Bob Carpenter (bc10+@andrew.cmu.edu)
Mon, 11 Sep 1995 11:48:48 -0400 (EDT)
Fernando, as usual, is exactly right. There are some interesting examples of such translations that deal with the semantic side of things, too, in his and Stu Shieber's book on Prolog and Natural Language. There the goal was to eliminate left-recursion (rules like N --> N PP) without changing the semantic results (in order to apply simple backtracking top-down parsers). One example of an empty category that couldn't be eliminated would be one of the form VP/VP that had a semantic content which computed Moens/Steedman style event-type coercions. Recall that the semantic content of such operations, for instance, allow a meaning of a punctual verb like 'to sneeze' to be iterated so that they can be used in examples like 'sneezed for an hour'. The tricky thing is that they can cycle, because a process can be come punctual again, and so on. But if you look at HPSG, the empty categores were eliminable. In general, they can be eliminated in favor of additional rule schemes (like the new work on phrase subtyping that eliminates the empty relativizer) or in favor of lexical rules (like the complement-extraction lexical rule). - Bob Carpenter
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:33:04 PST