Introduction
Problems in corpus analysis
Using small hand-sorted corpora and large computerised corpora each has problems. Findings from a small corpus may not be generalisable because contextual factors can skew the data. On the other hand, most large corpora provide the researcher with only outline information about context. Further, patterns may be missed in searching large corpora, because the researcher searches for particular linguistic forms: if he/ she has not identified a particular form as worthy of study, it may not emerge from the data during the analysis. In this paper we discuss how we have combined an analysis of a small corpus with the examination of concordances from a large corpus in an attempt to tackle the problems of each type of study.

Summary of linguistic findings
We studied metaphors in spoken discourse and their co-text. Our main findings were as follows:

· Metaphors in spoken discourse consistently co-occur with devices often described as ‘hedges’.

· It is common for more than one such device to be used with a metaphor.

· The devices identified serve a range of pragmatic functions. In particular, they offer c-cues to intended interpretations of metaphorically used language.

We have called these devices ‘tuning devices’; as will be argued, this term is intended to better reflect their orientation to the content of the utterance and to the hearer.

Methodology
Small to large corpora
A small corpus was analysed by hand, and words and phrases found to co-occur with linguistic metaphors were studied in a large corpus using a concordancing program. The small corpus consisted of 28,285 words of transcribed talk recorded in seven discourse events in a primary school through a microphone worn by a ten year old girl. Accompanying the transcripts are field notes taken by an observer, and copies of written texts used in the activities. This corpus was read and examined in its entirity a number of times. It was subjected to linguistic and discourse analysis, to investigate the nature of its linguistic metaphors and their pedagogic and interpersonal functions (Cameron, forthcoming). In the second phase of the investigation, linguistic metaphors and their frequent collocates found in the small corpus were studied using a section of the Bank of English, consisting of approximately 9 million words of spoken data. Concordances were examined for literal and metaphorical meaning, and for pragmatic and discoursal function. Where a lexical item occurred more than 1000 times in the corpus, the search was limited to 1000 randomly selected citations.

Specific to general use
With a small corpus collected in a single discourse context, generalisability is a problem. The co-occurrence of tuning devices and metaphors appeared to be of importance, but might have characterised the speech of individual participants rather than being a more general feature of language use around metaphor. By using our small-corpus findings as the starting point for the analysis of a large corpus, we were able to establish whether that was the case. Our findings from the large corpus analysis suggested that many of the tuning devices we found are in general use, with a broad range of pragmatic functions. However, one marker of metaphor found in the small corpus was rare in the larger corpus, which we ascribed to its specialised instructional use.

Meaning and use to form
By beginning our investigation with the study of a small corpus, we were able to focus on meaning and use. Through a detailed examination of the whole corpus, including a number of inter-rater checks, we were able to identify each linguistic item that could be included in our definition of ‘metaphor’. We examined their linguistic context and identified frequent collocates, considering their meaning and use. We then moved on to study these linguistic forms in the larger corpus. This is a reversal of typical corpus analytic procedures, which usually start with the identification of instances of a linguistic form, only later moving to the analysis of its meaning and use.

The first phase of the investigation suggested that nominal metaphors were often hedged (Glucksberg and Keysar 1993) by markers such as like and sort of. It appeared that the discourse function of these devices is to ‘tune’ the metaphor to the listener’s needs. In the classroom context in which the small corpus was gathered, this usually involved a teacher tuning metaphor for students. Tuning took the form of trying to prevent a metaphor from being understood literally, toning down the potential strength of a metaphor, or indicating the nature of the mapping to be made between Topic and Vehicle; for instance, approximation or comparison. Tuning devices were also used to prevent students from interpreting metaphorically a statement that was intended to be taken literally. These tuning devices were also present in the large corpus, where we found that they were used with similar functions. The difference in genre of the two corpora appeared to be responsible for some differences; in the larger corpus it was less common for speakers to explicitly direct listeners to a particular interpretation. On the other hand, it is common for speakers to use tuning devices to negotiate the interpersonal aspects of using figurative language (Cohen 1979, Gibbs and Gerrig 1989). In many cases these devices seemed to have a wider function of signalling an unexpected or pragmatically difficult stretch of text. In both corpora there was no detectable difference in the use of tuning devices depending on whether a metaphor is conventional or innovative. What seems to matter is its expectedness in a particular discourse context rather than degree of conventionalisation per se.

Implications
In the corpus citations studied, speakers used tuning devices to cope with the ever present gap in experience, knowledge and attitude between themselves and their hearers. In a small corpus whose discourse context is known to the researcher, the extent and nature of this gap can be estimated, and used to analyse the role of tuning devices (and, potentially, other markers of interpersonal negotiation of meaning). In examining a larger corpus the researcher generally lacks specific knowledge of context. Combining small and large corpora allows researchers to exploit detailed knowledge of context in analysing meaning and use of interpersonal markers, but also to draw more generalisable findings concerning their linguistic patterns of use.

Cameron, L. forthcoming. Metaphor in Educational Discourse. London: Continuum Press.
Cohen, T. 1979. ‘Metaphor and the cultivation of intimacy’, in Sacks, S. ed. On Metaphor. Chicago: University of Chicago Press.
Gibbs, R. W. and Gerrig, R. J. 1989. ‘How context makes metaphor comprehension seem ‘special’’, Metaphor and Symbolic Activity. 4(3). pp. 145- 158.
Glucksberg, S. And B.Keysar (1993). How metaphors work. In A. Ortony (ed) Metaphor and Thought (2nd edition). Cambridge: Cambridge University Press. 401 –424.