Measuring syntagmatic Fixedness of Multi-Word Expressions

Syntagmatic fixedness is an important feature of multi-word expressions (MWE). However, syntagmatic fixedness is gradual and various semantic and syntactic relations hold among the parts of MWEs. This poses intriguing problems for lexicography, linguistic description and language processing. In this paper we propose a computationally inexpensive and intuitive approach to the measurement of syntagmatic fixedness based on positional co-occurrence data that is not easily captured by simple statistical significance tests. The types of relations between frequently and systematically co-occurring lexical items have been the subject of studies in phraseology, corpus lexicography and corpus linguistics. MWEs have been classified according to different criteria: the compositionality of their meaning, their syntactic structure (phrasal vs. sentential), their internal structure, their grammatical well-formedness, their communicative function, their metaphoricity. For classifications of MWEs cf. (Moon, 1998; Cowie, 1998; Burger, 1998). Types of MWEs most frequently cited include named entities, idioms, proverbs, similes, routine formulae and sayings. To this we might add conventional metaphors which often involve the co-occurrence of lexical items. In the following we will focus on three important concepts in this discussion, namely cooccurrence, collocation, and idiom. Our emphasis is on co-occurrence between tokens of the underlying corpus but other notions are possible. In general, co-occurrence is seen as the existence of words in structural or positional adjacency or proximity (Evert, 2005). The concept of co-occurrence is discussed in more detail in section 3. Whereas the term co-occurrence is basically used in a rather uniform way, the term collocation has been used in a variety of ways, which can be summarised as follows: