Abstract: Henry James’s style is famous, or perhaps infamous, for its complexity. Numerous critics have commented on James’s ‘difficult’ writing, and also noted that James’s style becomes more complicated in the later novels. But while the ‘difficult late style’ is noted, few critics describe its characteristics in detail, and hardly any look at the syntax of James’s writing.

This seminar presents results from a pilot project which compares a chapter from Washington Square, an early James novel, with one from The Golden Bowl, a late novel, using the methodology of corpus stylistics. An annotated database was compiled from the two chapters, making it possible to search for particular grammatical features and compare how much they are used.

After distinguishing between sentence length and complexity, the standard measure of syntactic complexity, the number of dependent clauses, will be used to compare the two chapters. There will be a particular focus on James’s use of embedded elements, which are quantified in a ‘delay score’ comparable to Ohmann’s non-quantified ‘self-embedding’. The contrast between sentences containing direct speech and non-speech narrative will also be demonstrated.

In conclusion, the findings, based on this small sample, show an increase in complexity from the early to late texts, but also suggest that much of the readers’ impression of complexity may be based on a few extraordinary sentences. Finally the extension of this pilot into a full research project will be described, and and the suggestion made that this methodology might be used for other work on literary style.

Abstract: ‘Parsing’ is the process of applying a linguistic tree analysis to sentences. The problem is: which grammar should linguists use? Not only is there little agreement about grammars among linguists, but there is no agreed methodology for demonstrating the benefits of one grammar over another.

One simple method for comparing frameworks is based on the idea that a ‘better’ grammar is one that is more descriptive. If you can reliably retrieve individual linguistic events from one corpus but cannot from another (i.e. reliably distinguish concepts), then you can say that the first is ‘better’ than the second.

However, this is an essentially circular argument. Who is to say whether one event is more important than another? It is also atomistic. You can just keep adding terms to a grammar to make it ‘better’ . Most of all, it seems to be unsatisfactory because grammar is as much about structure as it is about the individual elements (words, nodes and relationships) that make up that structure.

In this paper we present a new methodology for linguistic interaction experiments developed at the Survey which allows us to do two things:

1. compare the use of a tree grammar over a flat grammar (part-of-speech tags) for capturing the same phenomenon, and
2. demonstrate evidence of interaction over grammatical embedding (phrases within phrases) - a phenomenon only visible with a tree grammar

The method is also a valid experimental perspective in its own right.

We conclude with a discussion of the implications of our methodology for a series of applications, including optimising and evaluating grammars, modeling case interaction, contrasting the grammar of multiple languages and language periods, and investigating the impact of psycholinguistic constraints on language production.

Slides (PowerPoint)
Paper (PDF)
Interaction trend analysis preadsheet (Excel)