Volume 54, Issue 2

The number of linguistic analyses conducted on the basis of authentic language corpora has steadily grown during the last decades. The aim of this paper is to investigate the reliability of individual corpus-based studies by comparing them to studies covering similar or identical grounds. As the subject of investigation the present perfect has been chosen. Ten major corpus-based studies of the present perfect will be surveyed focussing on three main aspects: the frequency of present perfect verb phrases, the temporal specification of utterances in the present perfect and the functions assigned to present perfect constructions.

This paper focuses on adjectives complemented by that-clauses as in I am happy that you are well or It is clear that the views of the unions were ignored. A new classification of adjectives will be presented which is based on an empirical analysis of the subject types found in the matrix clause. The classification of adjectives can be corroborated by distributional data. It will be demonstrated that all adjectives followed by that-clauses can be grouped into two classes of adjectives. Each class shows typical co-occurrence patterns which are reflected in the lexico-semantic characteristics of the adjectives. The new classification also covers cases which are referred to as 'subject extraposition'. These cases are included in the classification and will be explained in terms of the same syntactic and semantic criteria which apply to all other adjectives

This paper discusses distributional differences between the two additive particles also and too on the basis of corpus evidence. Three hypotheses are explored: (i) also and too are characteristic of different registers or styles, (ii) the use of also and too is sensitive to structural properties of the 'added constituent', and (iii) the use of the two particles depends on the distribution of 'added material' over the sentence. The discussion centres around the question to what extent corpus-based methods are appropriate to test these hypotheses. While hypothesis (ii) can straightforwardly be tested (and confirmed), the other two hypotheses pose methodological challenges of different types. Corpus-based test procedures are outlined for both hypotheses and preliminary results are given, but it is pointed out that specific types of questions could be answered more easily by applying common elicitation procedures

There is a stark contrast between the random sample model underlying the statistical analysis of corpus frequency data and our intuitive knowledge that sentences are more than random bags of words. The 'library metaphor' illustrates how randomness results from the selection of a corpus as the basis for a linguistic study. At the same time it reveals two reasons why corpus data do not fully meet the assumptions of the random sample model. Finally, practicable methods for identifying and quantifying non-randomness are introduced and demonstrated on the example of passive verb forms.

Over the past few decades, corpus linguistics has evolved into a fullyfledged methodological approach with an increasing number of scholars using various different methods. In this rather programmatic paper, I will argue, however, that corpus linguistics has, in some respects at least, still some way to go in terms of developing rigorous tools and methods and using them more often. More specifically, corpus linguistics - as the young discipline it still is - still has much to learn from other disciplines; a prime candidate in this respect is psycholinguistics. I will try to support this claim with arguments from several case studies.

This paper is based on a nascent project on Lancashire dialect grammar, which aims to describe the relevant features of this dialect and to engage with related theoretical and methodological debates. We show how corpora allow one to arrive at more precise descriptions of the data than was previously possible. But we also draw attention to the need for other methods, in particular modern elicitation tasks and attitude questionnaires developed in perceptual dialectology. Combining these methods promises to provide more insight into both more general theoretical issues and the exact nature of the object of study, namely Lancashire dialect.

This paper explores the relationships between traditional genetic trees of languages and (unrooted) similarity trees. Genetic trees are typically based on lexical and phonological data and ignore information from any other level of analysis. Similarity trees can directly be computed from corpus data. Since a corpus can be seen as corresponding strings (text, annotation tags), string comparison methods from bioinformatics can be used to compute 'distance scores' between languages. Similarity trees based on phonological and syntactic features are presented for five historical varieties of German, and the scope and limits of such surface-based methods in the calculation of language relationships are discussed