Book Reviews
Corpus Linguistics
Tony McEnery and Andrew Wilson
(Lancaster University)
Edinburgh: Edinburgh University Press
(Edinburgh Textbooks in Empirical
Linguistics, edited by Tony McEnery
and Andrew Wilson), 1996, x+209 pp;
distributed in the U.S. by Columbia
University Press; hardbound, ISBN
0-7486-0808-7, $70.00; paperbound,
ISBN 0-7486-0482-0, $24.50
Language and Computers: A Practical Introduction to the Computer
Analysis of Language
Geoff Barnbrook
(University of Birmingham)
Edinburgh: Edinburgh University Press
(Edinburgh textbooks in empirical
linguistics, edited by Tony McEnery
and Andrew Wilson), 1996, ix+209 pp;
distributed in the U.S. by Columbia
University Press; hardbound, ISBN
0-7486-0848-6, $70.00; paperbound,
ISBN 0-7486-0785-4, $24.50
Reviewed by
John M. Kirk
The Queen's University of Belfast
The appearance of not one but two introductions to corpus linguistics within the same
series shows the maturation and diversification of this fledgling subdiscipline within
linguistics. McEnery and Wilson offer an overview or annotated report on work done
within the computer-corpus research paradigm, including computational linguistics,
whereas Barnbrook offers a guide or manual on the procedures and methodology of
corpus linguistics, particularly with regard to machine-readable texts in English and
to the type of results thereby generated.
Whereas McEnery and Wilson recognize that the distinguishing features of cor-
pus linguistics rest with its computer-aided empiricism, they are eager to line it up
alongside cognitive rationalism in an effort to show the complementarity and interde-
pendence of the two. As they argue, the advantages of a corpus-linguistics approach
are that it is invariably systematic and rigorous, and that linguistics based on a corpus
acts as a yardstick or control to linguistics based on artificial or introspective data.
Of these current research paradigms, the authors' discussion offers fair and balanced
criticism.
In the central core of the book, McEnery and Wilson present overviews of the the-
ory and practice of corpus linguistics, the relative merits of qualitative versus quanti-
333
Computational Linguistics Volume 24, Number 2
tative approaches to language study, and a report (within linguistics) on subject-based
studies using a corpus linguistics approach of which the authors evidently approve.
The three central chapters constitute essential reading for every newcomer to the field.
In Chapter 2, the authors emphasize the key factors in a corpus-linguistics approach:
sampling, representativeness, size, format (and all their many sets of choices). In Chap-
ter 3, by emphasizing the dilemma of interpretation inherent in any claims about a
language as a whole that are based on no more than a sample, they focus on the
benefits of quantitative information that is statistically reliable and from which robust
generalizations may be made.
As graduates of Lancaster, whose mentors and now collaborators have included
Professors Garside and Leech, the authors reflect the preferences of their training: they
favor well-controlled (usually smaller) corpora (random stratified samples rather than
simply random samples) with good extra-textual information stored in headers, which
can be reused, with the same data yielding increasingly complementary results, such
as the Brown and Lancaster-Oslo/Bergen written corpora, or the Lancaster Spoken
English Corpus.
The many criteria expounded in these chapters will almost certainly lead others
embarking on the use of corpora to justify their many strategic decisions. Although
newer investigations might come to upstage some of the topic-centered studies high-
lighted here, the general criteria seem sufficiently assured to be of service for some
time to come--particularly those of verifiability, total accountability, and strength of
argument.
The final two chapters deal quite specifically with computational approaches and
numerous accompanying issues, and it is doubtful whether they will interest many
who are not natural language processors--certainly, there is little for the student of
English who is primarily interested in the analysis and description of data, for w h o m
these chapters are almost certainly beyond reach without further specialized training.
The chapters understandably reflect the authors' experience as programmers with Lan-
caster's pioneering projects: the development of automatic techniques for word-class
tagging and syntactic parsing (as evidenced by the development of the CLAWS tagger
and on the Lancaster Parsed Corpus and the Lancaster-Leeds Treebank), for automatic
machine translation (as evidenced by work on the Canadian Hansard Corpus and on
the Crater Corpus), and for the automatic identification of sublanguages (as in the IBM
manuals study), on all of which they write with conviction and enthusiasm. Whatever
the merits of these various approaches, the discussion lacks a demonstration of the
results of machine translation in action, so that readers might see how the approach
could benefit them. Nor are there any actual samples from the IBM manuals, whose
words and sentences and ratios of words to sentences are counted and presented on
graphs ad nauseam, nor a passage illustrating the finite subset of the language (re-
ferred to as "closure") in action. Indeed, the entire book is short on substantiating
examples, so that it rarely emerges what scholar X's work on subject Y contributes to
Y, however much favored by the authors.
Whereas McEnery and Wilson attempt the broad sweep across this burgeon-
ing field, Barnbrook attempts a much deeper analysis of the exploitation of corpora
(McEnery and Wilson's Chapter 2). There are separate chapters on using a computer
in the first place, issues arising from the choice of data and their capture, the use of
frequency lists, the generation of concordances, the analysis of collocations, and the
question of tagging, parsing, and other kinds of in-text annotation. All of these chap-
ters are very accessible to students of English because they are richly illustrated with
examples. This same is true of Barnbrook's next chapter, which provides an overview
of uses of corpora and their exploitation within natural language processing: stu-
334
Book Reviews
dents of English can see instances that relate to their use of computers (as simply in
word-processing) or examples of what is automatically analyzed or, in the case of the
so-called sublanguage of dictionary definitions, being investigated. Barnbrook's lucid-
ity, his practical guidance in showing the reader what to do and what comes out of
doing it, his copious examples and illustrations, all recommend his book highly. Just
as McEnery and Wilson reflect their Lancaster training, so Barnbrook's emphasis on
linguistic description reflects his association with the COBUILD dictionary and gram-
mar project at Birmingham. My only reservation about Barnbrook is his excessive use
of examples of spelling variants from Chaucer; whereas this use is understandable
in view of Barnbrook's doctoral research, even many native-speaker students are nei-
ther excited nor curious about the language of the past nor about spelling, so that its
use undervalues the approach in its perception of relevance, tending, I have unfortu-
nately found, to discourage. Corpus linguistics is so much more than variants within
late fourteenth-century spelling! With so much contemporary data running through
Birmingham's Bank of English, their use for illustrative or investigative purposes here
could only find approval for the book by a much wider and, I'd guess, more widely
international audience.
Although these books are undoubtedly playing to the same tune, with much com-
mon ground and overlap, their achievement is to reflect the different traditions emerg-
ing in the field by which the authors have become influenced and which ultimately
differ between "doing computing" and "doing language." Despite their differences,
they each show that corpus linguistics has two central planks: the ways in which
the computer is usable for language study and, on this basis, the generation of new
descriptions and understanding.
John M. Kirk lectures in English at The Queen's University of Belfast. He compiled the North-
ern Ireland Transcribed Corpus of Speech and is collaborating on the International Corpus of
English. He has published several articles in Corpus Linguistics and is working on an introduc-
tory textbook. Kirk's address is: School of English, The Queen's University of Belfast, Belfast
BT7 1NN, Northern Ireland; e-mail: [Link]@[Link]
335
