4th International Summer School in Computational & Mathematical Linguistics (sponsored by CNUCE & IBM Italia)

Pisa

Aug 5 – Aug 26 77

Up to 200 people attended the Summer School, mostly from Italy, France, USA, Canada and W.Germany; smaller national groupings included the English, Belgian, Dmtch, Czech, Yugoslavian and E.German (all about half a dozen each). There were also at least one Pole, one Vietnamese and an Indian from Moscow. Over two-thirds of those present were linguists and probably half of these had no computational experience. Almost all were engaged in academic research, mostly of post-doctoral or doctoral level. The summer school consisted of a dozen forma1 courses of lectures of varying lengths, workshops and informal discussions.

The courses reflected the interests of the majority in that the strictly computational element was slim. The star in this respect was M.K.Halliday (long regarded as the Grand Old Man of modern linguistics) whose course demonstrated the enormous complexity of understanding and producing natural language, and the complete lack of homogeneity in the typos of knowledge to do so. By contrast, Yorick Wilks (equally undoubtedly the Machine Intelligence Superstar of the Seventies) gave an optimistic and remarkably thorough overview of the various systems for understanding natural language developed over the last ten years. Wilks (a philosopher by training) has a witty lecturing style and maintains a lordly indifference to the petty squabbles of nomenclature which bedevil the field. He brought out the intrinsic similarities of much recent work, and stressed that the relation between these and recent developments in theoretical linguistics was closer than suspected. Wilks' course was complemented by a two-week course taught (in a ruthlessly pedagogic manner) by Luc Steels (U of Antwerp, now at MIT).Though called 'an introduction to computational linguistics', -this course in fact simply presented in some detail the formalisms used to express meaning in the field and gave more technical details of many of the systems described by Wilks. Typical material covered included LISP type lists, phrase-structure grammars, case-structure grammars, semantic structures (eg. Schank's "conceptual dependency" graphs, Winograd's PLANNER) and, most important, transition networks. Steels gave sufficient detail of how different parsing systems operated to suggest that implementation was comparatively trivial, and also discussed in more general terms other cognitive formalisms such as frames, scripts and the Knowledge Representation Language of Bobrow & Winograd. A specific example of a transition network was discussed in another course, given by Ron Kaplan (Xerox, Palo Alto) who, with R. Woods, designed the first successful ATN-based parser and has now expanded it considerably. (An ATN is a method of representing a given syntax or grammar in a Transition Network; the A stands for Augmented, and indicates that some or all of the transitions may include predicate tests, case functions, structure building acts etc., thus including a semantic component in the system)

In addition to these full length courses, I attended parts of a course given by George Lakoff (UCLA,Berkeley) on linguistic "gestalts" -a new concept in theoretical linguistics suspiciously similar to the AI "frame" concept , an informal discussion at which I spoke on the Oxford Archive, and an introductory LISP course. I also found a couple of potential Oxeye users and a lot of texts for the Archive. There was an attempt to organise a football match "Linguists v. Computers", but it failed through lack of computational support.

Approx 5 kg. of printed paper came put of the Summer School, and is on its way hither if anyone wishes to know more about it. I also have course notes for Steels' course and a useful bibliography.