Multidimensional (MD) analysis is a methodological approach that identifies co-occurrence patterns of linguistic features based on the factor analysis and characterises a text or a group of texts in terms of those patterns that are functionally interpreted.
Biber (1988), for example, identified a pattern in which a text with the high frequency of private verbs (e.g., anticipate, assume, believe) is also likely to include the high frequency of that deletion (e.g., I think he went to . . . ), contractions (e.g., he’s, can’t), and other features, and the low frequency of, among other features, nouns, prepositions, and attributive adjectives (e.g., the big horse).
The co-occurrence pattern of this kind is called a dimension, and different co-occurrence patterns represent different dimensions of variation of linguistic features. Each dimension is given a functional interpretation on the assumption that the features that behave similarly share their underlying communicative functions. In Biber (1988), the dimension above was labeled Involved vs. Informational Production. If a text has the high frequency of the features with positive loadings (e.g., private verbs, that deletion), the text is likely to have a more involved function, while if a text has the high frequency of the features with negative loadings (e.g., nouns, prepositions), the text is likely to have a higher informational focus.
In MD analysis, each text can be situated along each dimension through dimension score, and by computing the mean dimension score of a group (e.g., genre or register), we can characterise the discourse of the group by the identified dimensions. In Biber (1988), high on the Involved vs Information Production dimension were the registers such as telephone conversations, face-to-face conversations, personal letters, and public conversations, while the registers that are situated towards the lower end included official documents, academic prose, press reportage, and press reviews. You can see that the language use in the latter group of registers is much more information-centered than the language use in the former.
In our project, we will carry out an MD analysis on 11 journals. Our main interests are discoursal differences between journals, particularly between those of monodisciplinary and interdisciplinary ones, and the chronological change of Global Environmental Change, an interdisciplinary journal.
MD analysis has been applied to a variety of domains including specialized areas such as academic discourse. Compared to previous MD studies, however, our target domain is even narrower (i.e., research papers in limited areas). We will see what kind of dimensions will emerge out of our corpus.