The Construction of Local Grammars

Abstract

Grammar, or, as it has now been called, linguistic theory, has always been driven by a quest for complete generalizations, resulting invariably in recent times in the production of abstract symbolism, often semantic, but also algorithmic. This development contrasts with that of the other Natural Sciences such as Biology or Geology, where the main stream of activity was and is still now the search and accumulation of exhaustive data. Why the study of language turned out to be so different is a moot question. One could argue that the study of sentences provides an endless variety of forms and that the observer himself can increase this variety at will within his own production of new forms; that would seem to confirm that an exhaustive approach makes no sense. As a simple computation shows, there are more than 10 50 sentences having at most 20 words, a number which seems to deprive of any meaning the possibility of performing a systematic inquiry. However, the same could be said in Astronomy, Botany or Entomology, since the potential number of observations of individual stars, plants or butterflies is also limitless. Note that nonetheless, establishing catalogs of objects (and devising suitable criteria to do so) remains an important part of the activity in these fields. It is not so in linguistics, even if lexicographers do accumulate and classify words very much as in the other sciences. But grammarians operating at the level of sentences seem to be interested only in elaborating general rules and do so without performing any sort of systematic observation and without a methodical accumulation of sentence forms to be used by further generations of scientists. It is not necessary to stress that such an accumulation in any science is made possible by constructing and using suitable equivalence relations to eliminate 1 I indebted to Morris Salkoff for important suggestions that improved this article. what are deemed to be accidental variations, irrelevant to the specified goal of the catalog. 2

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 131 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.