantic representations, as opposed to more automatic techniques that are purely statistical. We will explore in depth alternative styles of semantic representations, and compare and contrast their contributions to finding a useful, common semantic representation that can bridge lexical and structural gaps both mono-lingually and multi-lingually. We will look particularly closely at semantic role labeling as an instance of automatically assigning semantic representations. We will also explore alternative styles of semantic annotations and their cross-linguistic application.

Suggested Schedule and Readings - Open to Modification

Introduction and Module 1: the Lexical Semantics of Verbs - Chap 1

Module 2: Available Lexical Resources - Chap 2

Module 3: Representations of Events - Student Presentations Begin

Module 4: Automatic Semantic Role Labeling - Chap 3

Module 5. Prelimnary Student Project Presentations Continued

Module Last. Project Presentations

Module Extra. Possible Additional Paper Presentations

  • Additional Semantic Role Labeling papers from book bib
  • Innovative Approaches to SRL
  • Applications of SRL
  • Induction of Semantic Relations
  • WSD
  • NAACL03 Workshop on Text Meaning
  • HLT-NAACL 2003 ACL Anthology Web Page