Semantic labelling for NLP tasks
In Association with
Motivation and Aims
Although it is generally assumed that improvements in language processing will be made through the integration of linguistic information and statistical techniques, the reality is that language is very diverse and looking for specific patterns of words that repeat enough to be statistically significant tends not to be a very fruitful task: sequences longer than three words are not generally repeated often enough to be statistically significant. At the same time, the identification of named entities: names, dates, places, organizations etc., has proved to be a very usefulpreliminary task in many natural language processing systems. We are interested in pursuing approaches which extend this notion by identifying and labeling other semantic information in a text, in such a way as to allow repeatable semantic patterns to emerge. Our interest is in attacking the data sparseness problem by exploring ways to collapse (semantically) related phrases which are expressed by different word sequences.
As this seems closely related to previously proposed class-based language models (see for example Brown et al. 90 in Computational Linguistics), it is different in that the empirical notion of classes used in the previous work (e.g. classes made up of collocationally similar words) are replaced by semantically justified sets.
Notice how Name Entity (NE) tagging and Word Sense Disambiguation (WSD) represent, in terms of granularity and representational complexity, two extremes of a single general problem: semantic disambiguation. Semantic disambiguation serves thus the purpose of improving the generalization power of statistical models. One of the questions here is how to determine a suitable level of clustering (for NE identification and for WSD) that would lead to high accuracy and to performance improvement by obtained statistical models.
Reason of Interest
It is to be noticed that several independent research efforts that focused recently on the statistical treatment of semantic phenomena (e.g. WordNet navigation as a stochastic process, as studied in Light and Abney or in Ciaramita & Johnson, 2003) correlated highly with the research program proposed above.
The workshop will offer a forum where experience from lexical semantics and statistical learning will be presented and fruitfuldiscussion among researchers in both fields will be promoted. The workshop is expected to attract researchers and practitioners from a range of areas as well as developers of large scale semantic resources who are interested in effective methods of semantic labeling.
Topics (to be addressed in the workshop include,
but are not limited to)
The workshop will be a half-day event with position statements from invited speakers (half an hour each) with two hours for 4-6 presentations of scientific papers. Submissions are intended to present works in progress and more completed works which fall within the scope defined by the topics listed above. A final 1 hour open discussion among all the workshop participants will be moderated by the organizers. In order to stimulate an interesting general discussion, each member of the program committee will be invited to submit a position statement of max. 1000 words.
Participants are invited to submit an extended abstract of max. 3500 words concerning one or more of the topics of interest. Each accepted paper receives a slot of 25 minutes for presentation (15 minutes talk and 10 minutes for discussion). Each submission should show: title; author(s); affiliation(s); and contact author's e-mail address, postal address, telephone and fax numbers. Submissions must be sent electronically in PDF to the following adddress:
Dept. of Computer Science, Systems and Management
Proceedings and Publications
Proceedings of the workshop will be printed by the LREC Local Organising Committee.
Organizers are negotiating for the publications of a special issue on “Semantic tagging/labelling for NLP tasks” with the Computer Speech and Language Journal and selected papers will appear on in that issue.
Louise Guthrie - University of Sheffield, UK
Roberto Basili - University of Rome, Tor Vergata, Italy
Eva Hajicova - Charles University, Czech Republic
Fred Jelinek - Johns Hopkins University, Maryland, USA
For any information related to the organization, please contact:
Computer Science, Systems and Management
+39 06 72597391