Economy in formulating typological generalizations

Abstract

Typically, a linguistic typology deÞnes all logically possible types and states which of these types are actually attested and which are not. The task then is to describe such a typology, preferably in the most economic way. In this paper, a descriptive principle is justiÞed to the effect that for any typology with at least one unattested type there exists a minimal description, consisting of a conjunction of non-statistical (implicational) Greenberg's Appendix II (1966) and Hawkins' Expanded Sample. Hawkins (1983) has noticed that Greenberg's universals do not describe all and only the attested types in Appendix II, but our analysis shows that Hawkins himself has not been fully successful in describing the typology in his Expanded Sample either. universals, deÞning all and only the attested types. A method is proposed that Þnds the minimal description(s) of a typology, and a computer program is sketched that executes the method, illustrating it on the typologies in

Language Universals: Their Discovery and Verbalization by Machine

“…Indeed, a linguistic typology is equivalent to a propositional function, and therefore, as known from propositional logic, for any propositional function there exists a propositional expression that generates it. As a corollary, for any linguistic typology there exists a set of non-statistical universals, describing all and only its attested types (Pericliev 2002).…”

An Appraisal of UNIVAUTO – The First Discovery Program to Generate a Scientific Article

“…A common pragmatic criterion for evaluating discovery systems is the publication of their discoveries in the specialized domain literature. According to this criterion, UNIVAUTO performs well: its outputs have found outlet in several linguistic publications ( [11], [13], [15], [16] (the latter under submission)).…”

“…Indeed, a linguistic typology is equivalent to a propositional function, and therefore, as known from propositional logic, for any propositional function there exists a propositional expression that generates it. As a corollary, for any linguistic typology there exists a set of non-statistical universals, describing all and only its attested types ( [15]). Secondly, our system found alternative sets of (non-statistical) universals describing the same typology.…”

A linguistic discovery program that verbalizes its discoveries

“…Also, UNIV can compute (implicational) universals valid in (at least) a user-specified number of positive examples (=languages), as well as compute the statistical significance of universals (based on the χ 2 statistic). A minimal set-cover subroutine may guarantee the discovery of the smallest set(s) of universals, generating a typology (Pericliev 2002).…”

“…UNIVAUTO was run on various data sets (word order, phonology, morpho-syntax), with very promising linguistic results. The published outcomes of UNIVAUTO so far include : two whole journal articles Pericliev (1999Pericliev ( , 2000 based on data from Greenberg (1966) (with no post-editing, the first one with no disclosure of articles' "machine origin"); around 50 statistically significant phonological universals based on Maddieson's UPSID-451 database, published without post-editing at the Universals Archive at the University of Konstanz; the substance discovery (rather than verbalization) part of Pericliev (2002). To the best of our knowledge, this is the first computer program to generate a whole scientific article.…”