Ivan A. Sag (sag@skol.stanford.edu)
Tue, 5 Sep 95 14:13:06 PDT
Mark Johnson writes, >>... it seems clear that if we adopt a static, structuralist, >>conception and insist that all relationships are strictly local within >>the ``phrase structure'', then we must also adopt something like the >>SLASH-passing treatment of long-distance dependencies. and, if Bob Carpenter's summary is accurate: >>Johnson argues (or at least did two weeks ago in Barcelona) that HPSG >>is just doing all kinds of program/grammar transformations by hand >>that should be left to the compiler writer. then I think I disagree with Mark's perspective. Consider (let me do my own Ivan Sag impersonation :-)) that we now know of at least the following 9 languages exhibiting grammatical or lexical phenomena that appear only within (or alternate with some other form that cannot appear within) an unbounded syntactic dependency: Irish Complementizer Alternations (McCloskey 1979, 1989) Chamorro Verb Morphology (Chung 1982, 1995) Palauan Verb Morphology (Georgopoulos 1985) Icelandic Expletive Subjects (Maling and Zaenen 1978) Kikuyu Downstep Suppression (Clements 1984) Thompson Salish Verb Agreement (Kroeber 1993, ms.) Moore Morphology (Haik 1990) French Stylistic Inversion (Kayne and Pollock 1978) Spanish Stylistic Inversion (Torrego 1984) Bob Levine or Tom Hukari may be able to add more, but these are enough to make the point that syntactic constructions and lexical elements in many genetically unrelated (pace Greenberg) languages must make reference to the global syntactic context: i.e. whether or not they are in an unbounded dependency construction. Note further that with respect to some of these phenomena it makes a difference whether the extracted element is an NP or an oblique dependent (e.g. Thompson Salish) and in others, it matters which of the lexical item's dependents the dependency travels through (e.g. Chamorro). Given the undisputed fact that lexical subcategorization is local in some sense (the issue is in WHICH sense), it is natural to view the SLASH analysis of UDCs as a *theory* in consequence of which certain global information is made available locally. One might thus even see the phenomena referred to above as directly predicted by the SLASH theory of UDCs. Moreover, these predictions distinguish the SLASH-based theory from a number of published alternatives that must be modified (sometimes in quite ad hoc ways) in light of these data, or perhaps rejected entirely. On Mark's view, I guess we'll have to leave the writing of the lexicon and the treatment of certain grammatical constructions to the compiler writer, too. Let's hope they add syntax courses to the computer science curriculum. Ivan
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:33:03 PST