Mathematical and Computational Analysis of Natural Language: Selected Papers From the 2nd International Conference on Mathematical Linguistics, Tarragona, 2-4 May 1996, and: Issues in Mathematical Linguistics: Workshop on Mathematical Linguistics, State College, PA, April 1998 (review)

These two volumes contain, respectively, 24 and 9 papers by disjoint sets of contributors, all of whose addresses are given. Space precludes mentioning more than a few.

The 1998 volume is more purely mathematical. Several papers address parsing, translation, and algebraic semantics. Two unusual topics that caught my eye are Shinya Matsunami, ‘A mathematical model for the post-creole continuum in Hawaii’ (327–40), which formalizes the notion of ‘overlapping lects’, and Bilge Say and Varol Akman, ‘An information-based treatment of punctuation in discourse representation theory’ (359–73), which brings punctuation within formal semantics and gives a useful bibliography.

The 1999 volume contains a long, interesting, and provocative paper, ‘Some uses and abuses of mathematics in linguistics’, by Alexis Manaster Ramer (73–130), unfortunately marred by typographical fumbles. For example, on p. 115, ‘76% centuries’ should be ‘76 centuries’, and on p. 105, the quote from Joseph Greenberg and Merritt Ruhlen ends one paragraph earlier than the quotation marks indicate. Manaster Ramer’s contention is that linguists pay undue reverence to mathematical work and in particular that they are often taken in by good mathematics badly applied.

Tackling arguments for and against the context-freeness of natural language syntax, Manaster Ramer points out something that, in retrospect, is obvious: Syntax is not the study of sets of strings; it is the study of strings plus their linguistic functions. Consider for example the argument that Zürich German is noncontext-free because it has serial verbs with abab agreement sequences. Is this argument refuted by the fact that NPs can be put in as optional adjuncts, thus expanding the set of strings into something that can be generated by a context-free grammar? Manaster Ramer says no; the point is, the context-free grammar may generate the right strings, but there is no way it can describe their functions correctly; it has to ignore a linguistically significant distinction.

After brief remarks on language learnability, Manaster Ramer moves on to the factor of chance in language comparison, and in particular the pioneering probability studies of Donald Ringe. As I understood it, Ringe’s intent was to open a long-neglected can of worms, not to have the last word; that Ringe made simplifying assumptions is hardly news, nor does it vitiate his arguments. Thus, some of Manaster Ramer’s criticism seems to me to miss the point, but he and Ringe would surely agree that the probability of spurious language grouping due to chance similarity is appreciable.