No label defined (Q4244418)

The paper describes an implemented system for the automatic acquisition of arguments for (some) common verbs from labelled texts. The acquisition is carried out making use of a valency database containing subcategorization information. The contents of the database is verb lemmas with associated argument frames, plus cues that lead a matching algorithm towards the identification of the verbal arguments. With the help of the matching algorithm and of a simple regular expression grammar, the valency information found in the database can be semi-automatically validated with respect to a corpus, and can be verified, rejected, or updated with empirical results. The linguistic information in the valency database has been (automatically) acquired from a verb valency machine readable dictionary and the Gothenburg lexical database. The argument identification system has been tested on linguistic material taken from the Stockholm-Umeå Corpus and from newspaper articles downloaded from the Internet. The main purpose of the presented work is to maintain a complete knowledge-base of vital importance in a broad range of applications, such as lexically specified grammars (e.g. HPSG theory), parsing and language learning etc. For the entire collection see [Zbl 0912.00026].