Web-Based Corpus Software

What is a web-based corpus and what is web-based corpus software? The answer is, strictly speaking, that there is no such thing as web-based corpus software. However, one should not be discouraged by this rather negative assessment. In fact, if one examines the title closely, different bracketings of the phrase might suggest interesting possibilities. For example, if one chooses to write it as ‘(web-based corpus) software’, the emphasis falls on the idea of the World Wide Web as a large corpus. It is, however, a very chaotic one. It is chaotic in the sense that it is diffi cult for its users to account for and control the sort of phenomena such a large and dynamic repository might refl ect when explored, say, through an ordinary search engine. This makes the task of formulating and testing hypotheses extremely diffi cult. All sorts of ‘noise’ might creep in: there are texts written by native and non-native writers, computer-generated text (e.g. text resulting from the ubiquitous web-page translation services currently on offer), duplication, and other forms of text which do not conform to standard norms. Little, if anything, can be done to guarantee the quality or integrity of the data being used. Still, this chaotic, noisy environment can be of some use to the statistically minded (computational) linguist. To borrow an example from Manning and Schütze (1999), one could use the web to decide which of the following word sequences to treat as a language unit: ‘strong coffee’ or ‘powerful coffee’. A quick search reveals over 30,000 occurrences of ‘strong coffee’ against just over 400 occurrences of ‘powerful coffee’, thus indicating that the former forms a collocation pattern while the latter apparently does not. In contrast, should one wish to write ‘web-based corpus software’ as ‘webbased (corpus software)’, the emphasis clearly falls on ‘corpus software’, of which web-based corpus software would simply be one type. In other words, one could simply regard the Web as the medium through which better constructed, human-designed corpora can be searched and studied by a large