Roberto Basili (basili@info.utovrm.it)
Mon, 05 Jan 1998 13:51:29 +0100
[Apologies to those of you who receive multiple copies] ECML-98 Workshop: First Call for Papers ECML-98 Workshop: Towards adaptive NLP-driven systems: linguistic information, learning methods and applications Organized by : R. Basili, M .T. Pazienza (University of Roma, Tor Vergata), ITALY Since most of the applications, from syntactic to semantic, are lexicon driven, systematic and reliable acquisition on a large scale of linguistic information is the real challenge to Natural Language Processing (NLP). Empiricist view on Natural Language Processing and Learning has become recently more attractive for a wider research community: computational linguistics, artificial intelligence, psychology then seemed to converge on a specific data-oriented perspective aiming to overcome the traditional knowledge acquisition bottleneck. It has been often noted that the limited attention paid by the machine learning community to text and speech data seems unjustified. It is thus more and more evident that empirical learning of Natural Language Processing (NLP) can alleviate the NLP main problem by means of a variety of methods for the automatic induction of lexical knowledge. Lexical knowledge is often hard to compile by hand, and even harder to port and reuse. NLP application systems have still a low impact on real world problems, mainly due to the costs related to reusability and customization of the required lexicons. In particular changes in the domain, causes changes in the lexical information required in the underlying natural language. Empirical, symbolic machine learning methods can be perfectly suited for this task like automatic acquisition and adaptation of this klnowledge. Rule induction, symbolic approaches to clustering, lazy learning, and inductive logic programming, have been already proposed by a growing community that is entering the challenge for theoretical (i.e. methodological) and application purposes A variety of techniques seems to be combined in order to successfully design realistic inductive systems for text processing: the target of this research are methodological and design principles for systems combining linguistic and lexical learning capabilities for large scale language processing tasks. This is what we mean with adaptive NLP-driven systems. Within this research enterprise, some issues can favour a sinergistic process between NLP and ML areas: the access to large data sets, that are even increasing over time, due to the telematics facilities available nowaday; extending the set of typical classes of ML problems to other hard cases (particularly dense in the NLP processes); adding inductive capabilities to NLP system for tasks related to specific applications (i.e. Information Extraction). The proposed Workshop is thus aiming to stimulate reasearch and discussion on the following aspects : - Establishing results and evidencies on the suitability of different ML paradigms on specific levels of representation of lexical knowledge (morphology, syntax, linguistic inference among others) - Comparison of the quantitative approaches to lexical acquisition with empirical symbolic methods - Stimulating discussion on cognitive perspective of some models within a plausible architecture for Language Processing and Learning - Establishing results on the applicability of the extracted/induce knowledge within NLP systems, with respect to assessed evaluation criteria, typical of the ML and Language Engineering (LE) area - Case studies on adaptive NLP systems, i.e. effective NLP systems integrating linguistic inferences with inductive capabilities (WWW KB at CMU, ECRAN), - Critical review of existing experiences on adaptive NLP systems - Establishing guidelines for an evaluation framework of adaptive NLP systems : accuracy of the linguistic process, robustness of the induction process, ... - Promote cooperation among research groups in Europe and USA to exchange ideas, data and tools for design and experiment architectures for adaptive NLP systems WorkShop format : The Workshop is expected to cover the whole day. In the first session, a part from an invited talk, we expect to cover methodological issues. Papers related to advanced research on suitability of learning paradigms for the different target lexical information will be favoured. Prototypical examples in this area are studies on empirical learning of tasks like POS tagging, induction of grammatical information, symbolic learning of word sense disambiguation criteria and lexical semantic information. A panel discussion is expected to close the morning session and focus on principles of suitability for learning paradigms vs. lexical levels. In the second half of the day we expect to stimulate partecipants to cover application areas, like IR and IE, by a couple of invited talks on existing adaptive systems as a basis for presenting novel aspects on integration of NLP capabilities with learning from experience (examples, errors, performance). A set of at least other 3 or 4 papers is expected to concentrate on original research works that we know are currently under development in several reasearch centres in Europe (Sheffield University, Tilburg, Rome Tor Vergata and Torino University). A Panel discussion on the implication of the adaptive paradigm on existing and potential NLP systems will close the Workshop. Program Committee R. Basili (University of Roma, Tor Vergata, ITALY) M. Craven (Carnegie Mellon University, USA) W. Daelemans (University of Tilburg, NEDERLANDS) M.T. Pazienza (University of Roma, Tor Vergata, ITALY) L. Saitta (University of Torino, ITALY) C. Samuelssonn (Bell Labs, AT&T, USA) Y. Wilks (University of Sheffield, UK) Paper Submission: ============ Papers should not exceed 3000 words or 6 pages Hard Copy Submission: Three copies of the paper should be sent to: Roberto Basili Department of Computer Science, Systems and Production University of Roma, Tor Vergata Via di Tor Vergata 00133 Roma (ITALY) e-mail: basili@info.utovrm.it Electronic Submission: Electronic submission may be in either self-contained Postscript or RTF formats, to basili@info.utovrm.it For each submission -- whether hard copy or electronic -- a separate plain ascii text email message should be sent to Roberto Basili, containing the following information: # NAME : Name of first author # TITLE: Title of the paper # PAGES: Number of pages # FILES: Name of file (if attachments are submitted electronically) # NOTE : Any relevant instructions # KEYS : Keywords # EMAIL: Email of the first author # ABSTR: Abstract of the paper . . . . . . Timetable: Workshop Announcement and Call for Papers : 5 January 1998 Papers due : 15 February 1998 Notification of Acceptance : 5 March 1998 Final version due : 25 March 1998 ==== cut here ==== ------------------------------------------------------ Roberto Basili Department of Computer Science, Systems and Production University of Roma, Tor Vergata Via di Tor Vergata 00133 Roma (ITALY) e-mail: basili@info.utovrm.it tel: +39 - 6 - 7259 7391 fax: +39 - 6 - 7259 7460 ------------------------------------------------------
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:38:16 PST