Motivation and Aims

Provision of large-scale labelled language resources, such as tagged corpora or repositories of pre-classified text documents, is a crucial key to steady progress in an extremely wide spectrum of research, technological and business areas in the HLT sector. The continuously changing demands for language-specific and application-dependent annotated data (e.g. at the syntactic or at the semantic level), indispensable for design validation and efficient software prototyping, however, are daily confronted by the labelled-data bottleneck. Hand-crafted resources are often too costly and time-consuming to be produced at a sustainable pace, and, in some cases, they even exceed the limits of human conscious awareness and descriptive capability.

Possible ways to circumvent, or at least minimise, this problem come from the literature on automatic knowledge acquisition and, more generally, from the machine-learning community. Annotated data are bootstrapped by training a machine-learning classifier with a small sample of pre-annotated data and by using the induced classifier to annotate more data. Co-learning provides an alternative methodology, which essentially consists in iterative cooperation of two or more independent learning systems. Another promising route consists in automatically tracking down recurrent knowledge patterns in unstructured or implicit information sources (such as free texts or machine readable dictionaries) for this information to be moulded into explicit representation structures (e.g. subcategorisation frames, syntactic-semantic templates, ontology hierarchies etc.).

We believe that all these attempts at bootstrapping labelled data are not only of practical interest (for continuous updating, management and validation of dynamic resources), but also point to a bunch of germane theoretical issues. In particular, the workshop intends to focus on the issue of interaction between techniques for inducing structured knowledge from raw data and formal methods of linguistic knowledge representation. Gaining insights into this issue is an essential requirement for explaining the effective use of linguistic knowledge by cognitive agents. Although the cognitive and engineering views of the form and acquisition of linguistic knowledge need not be related, data from neuroscience and psychology are indeed relevant when evaluating different ways of representing information in artificial systems, and different models for linguistic knowledge acquisition.

We encourage in-depth analysis of underlying assumptions of the proposed bootstrapping methods and discussion of possible relevant connections with existing annotation and representation schemes. This investigation is likely to have significant repercussions on the way linguistic resources will be designed, developed and used for applications in the years to come. As the two aspects of knowledge representation and acquisition are profoundly interrelated, progress on both fronts can only be achieved, in our view of things, through a full appreciation of this deep interdependency.

Topics of Interest

Possible themes for participation are:

• Development of 'data-driven' annotation/representation schemes;
• Dynamic update, customisation and tuning of labelled resources through acquired data;
• 'Hybrid models' of linguistic knowledge extraction, whereby machine learning methods are integrated with formal structures of knowledge representation;
• Incremental linguistic knowledge-bases;
• Formal representation and structuring of information flow automatically acquired from texts;
• Knowledge acquisition and linguistic resources lifecycle;
• Linguistic knowledge acquisition and representation in cognitive tasks.

Important Dates

Submissions

The organizers welcome contributions describing existing research related to the topics of the workshop. Each presentation will be 25 minutes long (20 minutes for presentation and 5 minutes for questions and discussion). Submissions should include: title; author(s); affiliation(s); and contact author's e-mail address, postal address, telephone and fax numbers.
Abstracts (maximum 500 words, plain-text format) must be sent to: simo@ilc.pi.cnr.it

The final version of the accepted papers should not be longer than 4,000 words or 10 A4 pages. Instructions for formatting and presentation of the final version will be sent to authors upon notification of acceptance.

Organising Committee

Provisional Programme Committee

Contact Person

Workshop Registration Fees

• If you are not attending LREC: 140 EURO
• If you are attending LREC: 90 Euro