Evidence against the “Grammatical”/“Ungrammatical” Distinction

Publisher Website
Google Scholar
Abstract
The majority of computational linguists develop systems which analyse NLs using some form of generative grammar which defines a clearcut class of well-formed sentences. But computational linguists who work with corpora of authentic NL material often doubt the validity of any clearcut distinction between grammatical and ungrammatical sequences. Statistics on the distribution of different types of noun phrase in a 40,OOO-word sample of written English are used to show (i) that there is a continuous gradient from very common to very rare constructions, and (ii) that alternative constructions grow more numerous at lower frequency-levels in a regular fashion which implies that a significant proportion of grammatical constituents in a text will belong to extremely rare types.

This publication has 0 references indexed in Scilit: