76  CHAPTER 2.

IDIOMS
VILNIAUS UNIVERSITETAS

VILNIAUS UNIVERSITETO LEIDYKLA

VILNIUS, 2017
Apsvarstė ir rekomendavo išleisti
Vilniaus universiteto Filologijos fakulteto taryba
(2017 m. kovo 3 d., protokolas Nr. 9)

Recenzentai:
Prof. dr. Inesa Šeškauskienė (Vilniaus universitetas)
Dr. Jolanta Kovalevskaitė (Vytauto Didžiojo universitetas)

ISBN 978-609-459-820-3

© Rita Juknevičienė, 2017

© Vilniaus universitetas, 2017
CONTENTS

Acknowledgements.................................................................................................4
Preface.........................................................................................................................5
Chapter 1. Preliminaries.......................................................................................7
1.1 Phraseological research in linguistics................................................11
1.2 Terminology..................................................................................................13
1.3 Classification of phrases..........................................................................17
Further reading....................................................................................................21
Study tasks.............................................................................................................21
Chapter 2. Idioms..................................................................................................25
2.1. Definitions.....................................................................................................27
2.2 Classification of idioms.............................................................................29
2.3 Designing a study on idioms..................................................................30
Further reading....................................................................................................32
Study tasks.............................................................................................................32

Chapter 3. Collocations.......................................................................................34
3.1 Definitions......................................................................................................37
3.2 Classification of collocations..................................................................40
3.3 Association measures...............................................................................42
3.4 The extended unit of meaning...............................................................47
Further reading....................................................................................................48
Study tasks.............................................................................................................48
Chapter 4. Lexical bundles, n-grams, recurrent sequences.................... 51
4.1 Definitions......................................................................................................51
4.2 Qualitative analyses of lexical bundles..............................................55
4.3 Methodological considerations.............................................................59
Further reading....................................................................................................61
Study tasks.............................................................................................................62
Postscript................................................................................................................65
REFERENCES...........................................................................................................67
I. Corpora...............................................................................................................67
II. Dictionaries......................................................................................................67
III. Software...........................................................................................................68
IV. Literature.........................................................................................................68

CONTENTS  3
ACKNOWLEDGEMENTS

This course book is a result of my work at the Department of English Philology

in Vilnius University. I would like to thank my colleagues for letting me enjoy a
sabbatical term in autumn 2016. Warm thanks are also due to all my students for
being a source of constant inspiration. I hope this course will encourage you to
never stop marvelling at the secrets of language.
A very important milestone in the understanding of research in phraseology was
my study visit at Cardiff University in February 2015. It was made possible through
funding provided by the Research Council of Lithuania. I would like to thank my
academic hostesses Alison Wray and Tess Fitzpatrick for their encouragement and
support.
I was very lucky to have two wonderful reviewers—my grateful thanks go to
Jolanta Kovalevskaitė and Inesa Šeškauskienė for their valuable comments and
advice. Many questions and suggestions from my colleagues at the Department also
helped me make the course book more accessible to the intended readers.
Lastly, I would like to thank my family for being what they are and never allowing
me to lose the sense of reality.

4  E N G L I S H P H R A S E O LO G Y A N D CO R P O R A
PREFACE

The present course book is a summary of lecture notes compiled for

undergraduate students of English Philology at Vilnius University for the course on
Phraseology. What started as an optional course on the study of phrases was gradually
developed into a seminar on research writing in the field of phraseology. The design
of the seminar requires that students gain basic knowledge in phraseological
research, develop skills in the use of a number of corpus tools, acquire elementary
understanding of several statistical tests relevant for the analysis of corpus data
and, last but not least, are introduced to the process of research writing. This course
book was thus primarily conceived as a manual for students enrolled on the course
and interested in carrying out their own phraseology-based research projects for
their term papers. Therefore, it introduces the reader to major research strands in
contemporary English phraseology, discusses its terminology and methods as well
as offers tasks for individual exploration.
Despite its relatively short history, phraseological research is currently under-
going a rapid growth, so it would be impossible to do justice to a plethora of stud-
ies within one academic course lasting for one semester. Therefore the coverage
of literature in this course book is inevitably limited. Two major considerations lie
behind the selection of studies and scholars referenced here. Firstly, an attempt
was made to use for reference the seminal publications which serve as background
reading in the field. Secondly, the choice of research articles for discussion is ex-
pected to cater for students facing their first academic challenges. Many of them
are exemplary works whose research design could be replicated by budding lin-
guists wishing to carry out small-scale individual projects on linguistic data from a
different language or with a different set of phraseological expressions. Hopefully,
the course will provide a reasonable methodological basis and encourage students
to take a step further and start developing their own original ideas for BA and MA
theses. As such, the present course book should be regarded as an introductory
reader rather than the ultimate resource. Throughout the book, the reader will find
not only explanations and illustrations but also references for further reading. More

E N G L I S H P H R A S E O LO G Y A N D CO R P O R A  5
generally, the course book may be beneficial to anyone willing to get hands-on expe-
rience in corpus data extraction and analysis and will thus serve as an introduction
to corpus-based and corpus-driven analysis of phraseological units.
Phraseological research overviewed here represents a number of linguistic
schools and traditions and, inevitably, makes use of diverse terminology. For
the purposes of brevity and clarity and in order to maintain an explicit link to
phraseology as a field of linguistic study, I will be using two equivalent terms:
phrase and phraseological unit, of which the former is perhaps more convenient for
its brevity. Both terms will refer to any type phraseological unit which is broadly
understood as a multi-word sequence that has a (semi-)fixed form and is partly or
fully non-compositional. Specific terms, however, will be introduced and used to
discuss different types of phrases described in individual studies and defined in a
more detailed manner.
Lastly, to make the most of this course book, the reader should register with any
website giving access to a corpus of English. The registration is free yet registered
users may enjoy many more search options than those who do not register. The
two widely known options are the British National Corpus (BNC) which can be
accessed on different websites, one of them hosted by the University of Lancaster
(UK) at http://bncweb.lancs.ac.uk; its current version was developed by Sebastian
Hoffmann and Stefan Evert. The same corpus and, in addition, access to the largest
language corpus ever compiled in the world, i. e. the Corpus of Contemporary
American English (COCA), are provided by Mark Davies from the Brigham Young
University (USA) at http://corpus.byu.edu/coca/. It is these two corpora that have
been used for preparation of this course and that would be relevant for individual
tasks.
The body of the course book is arranged in four chapters. Each chapter starts
with a brief explanation of relevant terminology and definitions, an overview of
the most important methodological issues and/or relevant approaches to the
study of phrases in question, and then invites the reader to embark on a journey of
phraseological discoveries by doing a set of study tasks.
Let it be an inspiring experience!

RJ
January 2017

6  E N G L I S H P H R A S E O LO G Y A N D CO R P O R A
CHAPTER 1.
PRELIMINARIES

The term phraseology is defined in the Oxford English Dictionary (www.oed.

com) as:
1. a. The selection or arrangement of words and phrases in the expression of
ideas; manner or style of expression; the particular language, terminology,
or diction which characterizes a writer, work, subject, language, place, etc.
b. Music. Arrangement or construction of musical phrases.
†2. A collection or handbook of the phrases or idioms of a language; a phra­
sebook.
†3. The use of phrases in speech.
In recent years, however, linguists use phraseology (from Greek phrasis ‘utterance’
+ logos ‘science’) to refer to an interdisciplinary research field which deals with
a broad variety of fixed and semi-fixed expressions. Such expressions function
in language as whole units and are increasingly seen by scholars as independent
constituents of vocabulary. In its broadest sense, the notion of fixed expressions
covers such multi-word units as collocations, formulaic sequences, idioms, lexical
bundles, n-grams and many others which will be jointly referred to in this course
book as phrases or phraseological units. Hence phraseology is understood as the
study of phrases.
Recent developments in corpus linguistics and vocabulary studies have provided
ample evidence of phrasal, or formulaic, nature of naturally produced language.
Apparently, a large part of what we say or write is not made up of discrete words
as a glance at an alphabetically arranged dictionary page might suggest, but rather
consists of multi-word units, or phrases. According to different estimates of the
English vocabulary, most words in speech or written language belong to some type
of phrase. The estimates of the level of formulaicity of language (Table 1.1), however,
are rather different.
Variation of the estimates of formulaicity arises from the choice of items, or types
of phrases considered, methods of calculation, and the type of language examined
(spoken or written). Moreover, most counts exemplified in Table 1.1, except for

C H A P T E R 1. P R E L I M I N A R I E S  7
Table 1.1 Estimates of degree of formulaicity of English
Source Reported evidence
Altenberg Altenberg reports data from the London-Lund Corpus
(1998, pp. 101–102) of spoken English which consists of ca. 500,000 words:
“A rough estimation indicates that over 80 per cent of
the words in the corpus form part of a recurrent word-
combination in one way or another.”
Biber et al. “In conversation, about 30% of the words occur in
(1999, p. 995) a recurrent lexical bundle; if two-word contracted
bundles are also considered, almost 45% of the words in
conversation occur in a recurrent lexical bundle.
In academic prose, about 21% of the words occur in a
recurrent lexical bundle.”
Erman and Warren In their study, Erman and Warren focused on so-called
(2000, p. 37) prefabs defined as combinations of at least two words.
They found that in spoken English prefabs account for
58.6% of the language and in writing for 52.3%. So
on average 55% of English could be considered to be
formulaic.
Juknevičienė Two-word lexical bundles account for 73.4% of all the
(2011, p. 65) words in the LOCNESS corpus which consists of academic
essays written by undergraduate students whose mother
tongue is English. Three-word lexical bundles, however,
cover only 18.2% of the corpus.
Vilkaitė The study is concerned with coverage estimates of
(2016, p. 40) collocations, phrasal verbs, idiomatic phrases and
lexical bundles in four registers (academic prose, fiction,
newspapers, and conversation) of English as they are
represented in the BNC Baby version. The research shows
that taken together all four types of “formulaic categories”
account for 32% of academic prose; 36% of fiction; 24%
of newspaper language; 69% of conversation.

8  CHAPTER 1. PRELIMINARIES
Vilkaitė (2016), are based on the frequencies of one type of phrase (i. e. recurrent
sequence, lexical bundle or prefabricated word combination) and disregard any
other possible multi-word combinations, for example, collocations or phrasal
verbs, which are often non-contiguous and thus much more difficult to capture.
Apparently, the given numbers remain rather crude measures of formulaicity yet
they demonstrate that phrases, whatever the type, account for an impressive part
of naturally produced English.
Another observation from corpus research gives even more prominence to the
status of phraseological units in vocabulary. Evidence abounds that in terms of
frequency certain phrases compete with single words, which only proves that such
phrases have more currency in language than many less frequent words. O’Keeffe
et al. (2007, p. 69) found that individual two-word phrases, which are called chunks
in this book, are more frequent in the CANCODE corpus of spoken English (the size
of the corpus is 5 million words) than single words. For example, you know occurs
28,013 times in CANCODE, and there are only 33 single words whose frequency
is higher in this corpus. Clearly, the phrase is a high-frequency item in this corpus
and would be very important to anyone aspiring to describe spoken English as it
is represented in this corpus. But is it an equally important phrase in some other
varieties of English?
A screenshot of a search output page from COCA (Figure 1.1) contains frequencies
of the phrase you know in American English.

Figure 1.1 Distribution of you know across different sections of COCA.

C H A P T E R 1. P R E L I M I N A R I E S  9
The charts provide convincing evidence that you know is characteristic of spoken
rather than any other register of American English. Its normalised frequencies per
million words (see the fourth column in Figure 1.1) allow us to draw straightforward
comparisons: you know occurs six times more frequently in speech than in fiction1
and 96 times more frequently than in written academic language. It is also interesting
to note that the frequency of that phrase has been increasing in the subcorpora of
COCA representing different periods in time. Yet we would need to delve deeper
into the corpus data and examine contexts and collocates of the phrase in order to
explain trends behind the numbers. Moreover, if we examine the word frequency
list of the COCA corpus, i.e. a list of all the words in the corpus reported in the
order of their absolute frequency (available at http://www.wordfrequency.info),
more unexpected discoveries are to be made. With its total frequency of 367,470
occurrences, the phrase you know ranks between the 108th and 109th most frequent
words in COCA (Table 1.2), which are back (used as adverb) and any (determiner),
respectively.

Table 1.2 A sample of word frequency rankings in COCA

Rank Word Part of speech Frequency
104 one pronoun 369,553
105 very adverb 391,821
106 her pronoun 397,950
106 even adverb 361,067
108 back adverb 367,844
you know phrase 367,470
109 any determiner 348,100
110 good adjective 353,973
111 woman noun 341,422
112 through preposition 340,921
113 us pronoun 351,088

So you know is more frequent in this corpus, which undoubtedly remains the most
representative corpus of American English up-to-date, than such functional words
as through, down, after, modal verbs may, should, need, not to mention many simple
open-class words, for example, good, work (verb), call, try, ask etc., all of which
appear much lower in the word frequency list.
1 To calculate how many times one normalised frequency differs from the other, we divide one by
the other. For example, spoken vs. fiction (Figure 1): 2,711/445 = 6.1 times; spoken vs. academic:
2,711/28.10 = 96.5 times.

10  CHAPTER 1. PRELIMINARIES
 Furthermore, the frequencies of the constituent words of the phrase provide
more food for thought: the pronoun you is the fourteenth frequent word in COCA
(3,081,151 occurrences), and the verb lemma know takes the 47th position (892,535
occurrences). If we consider their individual frequencies, it is obvious that the
pronoun often occurs in contexts other than those with the verb know, but over
41% of occurrences of the verb form know are in the phrase you know. Undoubtedly,
the phrase you know must be very important in the overall use of the verb know. To
compare, I know returns only 100,298 matches in COCA, which implies that the verb
has more uses in English in combination with the second rather than first person
pronoun. Admittedly, superficial frequency data does not offer any explanations
why the distribution is what it is but it serves as a good starting point for further
analyses. So even if the frequency data does not yet push the word as the central
unit of language from its pedestal, it certainly provides a legitimate ground for
treating certain phrases as equals to words.

1.1 Phraseological research in linguistics

Frequency data of individual phrases has several implications. To start with,

it justifies the decision of contemporary lexicographers to give the status of
headwords to certain multi-word units rather than merely enter them as illustrative
examples in the entry of the lexical headword. In other words, frequency raises
the status of certain phrases to that of single words. One of the first attempts to
do it was undertaken by a team of linguists lead by John Sinclair in the Collins
COBUILD project which, among many other contributions to linguistics, started
a new family of English dictionaries based on an electronic corpus of English. In
the Collins COBUILD English Dictionary for Advanced Learners (2001, 3rd edition or
any later edition), many phrases are entered as headwords, for instance, of course
(adverb), all clear (noun), not to mention phrasal verbs and compound nouns. For
many such fixed phrases, the compilers of the COBUILD dictionaries used specific
labels to mark both their morphological properties and phrasal nature, for example,
phrasal verb (look after, give in), phrasal co-ordinating conjunction (or else),
phrasal subordinating conjunction (just because, as if), phrasal modal (going
to, is supposed to), phrasal preposition (due to, as well as). Compilers of this
dictionary also identified a group of phrases defined as ‘groups of words which are
used together with little variation and which have a meaning of their own’ (Collins
COBUILD, p. xxix), for instance, you know, and the like, keep going etc.

C H A P T E R 1. P R E L I M I N A R I E S  11
 Apart from lexicography, the discovery of word co-occurrences has numerous
implications for the development of automated translation tools. We might
hypothesise that in the eyes of laymen machine translation is one of the most obvious
applications of linguistic research. Its importance can hardly be overestimated
as the society at large more often than not questions the aims and goals of
linguistic research. The development of automated translation tools convincingly
demonstrates how insights from corpus-driven and corpus-based phraseology
may be put to very practical uses. To obtain lexically and grammatically acceptable
translation, the artificial computer-based brain needs to be taught (or rather
programmed appropriately) which word sequences should be processed and
translated word by word and which should be processed holistically as multi-word
units. Clearly, only after linguists have provided a comprehensive list of phrases, this
information can be used by developers of translation tools. A simple experiment
on Google Translator shows what happens when the existence of phrases is not
recognized by the translation tool.
The collocation do research is translated into Lithuanian by Google Translator
as daryti mokslinius tyrimus (verbatim: ‘do scientific research’), which suggests
that the automated translating tool recognizes the phrase and gives an appropriate
Lithuanian compound noun for the English research. An attempt to translate
to make noise, in contrast, gives a less successful rendering. It is translated into
Lithuanian as triukšmo (verbatim: noise-Gen-Sg), so the program fails to recognize
that it is a phrase and leaves out the verb altogether. Similarly, to cast a ballot ‘to
vote in an ellection’ is translated as mesti burtus ‘to toss up’; the idiomatic sequence
raining cats and dogs is translated literally as lyja katėmis ir šunimis rather than
by the equivalent Lithuanian idiom pila kaip iš kibiro (verbatim: ‘pouring like out
of a bucket’). Further experimenting with Google Translator shows that pila kaip
iš kibiro is rendered back into Lithuanian as as the rain pours, so the Lithuanian
idiom seems to be recognized yet its English equivalent is not provided. Clearly, this
tool remains limited when it comes to the translation of phrases yet its accuracy is
improving. A handful of examples cannot lead to any generalizations, but it could be
hypothesized that a comprehensive inventory of phrases, on condition that it is at
all possible, could certainly increase the level of translation equivalence above the
single word level.
Another vast area in which insights from phraseological studies of vocabulary
have many applications is second language acquisition (SLA) and teaching. It
has long been recognized that the lexical approach (Lewis, 1993; 2000) is more

12  CHAPTER 1. PRELIMINARIES
advantageous than instruction in grammar (morphology and syntax) separated
from the teaching/learning of vocabulary. The practice of teaching word lists has
long given way to phrase lists because memorizing individual words does not
teach the learner anything about combinability of words. Expansion of research
into phraseologies of non-native English was further made possible due to the
development of learner corpora which represent English (or any other language
acquired as foreign/second) produced by non-native speakers, or, in the case
of English, EFL learners. It will suffice for now to refer to a study by Pawley and
Syder (1983), which is considered by many scholars to be one of the first studies
of learner English in terms of lexical phrases. Pawley and Syder raised what today
might appear a simple question: why does it happen that grammatically correct L2
English still sounds unnatural? The answer, as the reader should be able to guess
by now, is fairly straightforward—apart from grammatical choices, there are also
lexical ones to be made. L2 learners are often unaware of the fact that, for example,
a fixed expression in one language (e. g. Swedish tusen tack) does not have a word-
for-word equivalent in English, even if its verbatim rendering would most probably
make sense to any English-speaking person (verbatim from Swedish: thousand
thanks). As a consequence, foreign learners might produce grammatically correct
sentences yet they fail to make the right lexical choices, which results in unnaturally
sounding language.
It is thus little surprise that numerous insights from corpus studies gave rise
to a new understanding of phraseology, a branch of linguistics which deals with
all kinds of multi-word units, irrespectively of their contiguity, formulaicity or
semantic compositionality. Phraseology in contemporary linguistics covers a broad
range of phrases while the term itself refers to either (1) a field of study dealing
with all types of more or less fixed multi-word expressions or (2) the use of these
expressions in natural language.

1.2 Terminology

Phraseology has ‘fuzzy borders’ (Granger and Paquot, 2008, p. 29) and is thus
often regarded as an interdisciplinary field because its research questions and
approaches draw upon principles and categories from many other branches of
linguistics. While the most obvious links point to the relationship of phraseology
with semantics, morphology, syntax and discourse (for a detailed discussion, see
Granger and Paquot, 2008), it also has areas of overlap with psycholinguistics,

C H A P T E R 1. P R E L I M I N A R I E S  13
phonology and many other fields. As a consequence, phraseological research may
resort to terminology originating in diverse linguistic disciplines. Therefore while
it is easy to agree that phraseology is a science about phrases, what a phrase is
remains a disputable issue—every scholar feels it right to come up with a definition
that is most suitable to his/her research design and tradition. To an uninitiated
layman, a phrase is first of all a fixed sequence of several words. A linguist, however,
may argue that certain sequences of words re-occur in our language in a more or
less similar shape whereas others are not fixed at all and can be freely modified by
inserting other words or change their grammatical form. Here are several examples
from the BNC:

(1) I’m a fairly tidy sort of person, so I do make the bed, and sort of tidy up in the
bedroom (…)
(2) As Beck recalled, ‘He came, he slept, he left without making the bed and I never
saw him.’
(3) Clothes lay on the floor and the bed hadn’t been made.

The examples illustrate different grammatical transformations that the phrase to

make the bed ‘neatly arrange the sheets and covers of a bed’ (Collins COBUILD) may
undergo in naturally used language. In this respect, it is different from, for instance,
be rolling in it ‘be very rich’ (Collins COBUILD), where the Continuous form is the
predominant one:

(4) With five top 40 hits under their belt and a top ten album, you might think
Take That are rolling in it. (BNC)2

What is important to note at this point is the fact that changes in the form of a
particular phrase might imply changes in its meaning. Hence, even though the
sequence rolled in it is perfectly possible in English, it usually has a different
meaning from the one illustrated in example (4). Although searches in the BNC give
no matches for rolls in it and rolled in it, but such sequences with non-idiomatic
meanings are attested in COCA, for example:

2 Interestingly, a search in COCA gave one instance where rolling seems to be used in the Present
Simple to express the meaning of ‘being rich’. The excerpt comes from a novel published in 2003:
I can’t save everybody, I’m asthmatic, and I don’t want a dog, especially not one who acts like he snorts
coke and looks like he rolls in it. It is an example of change in language, in this case, an idiom chan­
ging its allegedly fixed form.

14  CHAPTER 1. PRELIMINARIES
 (5) It was full of Gold 100-there must have been seventy rolls in it altogether, and
she loaded one into the camera (…).
(6) The sheets were stripped, but I had torn down the canopy and rolled in it until
I was trussed and bound.

What we observe here suggests that certain phrases have a fixed form, and changes
of their form might incur loss of the phraseological meaning. Others, in contrast,
are flexible and their grammatical transformations do not cause any change in
the meaning. So what is a phrase and how fixed it should be to be recognized as a
phrase?
One of the most frequently cited definitions of phraseological unit was proposed
by Alison Wray (2000). In 2002, she published a book titled ‘Formulaic Language
and the Lexicon’ which was concerned with exploration of what was often termed
before Wray’s publications as lexical phrases (Nattinger and DeCarrico, 1992). Wray
proposed new terminology, namely, formulaic sequences and formulaicity, which was
intentionally rather broad in its scope. According to Wray, a formulaic sequence is
a sequence, continuous or discontinuous, of words or other elements, which is, or
appears to be, prefabricated: that is, stored and retrieved whole from memory
at the time of use, rather than being subject to generation or analysis by the lan­
guage grammar. (Wray, 2000, p. 465; 2002, p. 9)
The definition encompasses four major criteria of the phrase and each of them, al-
beit in varying degree, provides a basis for identification of (semi-)fixed multi-word
unit, be it a prototypical idiom (to pull a person’s leg ‘to tease a person’3 ), a phrasal
verb (to give up ‘to resign, surrender’), or a collocation (stale bread ‘bread that has
lost its freshness’). Let us take a closer look at the four criteria:
1. The statement “a sequence (…) of words or other elements” implies that a
formulaic sequence consists of more than one word. The “other elements”
may refer to non-verbal sounds (erm, mhm) characteristic of spoken lan-
guage as in well erm I dunno.
2. A formulaic sequence has a fixed form. We have already seen that the form
may sometimes vary, but variability is limited for some phrases. For exam-
ple, raining cats and dogs would sound unnatural if it were said as *raining
dogs and cats or *pouring kittens and puppies. Such deviations from the es-
tablished form would lead to a specific stylistic—mostly, humorous—effect
3 If not stated otherwise, all definitions are taken from the Oxford English Dictionary Online
(www.oed.com).

C H A P T E R 1. P R E L I M I N A R I E S  15
 and would be done on purpose in order to evoke unexpected images and
implications.
3. A formulaic sequence is retrieved from memory at the moment of produc-
tion as a whole and not processed by combining individual words. This cri-
terion deals with the idiom principle described by Sinclair (1991) and it will
discussed in more detail in Chapter 3 (page 35).
4. A formulaic sequence has a distinct meaning in discourse. It can be fully or
partially non-compositional, e.g. take a look ‘to look’ vs. kick the bucket ‘(in
slang) to die’. In the latter case the meaning of the sequence cannot be un-
derstood from the individual meanings of its constituents.
The four criteria albeit in varying degrees form the basis for the definition of
any phraseological unit. The definition of formulaic sequence proposed by Wray is
very broad but, according to the linguist, it was intended as such in order to cover a
multitude of terms that had been used by other scholars.
Terminological variation in phraseological studies is indeed impressive. Consid-
er the following collection of terms reprinted from Wray (2000, p. 465):
Amalgams, automatic chunks, clichés, co-ordinate constructions, collocations,
complex lexemes, composites, conventionalized forms, fixed expressions, idioms,
formulaic language, formulae, fossilized forms, frozen metaphors, frozen phra­
ses, gambits, gestalt, holistic phrases, lexical simplex, lexical phrases, lexicalized
sentence stems, multiword units/items, multiword lexical phenomena, non­com­
positional, noncomputational, nonproductive, nonpropositional, petri­fi­cations,
phrasemes, preassembled speech, precoded conventional routines, prefabs, pre­
fabricated routines and patterns, ready-made expressions, ready-made utteranc­
es, recurring utterances, routine formulae, schemata, semipreconstructed phrases
that constitute single choices, sentence builders, set phrases, stable and familiar
expressions with specialized subsenses, stereotyped phrases, unanalyzed chunks
of speech, units
A brief overview of more recent literature on phraseology proves that even this list
is not exhaustive. It could be further supplemented with such terms as, for instance,
catchphrases, lexical bundles, phrasal verbs, proverbs, sayings, and most probably
many others.
Whatever the term, the scope of phraseology covers such phrases which meet
the four criteria discussed above. Scholars pursuing specific research designs may
apply additional criteria for phrases they choose to analyse. It could be the minimal
frequency in a corpus, dispersion across a specific number of texts that are part of

16  CHAPTER 1. PRELIMINARIES
a corpus, a predetermined syntactic construction (e.g. ‘verb + noun’) and others. So
any comparison of findings should be preceded by a cautious consideration of how
the operational definition of item in question has been formulated by the research-
er. As we know, every scholar is entitled to modify his/her approach and definitions
in order to give a better account of the research, which, luckily or unfortunately,
results in this terminological jungle.

1.3 Classification of phrases

Current literature describes two distinct approaches to identification of phrases,

namely, phraseological and frequency-based (Nesselhauf, 2005, p. 12; Granger and
Paquot, 2008, p. 28). The two approaches provide two bases for classification of
phrases which will be overviewed in this section.
The development of the phraseological approach goes back to the middle
of the 20th century and is inseparable from the works of the Russian linguists
Vinogradov and Amosova (Granger and Paquot, 2008, p. 28), who were concerned
with identification of phraseological units in Russian and English. The underlying
principle in the phraseological approach is based on the degree of idiomaticity that
a phrase possesses, or its (non)-compositionality. Phrases with varying idiomaticity
are placed on a continuum from free combinations, whose meanings can be
understood from individual meanings of their constituents, to idioms which are
non-compositional and cannot be understood literally.
One of the most often cited sources for such classification is Cowie (1981),
who described the continuum of idiomaticity as a cline from free combinations to
pure idioms. Two middle categories, namely, restricted collocations and figurative
idioms, are partly idiomatic. Table 1.3 shows how Cowie’s original classification was
further elaborated by Howarth (1998), who divided phrases (termed composites)
into lexical and grammatical. Lexical and grammatical composites differ in their
structure as the former consist of two open-class words and the latter include a
closed-class word (preposition). The degree of semantic compositionality across
the types of phrases gradually increases from left to right.
Free combinations are fully compositional and pure idioms are absolutely
opaque, or non-compositional; restricted collocations have one constituent that is
used “in a specialized, often figurative sense only found in the context of a limited
number of collocates” (Howarth, 1998, p. 28), for instance, make a suggestion where
the verb cannot be substituted by the synonymous do. To judge from later interpre-
tations of this classification (cf. Nesselhauf, 2005, pp. 14–15), the phrase blow a

C H A P T E R 1. P R E L I M I N A R I E S  17
Table 1.3 A continuum of idiomaticity (reprinted from Howarth, 1998)
Free Restricted Figurative Pure idioms
combinations collocations idioms
Lexical blow a trumpet blow a fuse blow your blow the gaff
composites own trumpet
(verb + noun)
Grammatical under the table under attack under the under the
composites microscope weather
(preposition +
noun)

fuse, as it happens, is perhaps not the best example as it has two readings in con-
temporary English: literal and idiomatic and thus perfectly matches the definition
of figurative idioms, understood by Howarth as phrases that “have metaphorical
meanings in terms of the whole and have a current literal interpretation” (Howarth,
1998, p. 28). For example, under the microscope can be understood as ‘placed on the
focal plane of the microscope to see an enlarged image’ and figuratively as ‘being
studied very closely usually because it is believed that something is wrong with it’
(Collins COBUILD). The distinction of the two types in the middle (restricted col-
locations and figurative idioms), in particular, is far from being simple. Nesselhauf
(2005, pp. 14–18) gives the most exhaustive account of varying interpretations of
two major criteria of collocations (opacity and commutability of constituents) and
resulting diversity of operational definitions that could be found in linguistic litera-
ture. Some of these issues will be revisited later in this course book when discuss-
ing collocations.
The frequency-based approach to the identification and classification of phrases
goes back to lexical studies of languages, English among them, which became pos-
sible after the emergence of corpora. In contrast to the phraseological approach
whose practical application largely rested upon the intuitive judgment of the re-
searcher who was viewed as the ultimate authority in the identification of phrases,
corpus search tools provide an objective and inherently different procedure. It is
a computer software rather than the human brain which computes and processes
frequencies of co-occurrences of words and determines which word combinations
‘deserve’ to be included in the program output because their frequency proves that
they are not random co-occurrences but statistically significant linguistic events.
Admittedly, or rather luckily, the human researcher cannot be totally eliminated

18  CHAPTER 1. PRELIMINARIES
from the process because the computer needs to be given specific search param-
eters. More importantly, only the human analyst can interpret automatically re-
trieved corpus data.
There are two major types of phrases that can be extracted from a corpus by
computerized search tools: (1) pairs of co-occurring words understood as “dis-
continuous combinations of two words”; (2) recurrent continuous sequences of
n words known as n-grams (Granger and Paquot, 2008). A detailed discussion of
these types of phrases will be provided in chapters on collocations and lexical bun-
dles (Chapters 3 and 4) whereas it will suffice to note at this point that identifica-
tion of frequency-based phrases is fully automated. The computer software does
not evaluate any semantic relations between constituents of phrases but merely
computes their frequencies or probability of co-occurrence and on this basis identi-
fies a sequence of two or more words as something that has currency in language.
An inquisitive mind might wonder to what extent the phraseological and fre-
quency-based approaches correlate and how much overlap is to be expected be-
tween the two. As it turns out, the frequency-based analysis offers many valuable
insights into what has been pre-determined subjectively. Firstly, corpus research
shows that what is conspicuous and visible to the human analyst, e.g. stylistically
marked non-compositional idioms, expressions of folk wisdom, proverbs etc., are
rather infrequent phenomena in natural language. Phrases which contain partial
compositionality and are not necessarily viewed as fixed phrases, for example, re-
stricted collocations (cf. Table 1.3), were found to be very frequent and very numer-
ous. Moreover, research in applied linguistics showed that it is those ‘fuzzy’ types
of phrases that often differentiate beginners and proficient language users as they
cause many difficulties to EFL learners.
According to Granger and Paquot (2008), the two approaches to phraseological
units cannot be easily reconciled; hence, they propose to make “a clear distinction
between two typologies: one for automated extraction and one for linguistic analy-
sis” (ibid., p. 41). As a result, the scholars suggest that researchers working with
automated extraction of phrases could stick to terminology pertaining to comput-
erized research tools and employ such terms as lexical bundles, n-grams, clusters
to refer to continuous sequences of words meeting a certain frequency threshold
and co-occurrences or collocations to name combinations of words identified on the
basis of specific statistical measures, for example, mutual information or t-score
(see Section 3.3 on association measures, p. 42). Terms used in linguistic analyses,
however, where the focus is not on the extraction method but rather on functions
and uses of phrases in discourse, could employ terms from a functional classifica-

C H A P T E R 1. P R E L I M I N A R I E S  19
tion (Figure 1.2). A detailed description of each subtype in this classification given
by Granger and Paquot looks very comprehensive. Yet it remains to see whether
linguists will adhere to this classification and systematically use the proposed ter-
minology.

Phrasemes

Referential function Textual function Communicative function

Referential phrasemes Textual phrasemes Communicative phrasemes

(Lexical) collocations Complex prepositions Speech act formulae

Idioms Complex conjunctions Attitudinal formulae
Irreversible bi- and Linking adverbials (including attitudinal sentence
trinomials Textual sentence stems stems)
Similes Proverbs and proverb
Compounds fragments
Phrasal verbs Commonplaces
Grammatical Slogans
collocations Idiomatic sentences
Quotations

Figure 1.2 The phraseological spectrum

(reprinted from Granger and Paquot, 2008, p. 42)

As it happens, researchers do not always follow their colleagues. It may sometimes

be explained by a specific design of individual research projects, traditions of
linguistic schools in different institutions and many other circumstances. It is
important to note, however, that whatever approach to the classification is chosen,
be it phraseological, frequency-based or functional, it is always necessary to provide
accurate operational definitions of phrases under study and in this way minimize
any possible terminological confusion.
Finally, a very broad distinction of phraseological units could be related to two
approaches to corpus analysis that appear in the title of this course book, namely,
corpus-based and corpus-driven (see Tognini-Bonelli (2001) for a detailed discus-
sion). When a corpus is merely used to exemplify uses of phrases or check their
morphological and syntactic transformations, the researcher is most probably car-
rying out a corpus-based study and has a list of phraseological units chosen for
the analysis. The corpus-driven approach, in contrast, begins from very different

20  CHAPTER 1. PRELIMINARIES
premises: the researcher does not have a primary data set and therefore resorts
to a corpus to find out what the linguistic reality is. Moreover, the absence of ini-
tial hypotheses means that the researcher is not certain what a corpus might yield
and thus the whole research is driven by whatever comes out of the corpus data.
So while phrases falling under the phraseological definitions are more closely re-
lated to the corpus-based approaches (e.g idioms), frequency-based identification
of phrases is usually part of a corpus-driven research project.

Further reading

 Phraseology as research field; challenges and scope: Ellis (2008); Gries (2008).
 The word vs. lexical item vs. phraseological unit: chapter “The lexical item” in
Sinclair (2004).
 Corpus-driven phraseology: Stubbs (2007); Biber (2009).

Study tasks

1. Consider methodologies used to calculate the degree of formulaicity in studies

referred to in Table 1.1. What are their advantages and disadvantages? What
exactly was being counted in each study?
2. An interesting comment on estimates of formulaicity was made by Wray (2000,
p. 485), who questioned their validity on the grounds that corpora as such rep-
resent a community of speakers. In her view, corpus-based frequency counts
are insensitive to idiolects and “agreed preferences of a speech community”,
both of which might inflate or, on the contrary, minimize the proportions of
what she calls “formulaic material” (ibid., p. 466). Bearing in mind principles
of corpus-based or corpus-driven research, how would you respond to Wray‘s
remark? In your opinion, which extra-linguistic factors (e. g. age, gender, edu-
cation, or occupation) may have an impact on the degree of formulaicity of a
discourse community?
3. Analyse the dispersion of the phrase you know in different subcorpora of the
BNC. Then compare co-occurrence frequency of the lemmas you and know in
British English with the COCA data given on page 11. Write up your observa-
tions in a coherent paragraph (ca. 200 words).
4. Compare several dictionaries of English for foreign learners, e.g. Oxford Advan-
ced Learners’ Dictionary, Cambridge English dictionary, Longman Dictionary of

C H A P T E R 1. P R E L I M I N A R I E S  21
 Contemporary English, for their approach to phraseology. How do they differ
from the practice applied in the Collins COBUILD English Dictionary (see p. 11)?
Do they enter any multi-word expressions as headwords in the body of the dic-
tionaries, or are phrases always given as run-on entries? Write a paragraph (ca.
150-200 words) describing lexicographers’ strategies for inclusion of phrases
in a dictionary of your choice.
5. Reflect on your experience of learning English. Did you start with a list of indi-
vidual words or rather tried to memorise complete phrases? In your opinion,
how should young learners (aged 6–9 years) be taught a foreign language—by
asking them to memorise individual words or rather providing them with lists
of phrases? Explain your answer.
6. Apart from SLA, lexicography and translation, phraseological expressions are
also being investigated in psycholinguistics, neurolinguistics, sociolinguistics
and many other branches of linguistics. Find a recent research paper concerned
with the analysis of any phraseological expressions and write its summary (ca.
300 words). Focus on research aims and the type of phraseological expression
under study.
7. Marcinkevičienė (2010, p. 159) found that collocational pairs of words ac-
counted for 68.1 per cent of the Corpus of Contemporary Lithuanian (the size of
the corpus is 100 million words). How do you think this data from a corpus of
Lithuanian, which is a morphologically inflected language, might correlate with
estimates of formulaicity of English?
8. Phrases are ubiquitous. To do a simple estimate of the degree of formulaicity,
read the following message from the Director of the Centre for Languages, Lin-
guistics and Area Studies in the UK (https://www.llas.ac.uk//about.html). Can
you identify which words are constituents of any type of multi-word unit? Try
to count their proportion to all the running words in the text (the total num-
ber of words is 297). Compare your findings with evidence from other sources
given in Chapter 1.

Dear colleague,
As we sail into the lee of Christmas, most of us are looking forward to a
break and some time out. It continues to be a difficult time for languages
with a great deal of uncertainty about the future and a lot of pressures
on teaching, research, management/admin, outreach, enterprise and

22  CHAPTER 1. PRELIMINARIES
 all of the other activities we undertake. But the uncertainty has also
brought opportunities, even if grasping them takes time and energy to
think up new ideas, make business cases and submit bids.
In fact this term has shone a good deal of public attention on languages,
with continued media coverage of major projects, like Language Rich
Europe, and a concerted week of language-related events organized by
the British Academy. There have also been significant government con­
sultations on languages in schools and major implications flowing from
proposed changes in school education and teacher training more broad­
ly. There can be no doubting the increased concern nationally, and this
is also echoed across Europe where many countries are being shaken in
their complacency and beginning to worry about their capacity in lan­
guages.
One area of languages that continues to hold out hope for innovation
is online education. This is an area in which languages have tradition­
ally been very active. Technological changes are constantly challenging
old approaches and offering new possibilities. LLAS at Southampton is
about to host its 8th annual e-learning symposium on 24-25 January and
already it is clear that the event will be bigger than ever this year. We
have received twice as many session proposals as in previous years, and
demand for places is correspondingly high. Readers thinking of attend­
ing should note that the early bird rate is closing on 20 Dec. It may be a
reflection of where language learning is going.
Mike Kelly (follow Mike on Twitter at @ProfMikeKelly)
Director, LLAS Centre for languages, linguistics and area studies

9. Study the distribution and variability of phrases given below. Run different
searches on the BNC and COCA and establish which grammatical forms of
lexical verb(s) are typically used in these phrases. Apart from morphological
changes, is it possible to establish a specific syntactic pattern in which these
phrases usually occur?
to keep/get/stay in touch
to bring to light
to feel under the weather
to ring a bell
to get the sack
to spill the beans

C H A P T E R 1. P R E L I M I N A R I E S  23
10. Check the translation of different phrases on Google Translator. Try different
language pairs. Do you think it is possible to establish a tendency that would
account for inaccuracies in translation?

11. Consider phrases given below and choose a matching term for each of them
from the list given in this section (page 15). When in doubt, check the internet
for clarification of individual terms in the list.

To whom it may concern

To have influence
To put up with
On the other hand
Bureaucratic red tape could be reduced.
The black and white approach turned out unfruitful.
He knew the poem by heart.
I know I know, but I don’t quite agree with you here
He delivered the speech but he didn’t tell the truth.
Rolling stones are back.
Bond. James Bond.
Have you done your homework?
What about that guy over there?

12. Misuses of phraseological units may be deliberate. ‘Breaking’ a fixed expression

is one of the ways to create humour. Listen to a stand-up comedian’s perfor-
mance and find several examples of purposefully misused phrases.
13. Explain what the joke is based upon.

An instructor is talking to a group of tourists about safety measures. At

the end, he asks: “So what steps do you take if you see a grizzly bear
approaching?”
One of the tourists says: “Big ones. In the opposite direction”.

24  CHAPTER 1. PRELIMINARIES
CHAPTER 2.
IDIOMS

Although idioms are hardly ever associated with the development of corpus-
based or corpus-driven approaches to lexicon, they have been chosen as a point of
departure for the discussion of individual types of phraseological units. There are
several reasons for such a decision. As the reader might have understood by now,
idiom is often the first term that is evoked by any reference to phraseology. Appar-
ently, to both linguists and laymen, the fact that phraseology has already gone a
long way from the study of figurative expressions is not necessarily known. So it is
reasonable to start an overview of phrases from idioms which by many are seen as
the prototypical unit of phraseology.
The other argument for the inclusion of idioms in this course book is the amount
of ongoing research into idiomatic treasures of languages, including diachronic
and synchronic, contrastive and corpus-based studies, undertaken to inform the
practice of translation, SLA, stylistic analysis and many others. Although a simple
Google search is not the most reliable way of checking the currency of one or an-
other notion, it does offer a rather crude yet convincing picture of what is happen-
ing in the virtual world and, obviously, beyond. The frequency of different phra-
seological terms on the internet (Table 2.1) is noteworthy. While the dominance
of idiom could be explained by the long history of this term in language study, it is
also remarkable that the term n-gram(s) is clearly having many more references
in online resources than collocation(s), which at least chronologically could have
been expected to rank higher in terms of internet hits than the automatically ex-
tracted multi-word sequences. Obviously, idiom is the most frequent of all phrases
occurring in internet sources, even if the word itself may certainly be used in other
senses than the one related to phrases. Yet its currency most probably indicates
that it continues to interest linguists.
In natural language, at least as it is represented in corpora, idioms are relatively
infrequent. The only register where they have been found to occur in higher fre-
quencies is fiction but even here the number is less than five per million words

C H A P T E R 2. I D I O M S  25
Table 2.1 Results of Google search for different types of phraseological terms (searched
on 18 January 2017)
Search string Number of hits
idiom 21,100,000
idioms 25,400,000
n-gram 13,200,000
n-grams 5,190,000
collocation 5,290,000
collocations 6,420,000
“lexical+bundle” 16,700
“lexical+bundles” 29,200
“phrase+frame” 9,430
“phrase+frames” 16,700

(Biber et al., 1999, p. 1025). Another estimate is even less impressive—Moon

(1998, in Vilkaitė, 2016, p. 34) argues that the frequency of idioms is less than one
occurrence per million words. Yet idioms are quite conspicuous phrases in a text.
The reason is that idioms are unexpected and often illogical word combinations.
For example, the classic example it’s raining cats and dogs ‘raining heavily’ certainly
catches the eye because interpreted literally—and this is how it most probably is
processed when encountered in the text for the first time—it brings to mind an un-
realistic picture. To people who are less familiar with the history of British civiliza-
tion, deriving the motivation behind this idiom is perhaps impossible. Its origin, as
it turns out, goes back to English of the seventeenth century when
(…) there weren’t any drains to take away heavy rain, so many pets and stray ani­
mals drowned during heavy downpours. So it was not unusual to see their bodies
floating down the streets (Watcyn-Jones 2002, p. 61).
So the meaning of this idiom is not instantly obvious because it cannot be under-
stood from the individual meanings of constituent words. In linguistic termino­logy,
idioms are said to be non-compositional, which means the meaning of an idiom
is not composed as a sum of individual meanings of the constituents. Instead, the
whole idiom has its own distinct meaning which is often impossible to explain by
logical inference. Since idioms are striking and unusual, they often evoke a num-
ber of unexpected images and contribute specific emotive undertones to language.
To put it differently, jumps and twists in the meanings of constituent words add
colours to language.

26  CHAPTER 2. IDIOMS
 It is thus natural that early linguists’ interest in idioms was mostly related to
stylistics and eloquence of speech. Hence, such research often involved analyses of
literary texts. The tendency continues to persist in contemporary linguistics. For in-
stance, the stylistic potential of idiomaticity is at the focus of Gläser (1998) and Na-
ciscione (2010). Increasingly, however, studies of idiomaticity establish links with
other realms of language study. Many valuable insights are derived from cognitive
linguistics which, among other things, deals with metaphors. Cognitive linguists
are trying to delimit boundaries between idioms and metaphors. While some argue
that idioms are dead or frozen metaphors, others maintain that idioms offer new
insights into language processing:
Recent research in psycholinguistics shows that the meanings of many idioms are
motivated by people’s conceptual knowledge, which includes metaphorical and
metonymic schemes of thought. In this way, the study of idioms reveals significant
aspects of how people ordinarily think. (Gibbs, 1994, p. 277)
Advances in cognitive research provide linguists not only with methodology to
analyse idioms but also enhances our understanding of natural language processing
(cf. Gibbs, 1994, pp. 278–288). Finally, partly owing to the rise of automated language
processing tools, linguists also engage in contrastive analyses of idioms across
different languages, which has numerous applications in the fields of translation
and language teaching/learning.

2.1. Definitions

According to the Oxford English Dictionary Online, the word idiom has several
senses which are related to language, of which the one relevant in this context reads
as follows:
a group of words established by usage as having a meaning not deducible from the
meanings of the individual words.
Notably, the word idiom has many other non-linguistic uses; furthermore, even as
a linguistic term, it may have different interpretations. It is thus not surprising to
see linguists suggesting that in order to minimize confusion the term should be
avoided altogether (Naciscione 2010, p. 18) because it has been used in linguistic
literature to refer to many different types of phrases which differ in the degree of
compositionality. In this course book, the term idiom is used in its narrow linguistic
sense as it is defined in the OED definition given above. Idioms are different from

C H A P T E R 2. I D I O M S  27
proverbs and sayings which are typically full-sentence compact expressions of folk
wisdom (e.g. East or West home is best). It should be noted, however, that scholars
often modify the scope of this notion so that it suits their research designs. It is thus
always necessary to consider carefully specific approaches used by every individual
author and find out what exactly the term covers in a specific study.
Table 2.2 contains a selection of definitions of idioms and synonymous
phraseological units from linguistic literature. The citations are taken from a
number of studies and grammars and offer a number of different approaches to the
understanding of this type of phrases. Yet even this small selection demonstrates
terminological variation in approaches to what may appear a fairly clear-cut
phraseological category. Since the approaches in some of the studies referred to in

Table 2.2 Definitions of idiom and similar phraseological units

Source Definition
Gibbs “Idioms have traditionally been defined as expressions whose
(1994, p. 91) meanings are noncompositional or not functions of the
meanings of their individual parts (…).”
Sinclair “An idiom is a group of two or more words which are chosen
(1991, p. 172) together in order to produce a specific meaning or effect in
speech or writing. (…)
The individual words which constitute idioms are not reliably
meaningful in themselves, because the whole idiom is
required to produce the meaning.”
Biber et al. “Idiomatic phrases—expressions with a meaning not entirely
(1999, p. 1024) derivable from the meaning of their parts (…).”
Fernando “A working definition of a pure idiom which is adequate for the
(1996, pp. 35–36) present is ‘a type of conventionalized, non-literal multiword
expression’. Spill the beans, for example, has nothing to do with
beans. In contrast to its literal counterpart meaning ‘letting
fall leguminous seeds’, a non-literal meaning is imposed on
the idiom as a whole: ‘commit an indiscretion’.”
Ishida “Idioms are multi-word expressions with the following three
(2008, p. 276) properties:
i. formal frozenness (…)
ii. syntactic frozenness (…)
iii. semantic frozenness (…)”
Naciscione “[T]he phraseological unit is a stable, cohesive combination of
(2010, p. 32) words with a fully or partially figurative meaning.”

28  CHAPTER 2. IDIOMS
Table 2.2 are broader or narrower in scope than the prototypical understanding of
the idiom, the scholars use other terms than idiom. As a consequence, terminological
variation in literature continues to flourish but it is most probably common practice
in any research field that only time and widespread recognition of individual terms
resolve the terminological confusion. For instance, Naciscione (2010) argues for
the advantages of the term phraseological unit which, in her approach, also covers
proverbs. In the opinion of Granger and Paquot (2008, pp. 42–43), phraseological
unit is a cover term that includes all types of phrasemes (cf. Figure 1.2) and is not
singled out in the classification which, interestingly, lists the term idiom. In order to
avoid misunderstandings, it is important to find out what exactly is meant by each
term. It may turn out that different terms refer to the same type of phrase, which
is obvious even from the quotes in Table 2.2 where definitions of the prototypical
idiom differ in scope.

2.2 Classification of idioms

The most straightforward way of grouping idioms could be derived from their
surface structure. Hence, idiomatic phrases in English can be

1 – noun phrases, e. g. pros and cons, chapter and verse ‘exact location or place’,
a drop in the ocean;
2 – verb phrases, e. g. to ring the bell, to make mountain out of a molehill, to come
clean, to go Dutch ‘pay your own bill’;
3 – prepositional phrases, e. g. in a nutshell, in the long run, on the spur of the
moment.

Biber et al. (1999, p. 1024) also mention wh-questions as another possible

structural pattern of idiomatic phrases, for example, what’s up?; what on earth …?
etc. Clearly, if the operational definition of idioms covers full-sentence items, for
instance, proverbs, the formal structural classification can be accordingly modified
to account for any relevant syntactic feature.
Differences in the degree of non-compositionality of idioms may serve as
another possible base for classification. For instance, Biber et al. (1999, p. 1025)
illustrate how two idioms differ in the extent to which their meanings “can be
derived from the components parts” by considering the following two expressions:
change one’s mind ‘rethink a decision’ and kick the bucket ‘die’. Clearly, the meaning
of the first expression is much closer to its literal meaning than is the case in the

C H A P T E R 2. I D I O M S  29
second expression and on this basis it is logical to speak about idioms with different
degree of non-compositionality.
A more thought-provoking approach to the classification of idioms was applied
in a recent project, ‘Widespread idioms in Europe and Beyond’ (see the website
at http://www.widespread-idioms.uni-trier.de/; also cf. Piirainen, 2008), where
idioms were classified into five groups according to their cultural foundations. For
instance, idioms with textual dependence are such phraseological units whose
interpretation and origin draw upon textual sources (the Bible, fairy tales or
classical literature). The other groups in this classification cover idioms originating
in pre-scientific conceptual domains, cultural symbols, aspects of material and
social culture. Apparently the aims of the project, formulated as identification of “the
core set of idioms that actually exist in many languages, Europe-wide and beyond”
(quoted from the project website) predetermined the basis for the classification
and allowed researchers to draw relevant comparisons and analyse the obtained
data. Whatever the approach to classification, it should be meaningful to the aims
of the study in question.

2.3 Designing a study on idioms

Research projects of idioms in one language or in a contrastive perspective across

several languages are popular among students of linguistics. A few important steps
should be considered in order to design a valid study. The following checklist is
meant to provide basic guidance to those interested in the study of idioms, especially
involving corpus evidence. For a more detailed description of methodology for
contrastive analysis of idioms, the reader is directed to Ishida (2008) which could
be considered an exemplary paper demonstrating the contrastive idiom analysis.

2.3.1. Choosing your data

Compiling a list of idioms that describe, for example, human traits of character
is not difficult. There are many dictionaries of idioms; plenty internet sites offer
specific compilations of phrases in one or more languages. So collecting primary
data from many sources is not particularly difficult, but it becomes clear very soon
that something should be done with the long lists of idioms. It is therefore advisable
from the very beginning to target a limited number of idioms related to one specific
issue. It could be, for instance, idioms describing one particular trait of character or
physical feature (brightness, stupidity, stubbornness, height etc.). It will certainly

30  CHAPTER 2. IDIOMS
yield a smaller data set but will offer more opportunity for an in-depth analysis,
which is always more informative and, as it happens, more interesting than a
superficial overview of a long list of items.

2.3.2. Objective judgement

In the studies of idioms, interpretation of idiomatic meanings, their implications
or sometimes connotations should be made as objective as possible. The point of
departure is always a lexicographic publication. Consulting several dictionaries in
order to obtain authoritative definitions is essential. Apart from it, one could also
engage human informants as an additional source of information. It is common
practice in linguistic research to use native speaker informants (or, on the contrary,
involve non-native speakers if the aims of the study in question deal with non-
native interpretations of idioms). A small experiment when several informants are
asked to comment on sample sentences containing idioms or identify idioms whose
meanings are (non-)equivalent, or provide translational equivalents etc. allows the
researcher to minimize subjectivity and, as a consequence, increase the validity of
the study. An extensive survey of informants is also possible, especially with plenty
of internet tools freely available, but it will inevitably increase the scope of study
and require basic statistical skills to process the data.

2.3.3. Co-occurrence analysis

In order to capture the meaning of an idiom, its use in natural language should
be carefully analysed. Here come corpora which represent different language
varieties and offer plenty of ways to reveal how a particular idiom, let us say, on
and off, is different from its non-idiomatic synonym occasionally, and whether it
is equivalent to its renderings into other languages (Lith. priešokiais, retsykiais,
kartais). It is essential to take into account naturally co-occurring data and examine
lexical and grammatical patterns in which the idiom is used. Also, grammatical
forms or constraints in transformation might prove that the idiomatic meaning is
predominantly realised in one particular form of the idiom. For instance, the idiom
soaked to the skin ‘very wet’ to judge from the BNC output seems to show preference
for the verb in the Past Simple tense whereas when the prepositional phrase to the
skin occurs with verbs in other tense forms, it is usually meant literally as in the
following example:
(7) Exercise stimulates blood flow to the skin and so gives rise to a healthy
appearance (...) (BNC)

C H A P T E R 2. I D I O M S  31
 Corpus-based analyses of idioms involve a close examination of concordances.
In a contrastive study where more than one language is considered, a researcher
might find himself/herself going through hundreds of instances, especially if the
item in question allows for the literal and idiomatic interpretations. Hence the
decision to set out on a corpus-based study of idioms inevitably means that it is
important from the very beginning to consider carefully how the scope of the study
could be delimited, as it was argued in section 2.2.1.

Further reading

 Idioms in the functional language perspective: Fernando (1996);

 Contrastive idiom analysis: Ishida (2008);
 Idioms in European languages: Piirainen (2008).

Study tasks

1. As you know, idioms are not necessarily reproduced in their prototypical form.
Quite often they are used with certain elements omitted and/or transformed.
Using the BNC or COCA, check variability of the idioms given below and decide
what is their most typical form in English:
Keep a straight face;
Make a mountain out of a molehill;
Pull someone’s leg;
To ring a bell;
To take pot luck;
To take something with a pinch of salt.
2. Gibbs (1994, pp. 92–93) gives an account of several experiments on how people
process idioms which have both a literal and idiomatic reading. For instance,
he kicked the bucket can be understood in two ways: literally as ‘he pushed
the bucket with his foot’ and figuratively as ‘he died’. The insights from such
experiments are twofold. It was found that people tend to go for the literal
interpretation first; yet once they know that the context supports idiomatic
interpretation, they choose to process phrases as idioms. In your opinion, which
of the readings of the following sentences given below (all of them taken from
the BNC) is faster or primary? In your opinion, what does it depend on? Try to
think about how you mentally process the sentence. How would you design an
experiment to verify your answer?

32  CHAPTER 2. IDIOMS
 1) No sooner was I off the train than the guard blew the whistle and the train
started and I had to run for it.
2) It could spot border violations and blow the whistle on breaches of cease-fire
agreements.
3) With continual borrowing over two years, the bank had blown the whistle.
4) We have teething problems to sort out.
5) ‘Despite various teething problems we are very happy in our new life,’ said
Mr Turner.
6) They blame budgetary pressures, persistent teething problems, and cost
overruns.
3. Translation of idioms is always challenging. On the one hand, to preserve stylistic
features of a text, written or spoken, a translator should aim at phraseological
equivalence and thus render source language idioms into equivalent idioms
of the target language. The difficulty lies in the fact that the translator needs
to find an idiom that has an equivalent meaning while the wording may be
very different for the source language. How would you translate the following
sentences from the BNC into another language?
1) They used to pull my leg, but once they’d had a go they soon changed their
minds.
2) They took to their heels and ran up the road.
3) It was interesting first time out, but after a couple of them I realised it wasn’t
my cup of tea.
4) Nothing’s going to be hidden, no skeletons in the cupboard, no dark secrets,
everything out in the open where I can deal with it.
5) Pushing the wheelbarrow should have been child’s play, but I still could not
get the hang of it.
6) Then suddenly, right out of the blue, it had gone straight down the drain.
7) When positive news did come, it again seemed to arrive out of the blue.
8) Bargain-hunters (…) have the right to demand their money back if what they
buy turns out to be a pig in a poke.

C H A P T E R 2. I D I O M S  33
CHAPTER 3.
COLLOCATIONS

One of the most significant contributions of corpus linguistics to the study of lexis
and, more specifically, phraseology is undoubtedly the discovery and description of
collocations. In contrast to idioms, collocations are said to be ubiquitous—albeit
frequent they are not easily spotted. Such phrases as take a risk, make fortune,
tender meat, high temperature are seemingly simple expressions and not eye-
catching vivid idioms. Moreover, collocations are often semantically transparent
because they do not carry hidden figurative meanings. Yet for whatever reason the
combinations *fetch/grab a risk; *do/produce fortune; *soft meat; *tall temperature
do not sound good in English, or are not acceptable as standard ways of expressing
the ideas. Apparently, there are certain restrictions in the combinability of words
while transparency and fixedness of collocations is a matter of degree.
It was only after the arrival of corpora that lexical co-occurrence patterns were
observed and described in a comprehensive manner. While linguistic research
into collocations started gaining pace after the publication of materials from the
COBUILD project which was jointly undertaken by Collins Publishers and the
University of Birmingham and headed by John Sinclair in 1980s, the notion of
collocation was not altogether new to linguists.
Collocation of words has long been recognized as a way to study meaning. The
idea of meaning through collocation goes back to J. R. Firth (1890–1960) who
provided the theoretical background to the contextual theory of meaning:
Meaning by collocation is a direct consequence of the fact that, for Firth, the
meaning of words lies in their use, and established usage will recognize words
“familiar and habitual company” (Tognini-Bonelli, 2001).
In other words, the co-text4 of a word gives a researcher essential evidence on
which any inferences about the meaning of that word could be drawn. Corpus

4 The term co-text is understood here as it was explained in Tognini-Bonelli (2001, p. 87): the term
is used “to refer specifically to the verbal environment that we are aiming to formalise, and the
term context to refer to the situational and cultural parameters involved in the interaction (…)
contextual elements, such as the relevant participants or the relevant object, will often have a
correlate linguistic realisation in the co-text.”

34  C H A P T E R 3 . CO L LO C AT I O N S
evidence shows that words are selective as to their partners, and the fact that one
word co-occurs more frequently with the other than we would expect by chance
most probably implies that we deal with a realisation of a specific sense of that
word. The classical quote of Firth “you shall know a word by the company it keeps”
is often cited in linguistic literature to explain the phenomenon of collocation and
its role in meaning.
As regards phraseological research, the discovery of co-occurrences of words put
onto the agenda of vocabulary studies a new type of phrases, namely, collocations.
It turns out that words do not occur one by one in naturally produced language—
having said one word we tend to choose another which goes well with the previous
one. For example, environmentally evokes friendly but not any other of synonyms of
friendly, for instance, amiable, amicable, good-natured, kind, pleasant etc. Similarly,
mistakes in English are made but not done whereas research, on the contrary, is
done. It was John Sinclair (1991) who brought the combinability of words into
mainstream linguistics and showed that words build up a network of links with
other words which occur in their co-text, or, to put it differently, attract each other.
In a way, he developed Firth’s ideas and argued that co-occurrence of words is the
way language realizes meaning, and evidence from language corpora is essential for
the study of meaning (cf. Sinclair, 2004, p. 134).
To explain how meaning is created Sinclair proposed two principles which are
inactivated in the process of language production: the open-choice principle and
the idiom principle (Sinclair, 1991, pp. 109–112). The open choice principle implies
that language is a structure consisting of slots which are filled in with grammatically
acceptable words. The ‘slot-and-filler’ model of language is at the basis of all
traditional grammars (ibid., p. 110). For example, let us consider an English sentence
having the following structure: S → NP V NP, for example, He likes ____, where the
empty slot could be realized by any noun phrase (Julia, her, surfing, cakes, reading
etc.) that is suitable according to the rules of grammar. Such decisions are governed
by the open-choice principle. The idiom principle, in contrast, works beyond the
limits of one sentence and is opposed to the open-choice principle. The idiom
principle requires language users to select such words which naturally combine
with each other. If we want to say that coffee has a high concentration of caffeine, it
is typical to describe it as strong, for instance: A coffee please, not too strong (BNC).
If grammaticality were the only consideration in this utterance, it would also be
possible to use synonymous words, namely, powerful or forceful, but they are clearly
unacceptable to describe the drink. Hence, the combination of coffee and strong is

C H A P T E R 3. CO L LO C AT I O N S  35
a realisation of the idiom principle when the words are co-selected not because of
their grammatical properties but owing to certain constraints in their meaning.
Furthermore, on the surface of it, the attraction of words to each other might
appear to be inexplicable as it is difficult to arrive at a logical explanation why,
for instance, the English adjective high attracts a different set of nouns than tall
when it is used in English (Table 3.1), because any dictionary will provide similar
definitions of high and tall suggesting that the two adjectives essentially mean the
same. But even a very superficial comparison of the two sets of collocates shows
that high collocates with nouns that have more abstract meanings whereas tall is
often used to characterise more concrete nouns.

HIGH + court, street, level, quality, TALL + man, figure, trees, woman,
rates, proportion, buildings, buildings, windows, chimneys, boy,
standards, unemployment, degree, tower, ships, girl, glass, order,
speed, school, ships, risk, pressure grass, houses

Figure 3.1 Noun collocates of HIGH and TALL

Such differences between the two adjectives are difficult to arrive at by intuition.
Admittedly, native speakers are often capable of determining whether two words
go together well but even their judgment may be limited and related to different
social factors, such as the level of (linguistic) education, occupation etc. Luckily,
corpus linguistics resolved the issue of subjectivity and replaced the human
intuition with corpus evidence. Sinclair (1991) convincingly showed that the
tendency of a particular word to appear in the context of some other words is
invisible to the naked eye and can only be reliably discovered by analysing corpora.
Hence, a representative corpus of a language is necessary is order to identify typical
‘companions’ of any word.
Before delving deeper into collocations, some clarification of terminology
is due. As the reader might have noticed, the term collocation may be used as an
uncountable noun to refer to the phenomenon of word co-occurrence. A collocation
usually refers to one particular phrase. Collocations consist of two words which
collocate with each other. Usually, the word which is being analysed or queried in a
corpus is referred to as the node whereas its ‘partners’ are called collocates. It should
also be noted that to collocate (verb) is pronounced differently from a collocate

36  C H A P T E R 3 . CO L LO C AT I O N S
(noun). When looking up for collocates of a particular word, we need to specify the
distance from the node to the left and right. The distance, also known as the span
of the window, is measured in words, the default in many corpus access tools being
five or four words. Meant humorously, the same terminology is sometimes used by
linguists to refer to people, and it is not unusual to hear them say that they often
“collocate” with one or another colleague.

3.1 Definitions
Back in 1965, a simple and beautiful definition of collocations in English and
French was proposed by Sir Paul McCartney in what turned out to become an
award-winning song of the Beatles:
Michelle, ma belle
These are words that go together well
My Michelle
Michelle, ma belle
Sont les mots qui vont tres bien ensemble
Tres bien ensemble
As it happens, the famous songwriter and singer covered at least two essential
aspects of collocation: it consists of more than one word; the words form a well-
sounding combination. Linguists, as could be expected, had more to add.
Generally speaking, the definition of collocations can be formulated in two ways:
it can draw on the phraseological or frequency-based approach to the understanding
of collocation (cf. section 1.3; also see Nesselhauf, 2005, p. 12). The two approaches
reflect not only the conceptual differences behind the notion but also pertain to the
ways in which collocations are extracted from a corpus.
The phraseological approach takes into account the degree of restriction in the
meaning of the constituents of a collocation and goes back to the classification of
phrases proposed by Cowie (1981). The term collocation in this approach refers to
phrases that are made up of constituents that are not freely substitutable; hence
Cowie’s term restricted collocations. For example, to hold talks ‘negotiate’ would
be considered a restricted collocation because the verb is used in its non-literal
meaning, the whole phrase is transparent, and it allows some substitution as in
to have talks (cf. Nesselhauf, 2005, p. 14). The main aspect of the phraseological
definition of collocations is commutability, i. e. the extent to which individual
constituents of a collocation may or may not be freely substituted with synonymous
words.

C H A P T E R 3. CO L LO C AT I O N S  37
 The phraseological approach most often points to certain peculiarities of
data collection. In practice it means that a set of collocations which are defined
phraseologically is inevitably a subjective selection of items performed manually
by the researcher from a raw output of a corpus analysis program or manual
text analysis. To put it differently, it is the researcher’s responsibility to decide
for each individual collocation if its constituents are freely substitutable or their
substitution is restricted. Admittedly, to minimise subjectivity of judgment, it is
possible to set up a transparent methodology. For instance, Nesselhauf (2005, p.
54–63) gives a detailed account of steps in the process of data selection which
was used to operationalise her phraseological definition of collocations. Manually
extracted collocations were carefully considered to decide how commutable they
are. The linguist used information given in monolingual English dictionaries which
“was then supplemented with information from corpus analysis and from native
speaker tests” (ibid, p. 54). So whatever was found in the dictionaries and the BNC
was put to test with native speaker informants who either confirmed or rejected
certain forms of collocations as (un)acceptable in English.
The frequency-based approach, in contrast, is more inclusive as it identifies
collocations on the basis of statistical co-occurrence. In principle, it means that
any combination, no matter how non-compositional or idiomatic it is, may be
considered to be a collocation if it meets certain statistical parameters. These
parameters are understood as co-occurrence of two words, i. e. constituents of a
collocation, significantly more frequently than it could be expected by chance. The
statistical definition of collocations is thus based on specific statistical measures
which will be overviewed in Section 3.3. In contrast to the phraseological approach,
in the statistical approach collocations are extracted from a corpus automatically
whereas their co-occurrence statistics is provided as a built-in function in most
corpus analysis tools that are available to researchers. So automatically obtained
data is objective even though it does require some interference of the human analyst.
The automatically produced list of collocates contains a considerable amount of
so-called ‘noise’, or linguistically uninteresting collocates. For instance, functional
words are sometimes less interesting to scholars who focus on lexical collocations,
but others, on the contrary, might be investigating collocations with prepositions.
It is thus the task of the researcher to go through the list and manually select the
relevant collocates.
The choice of one or the other approach to the definition of collocations always
depends on the aims and scope of research. Also, it is not unusual to see studies
where a combination of the two approaches is used. First, statistical collocations

38  C H A P T E R 3 . CO L LO C AT I O N S
are retrieved from a corpus and then a researcher manually filters all collocates
to select the most salient ones. Yet the approach used is not necessarily directly
obvious from the definitions found in literature. Table 3.1 offers a selection of
definitions of collocations from a variety of sources. Each of them was formulated for
different aims and published in books or research articles that were not necessarily
concerned with phraseology.

Table 3.1 Definitions of collocations

Source Definition
Firth Firth suggested that collocation is a way to analyse meaning:
(1957, p. 196) “Meaning by collocation is an abstraction at the syntagmatic
level and is not directly concerned with the conceptual or
idea approach to the meaning of words.”
Halliday and Hasan Collocation is “a cover term for the cohesion that results
(1976, p. 287) from the co-occurrence of lexical items that are in some way
or other typically associated with one another, because they
tend to occur in similar environments”.
Sinclair “Collocation is the occurrence of two or more words within
(1991, p. 170) a short space of each other in text. The usual measure of
proximity is a maximum of four words intervening.”
Kjellmer “[C]ollocations are defined as recurring sequences that have
(1991, p. 116) grammatical structure.”
Biber et al. Collocations “are associations between lexical words, so that
(1999, p. 988) the words co-occur more frequently than expected by chance.
(…) Unlike idioms, collocations are statistical associations
rather than relatively fixed expressions. Moreover, the
individual words in a collocation retain their own meaning.
However, part of the extended meaning of a word is the fact
that it tends to co-occur with a specific set of collocates.”
Nesselhauf “[C]ollocations are considered a type of word combination in
(2005, pp. 25–34) a certain grammatical pattern” (p. 25)
The definition “has only been developed for verb-noun com-
binations” (p. 25).
“The noun can be used without arbitrary restriction in the
sense in which it is being used, but the verb is, in the given
sense, to some degree arbitrarily restricted to certain nouns.”
(p. 33).
Vilkaitė “In this study, a corpus-based statistical approach of defining
(2016, pp. 32–33) collocations as frequent co-occurrences, without any addi-
tional semantic criteria is adopted. Examples of such colloca-
tions could be take care, last night, learning difficulties.”

C H A P T E R 3. CO L LO C AT I O N S  39
Whatever the definition, it is always important to carefully consider why the
linguists chose to write about collocations. Knowing the broader context, it is
easier to understand why one or the other approach to the definition is preferred
and appreciate why certain aspects of collocation are given more prominence than
others. Clearly, a mere choice of the term collocation does not presume a unified
approach to its operationalisation.

3.2 Classification of collocations

To classify collocations one may proceed from their formal structure and identify
different syntactic patterns, for instance,
(1) verb + noun, e. g. to shrug one’s shoulders, to compile a corpus
(2) adjective + noun, e. g. rancid butter, low pressure
(3) adverb + adjective, e. g. perfectly correct, utterly boring etc.
A detailed classification based on the word classes in which constituents of
collocations appear is provided in the “Oxford Collocations Dictionary for Students
of English” (2002, p. ix). Benson et al. (1993), compilers of the “The BBI Combinatory
Dictionary of English”, also identify structural groups of collocations which they
generally divide into two broad types: grammatical and lexical, as explained in the
following excerpt:
Grammatical collocations consist of a dominant word—noun, adjective/participle,
verb—and a preposition or grammatical construction. Lexical collocations, on
the other hand, do not have a dominant word; they have structures such as the
following: verb + noun, adjective + noun, noun + verb, noun + noun, adverb +
adjective, adverb + verb. (Benson et al., 1993, p. ix)
In many corpus-based and corpus-driven studies of collocations, linguists choose
to focus on one particular syntactic type of collocations as it helps to obtain a
reasonable and manageable amount of data. Coverage of a broad range of syntactic
types, however, is preferred in such studies where collocation is not an object in itself
but rather a means of investigating meaning of individual lexemes, synonymous
pairs of words etc. Then a full picture of all types of structural categories might
help a scholar to uncover specific patterns in the use of a word which, among other
things, often sheds light on nuances in the meaning of that word.
Since collocations are combinations of words on a cline of non-compositionality,
they can be also classified on the basis of as literalness. Hence, some collocations are

40  C H A P T E R 3 . CO L LO C AT I O N S
fully transparent (give a present) whereas others more opaque (give a headache).
Between the two extremes, one might identify a broad category of medium non-
compositionality (give a sign). One way of differentiating among the three examples
is focusing on the verb, in this case dealing with the action of giving, and the object
that is being given. Giving usually involves a transfer of an object from the agent
to the recipient, which is exactly the case in give a present: one person transfers
the object into the possession of the recipient. Headaches are clearly different in
this respect as they cannot be transferred in the same manner as presents; they
are sooner caused. Moreover, the doer of the action (agent) who gives a headache,
‘shares’ something he does not possess himself. Give a sign falls between the first
two cases as the object sign belongs to the agent although it is not possessed in the
same manner as a present, but neither is it so detached from the agent as in the case
of headache. The opacity of these three collocations largely pertains to the verb, yet
in other collocations, for example, to show the way, to get into the way and way too
much, opacity stems from the noun.
Finally, collocations can be classified on the basis of commutability, understood
as substitutability of their constituents. On the surface of it, commutability seems to
be easier to identify as it refers to the constituent which can or cannot be substituted.
For instance, high temperature is a collocation where none of the constituents
can be replaced without a major change in meaning (low temperature refers to a
different state of affairs), and it is impossible to say *tall/elevated temperature. But
the noun survey as in to conduct a survey collocates with several verbs of similar
meaning, namely, carry out, do, make, undertake, so the two collocations, namely,
high temperature and to conduct a survey, fall into two different groups on the basis
of commutability: one is completely fixed whereas the other allows variability of
one of its constituents.
Admittedly, a specific research design might dictate other approaches to
classification, so this overview is in no way exhaustive. For instance, a study
concerned with translation of collocations might propose categories related to
translatability or translation strategies suitable for rendering collocations from one
language into another. Hence, the choice of classification partly echoes the choice of
the defining approach—it should be first of all relevant and meaningful to research
questions formulated by the scholar rather than blindly follow an authoritative
source.

C H A P T E R 3. CO L LO C AT I O N S  41
3.3 Association measures

The discussion of ways to classify collocations so far has left out one of the
most popular attributes used to describe this type of phraseological units, namely,
weakness and strength, which are directly related to statistical parameters used
in collocational research. More specifically, they refer to the strength of attraction
between constituents of a collocation. Apparently, words, just like people,
‘attract’ each other, and the strength of attraction can be measured (although the
measurement has not yet been shown to extend to attractions between human
beings). This section will overview the basics of association statistics and will
introduce several statistical measures readily available on corpus access sites and
in corpus processing programs.
To understand the underlying arithmetical operations which are used to
evaluate the strength of collocations in a corpus, let us take a closer look at how
such measures are computed. What is measured, in fact, is the probability that one
word will trigger the occurrence of the other. How certain are we that, for example,
the use of global will necessarily trigger warming? Or vice versa, is it possible
to claim that warming is a good predictor of global? To calculate the degree of
attraction between two words, or even a word and a construction, the following
frequencies should be known (cf. Levshina, 2015, pp. 223–224): global warming
(599 occurrences in the BNC); global except its occurrences in global warming
(2922 occurrences); warming except its occurrences in global warming (491). The
data is usually arranged as a contingency table (see Table 3.2), which, besides the
three mentioned numbers, also includes the total size of a corpus that we need not
take into account to measure attraction but that is necessary for more sophisticated
statistical measures.

Table 3.2 Co-occurrence frequencies of global and warming in the BNC

global global
(all other occurrences of
warming without global)
warming 599 491
warming 2922 98,308,219
(all other occurrences of (the total number of words
global without warming) in the corpus minus three
numbers given in the other
three cells)

42  C H A P T E R 3 . CO L LO C AT I O N S
 The following computations should be performed with the data:
to measure the probability that global predicts warming: 599/(2922+599) =
0.17, or 17%;
to measure the probability that warming predicts global: 599/(491+599) =
0.55, or 55%.
So it is possible to conclude that warming is a much stronger predictor of global in
the BNC because there is a probability of 55% that, in this corpus, we will find this
adjective in the co-text of warming. In contrast, global has a weaker association
with warming so there is less chance that we will see warming in the same sentence
as global. Measures like these are known in corpus statistics as Attraction and
Reliance (Levshina, 2015, p. 228), both being fairly simple unidirectional measures
of association.
More statistical knowledge is needed in order to understand computation
of bidirectional measures which take into account not only the frequency of co-
occurrence of two words, but also the total corpus size. Luckily to many enthusiasts
of corpus linguistics at different stages of proficiency, such tools are built-in in most
corpus analysis tools and do not require a special training in statistical methods. It
is, however, useful to have a basic understanding of some widely-used association
measures in order to be able to interpret the automated computations. To begin
with, it will suffice to know that bidirectional measures are based on statistical
independence tests that evaluate how significant is the difference between the
actual co-occurrence frequency of two words in a corpus and their expected
frequency if no association between collocates is assumed (Levshina, 2015, p. 225).
The more significant the difference, the stronger is the association in question. The
ways to quantify that significance are many—Bouma (2009, p. 1) mentions over 55
measures, all involving specific statistical tests.
One of the most widely used measures is Mutual Information (MI). This measure
is the default in the COCA output of collocates; it is also reported on the BNC site on
the collocations output screen (one should select ‘Mutual Information’ in the drop-
down menu for statistics). Mutual Information
compares the probability that the two items occur together as a joint event (…)
with the probability that they occur individually and that their co-occurrences are
simply a result of chance (McEnery and Wilson, 2001, p. 86).
MI scores are fairly easy to interpret: the higher the score, the stronger the association
between two words. Table 3.3 lists a few collocates of the noun Christmas in the

C H A P T E R 3. CO L LO C AT I O N S  43
BNC output with different MI values listed. It should be remembered, however, that
collocates do not necessarily represent adjacent sequences of words. The program
reports any word that occurs within the window span of three words to the left/
right from the node Christmas. To check whether it is an adjacent or rather non-
adjacent sequence, one should view the actual concordance lines (easily done by
going to the column ‘Observed collocate frequency’ in the output page and clicking
on the number).

Table 3.3 A selection of collocates of Christmas from the BNC

Collocate MI score
eve 8.79
decorations 8.48
carol 6.81
greetings 5.82
preparations 4.80
hope 0.11
house 0.16
parents -0.35
services -0.73
problems -1.12

A high MI score proves that the two words (the node and respective collocate) are
strongly connected, for instance, Christmas + eve, Christmas + decorations. If the
score is negative, as in the last three rows of Table 3.3, the two words occur more
often in isolation than in the same window span. Values that are close to zero (hope
and house) point to such items whose occurrence in the proximity of Christmas may
be a chance event.
Another widely-used association measure is t-score. It is, however, markedly
different from MI, as t-score reflects typical associations with high-frequency words.
The difference between MI and t-score is related to the type of co-occurrences
that are treated by the program as significant. We have seen that MI collocates
are sometimes fairly infrequent words in a language (eve, carols) but they build
quite exclusive links with the node and form salient lexical collocations. When
computing MI scores, the program disregards the absolute (raw) frequency of each
word and gives more importance to such collocates that are normally rare words in
language (cf. Gries, 2015) and their co-occurrence with the node is thus treated as a
significant event. Collocates ranked by t-scores, in contrast, include high-frequency

44  C H A P T E R 3 . CO L LO C AT I O N S
words, functional word classes among them, therefore collocate rankings are very
different from MI lists.
To illustrate differences among various association measures, Table 3.4 shows top
twenty collocates of the noun lung in the BNC ranked by the strength of four types of
measures. Besides MI and t-score, it also includes z-score whose computation takes
into account not only the frequencies of two collocates but also the frequencies of all
other words that occur in the specified window span (McEnery and Wilson, 2001,
p. 86). The Log-likelihood measure, which is the default association measure in the
BNC output, is computed with regard to absolute frequencies of each collocate and
its list of collocates roughly falls in-between the MI and t-score lists.

Table 3.4 Top twenty collocates of lung in the BNC ranked by different association
measures
rank MI t-score Log-likehood z-score
1 parenchyma cancer cancer cancer
2 asbestosis and heart parenchyma
3 punctured heart function punctured
4 kai his disease asbestosis
5 nonsmokers function air heart
6 et-1 her his kai
7 lobes into and transplant
8 cancer from punctured function
9 obstructive air smoking lobes
10 transplant disease transplant obstructive
11 bronchitis your liver et-1
12 pulmonary my into nonsmokers
13 lymph with her liver
14 cancers the deaths smoking
15 kidneys in tissue disease
16 fibrosis smoking collapsed cancers
17 carcinoma risk chronic pulmonary
18 bursting patients risk deaths
19 respiratory liver parenchyma collapsed
20 liver deaths lobes chronic

As seen in Table 3.4, collocations with the highest MI are field-specific terms,
medical in this case, whereas the t-score list includes many high-frequency words.
Therefore, MI rankings might be relevant in studies dealing with terminology or more
salient collocations (cf. Wang, 2016, p. 61). If one is concerned with grammatical

C H A P T E R 3. CO L LO C AT I O N S  45
patterns of the node, collocates ranked by their t-score values are certainly more
informative. While the z-score association measure is said to be better suitable for
the extraction of collocates from a corpus of special text types, for example, literary,
the Log-likelihood measure seems to yield a combined version of the MI/z-score
and t-score list.
There is still no one clear answer in literature as to which of the available
measures of association is the best. An approach quite often applied in studies of
collocations is a combination of several measures when a researcher considers
rankings for several different measures and produces a combined list of collocates
that have high ranks by the majority of measures tested. Such approach is supported
in literature:
One of the lessons taught by systematic evaluation of association measures against
different gold standards is that there is not one association measure that is best in
all situations. Rather, different target collocations may be found most effectively
with different methods and measures. It is therefore useful to have access to a wide
array of association measures coupled with an understanding of their behaviour
if we want to do collocation extraction. (Bouma, 2009, p. 2)
So the real collocations are most probably not to be established by application of
one association measure. Interestingly, the only collocate that appears in all four
columns of Table 3.4 is cancer; collocates that rank high for at least three measures
are punctured, heart and function. This is hardly a list of collocates of the lexeme lung
that a native speaker of English would produce if asked to give the most common
words that lung combines with.
The remaining unanswered question is what kind of mental links one or another
association measure captures most effectively. The author of a recent textbook on
statistics in corpus linguistics very aptly observed that it is not quite clear what
kind of reality is reflected in corpora:
It is still unclear which [association] measures represent the information that
the speakers store in their minds more adequately than the others, since recent
empirical studies based on corpus-based and experimental evidence have yielded
divergent results. (Levshina, 2015, p. 238)
So it is perhaps impossible to give a simple answer to the question of whether a
particular phrase is a collocation because we still know very little about how
associations between words are processed in the brain. As it should be clear to the
reader by now, to call any phrase a collocation a linguist would first need to know

46  C H A P T E R 3 . CO L LO C AT I O N S
the approach to its definition (phraseological or statistical) used; the size of the
corpus and the association measure used to extract collocates. If a phrase meets the
phraseological definition, is extracted from a representative corpus of a language,
has a high rank in the lists based on two or more association scores, we might be
dealing with a real collocation. It remains to see, however, if findings from a corpus-
driven analysis are confirmed by psycholinguistic experiments.
If one may think that measures of association between words are difficult to
grasp, it should be remembered that they are merely capturing the linguistic
competence which is possessed by any native speaker of a language. Links between
words may not be visible to the naked eye yet native speakers have an intuitive
awareness of combinatory properties of words and their co-selective preferences.
Non-native speakers, in this respect being in a less advantageous position, need to
develop their collocational competence in order to be able to produce native-like
sounding natural language. So the association measures discussed in this section
are not complex; complexity is the feature of the empirical world and one of its
parts, i. e. our mental lexicon, which is so masterly and smoothly controlled by the
human brain.

3.4 The extended unit of meaning

The analysis of collocates, indicating lexical and grammatical patterns of word

co-occurrence, is closely related to the development of the theory of extended
unit of meaning (Sinclair, 1996). It introduced the lexical item as a new and more
broadly understood lexical structure which in the hierarchical structure of language
deserves a place above the word. The identification of the lexical items draws on
evidence from four levels of analysis required for the description of lexical items:
1. Collocation;
2. Colligation;
3. Semantic preference;
4. Semantic prosody.
Sinclair maintained that those “structural categories” had the potential of assuming
“a central rather than a peripheral role in language description” (2004, p. 39) as
they provide a comprehensive picture of any lexical unit, be it a word or a phrase.
His analyses of the phrases naked eye, true feelings and the lexemes brook (verb)
and place (noun) are convincing demonstrations of how the four levels may be
instrumental in language study (2004, pp. 24–48). More importantly, a purely

C H A P T E R 3. CO L LO C AT I O N S  47
corpus-driven approach was shown to provide an objective rather than intuition-
based account of language use.
So what are the four levels of analysis? Collocation, as the term suggests, is the
analysis of collocates of a particular word or phrase. The resulting collocational
profile of a lexical unit shows significant lexical realisations of words that occur
in the co-text of the item under study. Colligation refers to a set of grammatical
choices, namely, articles, prepositions and specific constructions which occur when
the item is used in natural language, or a corpus. Semantic preference pertains to
semantic fields which are represented by the lexical collocates of the item. Finally,
semantic prosody is the level of analysis that deals with pragmatic functions and
involves examination of, for instance, negative, positive or neutral implications, of
the analysed lexical items.
The four levels proposed as a means to describe any lexical item provide a robust
methodology to anyone willing to describe a particular word or phrase; they also
serve as a practical step-by-step approach to the study of synonymous words.

Further reading

 Cross-linguistic and contrastive studies of collocations: Molina-Plaza & de

Gregorio-Godeo (2010); Xiao & McEnery (2006);
 Collocation as a way to describe meaning: Moon (2008); Römer (2009a); Stubbs
(2007);
 Classification of collocations: Stulpinaitė et al. (2016);
 Collocations in second language acquisition: Durrant (2014);
 Collocational resonance: Williams (2008);
 Ways to test word associations: Zareva & Wolter (2012);
 The extended unit of meaning in vocabulary research: Tognini-Bonelli (2001).

Study tasks

1. Collocation and collocations were first described in detail when electronic

corpora came into being. Can you explain why lexical patterns were discovered
only through the study of corpora while grammatical patterns had been
extensively described by linguists in the pre-corpus era?
2. Extract frequency data of the words you and know from the spoken subcorpus of
the BNC and compare the strength of the collocation you know with its strength

48  C H A P T E R 3 . CO L LO C AT I O N S
 in the spoken part of COCA. Compare your findings with the data given on p. 11
in Chapter 1.
3. Check which word of the following phrases is a better predictor of the other
one? Is it the same in the BNC and COCA?
a splitting headache
a sweeping statement
tectonic plates
to clench one’s teeth
to look back
to tell the truth
to nod one’s head
to shrug one’s shoulders
totally crazy

4. A list of collocates of a particular word may differ from one variety of English
to the other (British vs. American, spoken vs. written). Compare collocates of
the following lexemes in different registers of English represented in the BNC
or COCA:
Nouns: point; sense; way
Verbs: like; give; set

5. Collect relevant corpus data in order to answer the following questions: What
is the collocational resonance (see Williams (2008) for the explanation and
methodology) of pen friend (British English) and pen pal (American English) in
current language use? Has it been changing over the last decades? Run searches
on COCA, COHA (Corpus of Historical American English accessible through the
COCA site) and the BNC.

6. Read the following extract from the style guide of The Economist about the
meanings of quite in different varieties of English. Extract adjective collocates
of quite from the BNC and COCA to verify the claims of editors at The Economist.
quite In America, quite is usually an intensifying adverb similar to
altogether, entirely or very; in Britain, depending on the emphasis,
the tone of voice and the adjective that follows, it usually means
fairly, moderately or reasonably, and often damns with faint praise.
(Economist Books, 2005, p. 119)

C H A P T E R 3. CO L LO C AT I O N S  49
7. Read another extract from the style guide of The Economist:
red and blue In Britain, colours that are associated with socialism
and conservatism respectively; in the United States, colours that are
associated with Republicans and Democrats respectively. (Economist
Books, 2005, p. 120)
How would you check whether the claims are valid? Use different corpora to
provide evidence. Pay attention to possible differences across registers: spoken,
written (newspaper language vs. academic) etc.

8. To see the difference between intuitive linguistic judgment and corpus evidence,
do a simple experiment. Ask your friends to write down 5-7 words that often
go together with despair, spoon, looming. Then extract collocates of these words
from a corpus of English. Do you observe any differences between human
intuition and corpus evidence?

9. Analyse corpus evidence to capture differences between the following pairs of

synonyms:
Verbs: attempt vs. try vs. endeavour; buy vs. purchase
Nouns: building vs. edifice; research paper vs. research article
Adjectives: big vs. large, little vs. small, beautiful vs. handsome;
subjective vs. biased; good vs. kind; quick vs. fast
Adverbs: soon vs. shortly; absolutely vs. totally
Write up your observations for one pair of synonyms in a coherent text
(ca. 1,000 words).

10. Write an essay (ca. 600 words) in response to the following quote from John
Sinclair (2008, p. 409):
[H]owever we circumscribe the unit of meaning, there will be connections
like tentacles stretching out to the surrounding cotext, supporting
or modifying the selection. Then, we have to concede that the normal
primary carrier of meaning is the phrase and not the word (…).

50  C H A P T E R 3 . CO L LO C AT I O N S
CHAPTER 4.
LEXICAL BUNDLES, N-GRAMS,
RECURRENT SEQUENCES

The two types of phrases discussed so far, i. e. idioms and collocations, share a
common feature—their constituents are related through meaning. In the case of
idioms, we may observe a rather unexpected semantic shift of the primary senses
of the words which produces non-compositional phrases. Collocations, in contrast,
are often semantically transparent even though they involve a certain restriction
in meaning which explains why the constituent words cannot be freely substituted
with synonymous words. The present chapter introduces an inherently different
approach to phrases which, on the surface of it, has nothing to do with word
meanings. Moreover, lexical bundles, to call them by one of the terms used to refer to
the new type of phrase, are automatically extracted from a corpus on purely formal
criterion of recurrence at a certain frequency in identical form consisting of n words.
Therefore they are often structurally and semantically incomplete, for example, it
was a, in order to, in the case of, you know and similar. Only at a later stage, lexical
bundles are investigated in terms of structure and functions in discourse. In this
respect, they are totally different from idioms and collocations. But let us take care
of first things first.

4.1 Definitions

In linguistic literature, the term lexical bundles and synonymously used terms
n-grams, clusters, chunks, recurrent sequences usually refer to the same type of
phrase. Table 4.1 below contains a selection of definitions taken from different
sources.
Different terms for automatically generated sequences of words often reflect the
tradition that scholars belong to, but essentially they denote a sequence or a string
of words, both lexical and functional, which appears in a corpus in the identical
form at a certain frequency. In the following, I will stick to the term lexical bundles

CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES  51

Table 4.1. Definitions of lexical bundles and related terms
Term Definition
Lexical bundle “Lexical bundles can be regarded as extended collocations:
bundles of words that show a statistical tendency to co-occur.”
(Biber et al., 1999, p. 989)
“Lexical bundles are recurrent expressions, regardless of their
idiomaticity, and regardless of their structural status.” (Biber
et al., 1999, p. 990)
“Lexical bundles are recurrent expressions, regardless of
their idiomaticity, and regardless of their structural status.
That is, lexical bundles are simply sequences of word forms
that commonly go together in natural discourse.” (Biber et al.,
1999, p. 991)
Recurrent “any continuous string of words occurring more than once in
sequences; identical form” (Altenberg, 1998, p. 101)
recurrent word “sequences of word forms of length n which recur in identical
combinations form with frequency greater than m from a corpus using spe-
cialised software” (De Cock, 2004, p. 228)
N-grams “repeated units of four words” (Forchini and Murphy, 2008)
“combinations of n words” (Römer, 2009b, p. 91)
Chunks “(…) recurrent strings of words, delimited by establishing fre-
quency cut-off points, for example, that a string must occur
at least 10 times per million words of text (…) and must be
distributed over a number of different texts, to qualify as a
bundle.” (O’Keeffe et al., 2007, p. 61)
“items in the automatically extracted strings which display
pragmatic integrity and meaningfulness regardless of their
syntax or lack of semantic wholeness” (O’Keeffe et al., 2007,
p. 64)

which is the one introduced by the authors of The Longman Grammar of Spoken
and Written English (Biber et al., 1999) where this type of phrase was extensively
described for the first time in linguistics (Cortes, 2015, p. 203).
To better understand how lexical bundles are identified, let us take a closer look
at the nature of these units. Imagine that we want to extract 4-word lexical bundles
from a particular corpus which is uploaded on a computer. To generate lexical
bundles, one has to use a corpus analysis program, the most popular in present-
day research being WordSmith Tools (Scott, 2005) or AntConC (Anthony, 2015).
There is also an online tool made freely available by Tom Cobb (http://lextutor.
ca/n_gram/). These programs have a special function to generate sequences of

52  CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES

recurrent words termed clusters in WordSmith Tools, n-grams in AntConC and on
the Lextutor site. When performing the function, the programs break up the entire
corpus into sequences of n words. The length of lexical bundles depends on the
aims of the study. In most studies of English, however, scholars prefer to analyse
4-word lexical bundles as they have been found to be more salient sequences than
the shorter ones. For example, the sentence There are several limitations associated
with this research (COCA) would be broken up into the following 4-word sequences:
there are several limitations
are several limitations associated
several limitations associated with
limitations associated with this
associated with this research
Usually the programs recognise sentence breaks by default, and the procedure
is repeated anew for each punctuated sentence in the corpus. As a result, in a few
seconds we get an output file in which all sequences of four words that are repeated
at least twice in the uploaded corpus are listed in the order of frequency. The list
may be easily exported into the required format and saved, for instance, as an Excel
file, for further analysis. A sample output list may look like the one given in Figure
4.1. What is instantly obvious is the fact that many lexical bundles are structurally
incomplete units (there are a lot, of the most important). The other peculiarity
of phrases of this type is their semantic transparency. Most of them are fully
compositional; yet there are also such discourse markers as on the other hand and
at the same time which are semi-compositional. Obviously, if any idiom is repeated
in a particular corpus more than once, it can make its way into the list of lexical
bundles extracted from that corpus.
So in order to obtain a list of lexical bundles, a researcher needs to make several
important decisions and formulate the operational definition of lexical bundles for
any study. The first aspect to consider, as already explained above, is the length of the
items. Many studies based on the English language deal with 3- and 4-word lexical
bundles, but the decision is directly related to the aims of the research. The criterion
of salience in the choice of the length was already mentioned above. In practice, it
means that the shorter bundles, for instance, 2-word, are usually incorporated in
the longer ones so it is reasonable to focus on a unit that is not to be split up into
shorter sequences. The shorter lexical bundles are naturally much more frequent
than the longer ones and manual revision of the automatically generated list
might lead to a situation where the researcher doesn’t see the wood for the trees,

CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES  53

 Frequency Dispersion
1 on the other hand 79 71
2 is one of the 51 43
3 one of the most 42 41
4 at the same time 41 32
5 all over the world 26 21
6 there are a lot 25 21
7 are a lot of 24 20
8 it is clear that 21 19
9 of the most important 19 18
10 it is possible to 19 15
11 first of all the 17 17
12 the most important thing 17 15
13 the fact that the 16 16
14 there is no need 16 15
15 is considered to be 16 13
16 will be able to 16 13
17 a lot of people 15 15
18 it is important to 15 15
19 that there is no 15 15
20 it is obvious that 15 14

FIGURE 4.1 Twenty most frequent 4-word lexical bundles in LICLE

which is another reason why the longer lexical bundles are often preferable. More
importantly, 2-word lexical bundles are less interesting for the qualitative analysis
(structural and/or functional) because it is not so easy to generalize about the
functions of such items as in the, on the, can be etc. Yet the short lexical bundles may
sometimes be handy in order to filter out from a corpus some specific items, e.g.
complex prepositions (as to, as regards, according to, out of), complex conjunctions
(as if, just because) and similar. Whatever the length, its choice is one of the first
decisions to be made.
The second important decision is related to the frequency threshold, or the
minimum cut-off point. In practice, it means that linguists shorten the raw list of
lexical bundles to those that have the greatest currency in the corpus and are the
most frequent. It is not unusual to see different cut-off points applied to lexical
bundles of different lengths. For instance, Biber et al. used the minimal frequency
of at least ten times per million words for four-word lexical bundles (1999, p. 990)

54  CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES

and five times per million words for five- and six-word lexical bundles (ibid., p. 992).
In studies where the approach to data selection is more stringent, the minimal cut-
off point may be set at forty times per million words, which is often the case in
studies based on corpora that are smaller in size and that, as research shows, yield
many more lexical bundles than large corpora (Cortes, 2015, p. 205). This decision
is also linked to the scope of a study—a sample adequate for a term paper might be
smaller than the one chosen for a more extensive study.
Finally, the last aspect to consider is the dispersion of a lexical bundle across
different texts that make up a corpus. A lexical bundle is only considered to be
such if it is spread across different texts rather than occurs in one text. This way
linguists ensure that whatever they analyse in the corpora is not idiosyncratic,
i. e. characteristic of only one text or person who wrote that text. This decision is
fairly flexible and is freely modified by linguists, usually in combination with the
frequency criterion when delimiting the list of automatically generated lexical
bundles. An interesting approach is described in Biber (2006, pp. 148–149) where
the researcher relaxed the parameters for retrieval of data from certain subcorpora
in order to avoid artificially inflated frequencies and obtain a reasonable sample of
data. While it is difficult to formulate a rule of thumb for dispersion, common sense
usually helps: if a lexical bundle occurs in two texts in a corpus that consists of 500
texts, any generalizations about currency of that bundle in the corpus would hardly
be valid. If it occurs in more than half of the texts, it certainly has a different status
in that corpus. Any dispersion in-between the two extremes, coupled with the
criterion of absolute frequency, usually leads to a more or less reasonable sample.
A lexical bundle is thus a phraseological unit defined on the basis of the following
parameters: 1) length, usually set between two and six words; 2) normalised
frequency per million (or any other number of) words; 3) dispersion: the minimal
number of texts in which it should occur to be included in the sample. While some
linguists do not fully define these criteria when they present the definition of lexical
bundles (see Table 4.1), these parameters are inevitably explained very accurately
when discussing research methods and operationalising the general definition of
lexical bundles.

4.2 Qualitative analyses of lexical bundles

Qualitative analysis of lexical bundles was first introduced in Biber et al. (1999)
where the major structural types and their distribution across two registers of
English (spoken conversation and written academic) were overviewed. A refined

CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES  55

classification of lexical bundles into structural and, for the first time explicitly,
functional types was proposed in Biber et al. (2004) and further revised, especially
the functional subtypes in Biber (2006). The structural classification is based on the
formal word class features and the functional classification takes into account the
functions of lexical bundles in discourse. While the two bases of classification cause
little argument, the ways of delimiting categories, as it will be demonstrated below,
leave much space for doubt and ongoing debate among linguists. The following
presentation of the two classifications largely draws on Biber et al. (2004).
Structurally lexical bundles fall into three major types:

I. Lexical bundles that incorporate fragments of verb phrases;

II. Lexical bundles that incorporate fragments of dependent clauses;
III. Lexical bundles that incorporate fragments of noun phrases and prepositional
phrases.

In addition, each type is subdivided further into several subtypes (Biber et al., 2004,
p. 381). The analysis of different registers of English shows that while lexical bundles
of type I are more frequent in spoken language, nominal bundles (type III) have
more prominence in the written registers of English (Biber, 2006). Applications of
the structural classification do not cause any disagreement among linguists and
seem to be generally accepted. More difficulty, however, arises from the application
of the functional classification.
The functional classification proposed by Biber et al. (2004) and revised in Biber
(2006) reflects the three metafunctions of language described in the systemic-
functional theory of grammar that was formulated by the famous British linguist
Michael Halliday. The three metafunctions are ideational (we use language to refer
to real and imagined world), interpersonal (we use language to communicate
stance) and textual (we use language to create coherent texts in speech and writing).
Accordingly, all lexical bundles fall into one of three functional types associated
with the metafunctions of language, namely,

I. Referential lexical bundles, including expressions of imprecision, quantity

specification, multifunctional reference;
II. Stance lexical bundles, including stance, attitudinal, modality, imperatives
etc. bundles;
III. Discourse-organising lexical bundles, including lexical bundles of topic
introduction/focus, topic elaboration/clarification, identification/ focus.

56  CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES

In Biber (2006), the fourth type, i. e. special functions lexical bundles referring to
politeness and inquiries, was added to account for certain bundles established in
spoken university registers.
The broad functional types are subdivided into several subtypes which may
differ from one variety of language to another. The reason is that a particular
function and, consequently, lexical bundles which express it, are well-attested
in one register but may be absent in the other registers of English. For instance,
referential bundles expressing imprecision (and something like that, and stuff like
that, or something) are typical for speech but hardly ever occur in written academic
English. Similarly, lexical bundles in accordance with the or with regard to may
be characteristic expressions of written language but quite rare in conversation.
Therefore, an analysis of the functions of lexical bundles usually begins with the
examination of concordances of individual bundles and thus involves a considerable
amount of text analysis and interpretation. Unavoidably, it leaves space for divergent
interpretations and different readings.
To give an example, let us consider the function of what is a typical lexical
bundle in any corpus of English, namely, one of the most. Structurally, this bundle is
NP-based as it consists of a noun phrase with of-phrase fragment (cf. Biber et al.,
2004, p. 381). In terms of its function in discourse, however, this lexical bundle may
be attributed to different categories. Biber et al. (2004) categorized it as a referential
bundle performing the function of identification/focus in most university registers,
excluding textbook language. In Biber (2006, p. 159) it was classified as a discourse-
organising bundle used for identification/focus in written university registers,
but on p. 166, it appears among referential bundles performing the identification
function in textbook English. The point is that it is not always possible to give one
interpretation of an instance of use. The following three examples from the BNC
illustrate challenges in the functional analysis:
(8) Bedford Park is regarded as the first Garden Suburb in England, and thus
has an international importance, while Shaw is one of the most important
architects of the Victorian period, best known as the designer of the old New
Scotland Yard building on the London Embankment.
(9) Dictionaries describe a monograph as an account of a single subject; by this
definition monographs make up one of the most common categories of art
publishing.
(10) All species will do well, but C luciliae is one of the most widely available and
dependable (…).

CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES  57

Apparently, the function of one of the most should be related to the adjective
immediately following the bundle and the broader context. In all cases, it is related
to identification yet what this identification is derived from is debatable. While in
(8) it is easy to link it to the author’s personal preferences and evaluation which
is signalled by important, which would mean that we have a stance lexical bundle,
in (9) and (10) the adjectives common and widely available imply focus on one
object out of many known. Is it a focus in order to refer to the object or in order to
narrow the following discussion? In the first case, we would categorize (9) and (10)
as referential expressions; in the second, they could be interpreted as discourse
organisers. It is therefore important to remember that a large part of interpretation
remains necessarily subjective; it is usually a slow process with many returns and
re-examinations.
An alternative functional classification of lexical bundles was proposed by
Hyland (2008a; 2008b) who investigated Master’s theses and dissertations written
in English. Hyland identified the following functional types of lexical bundles:
I. Research-oriented, including lexical bundles referring to location, procedure,
quantification, description, topic;
II. Text-oriented, including transition signals, resultative signals etc.;
III. Participant-oriented, including lexical bundles of stance and engagement.
Irrespectively of terminology used, Hyland’s approach largely overlaps with the one
proposed by Biber and his colleagues. The fact that one classification was published
earlier than the other perhaps explains why Biber et al. (2004) is a much wider
referenced source in literature than Hyland’s works.
A new development in the analysis of lexical bundles involves research into how
individual lexical bundles correlate with specific communicative purposes and
rhetorical moves characteristic of academic prose (Cortes, 2015, pp. 210–212; Cortes,
2013). The focus in such studies is on lexical bundles used in different sections of
academic research articles representing the four moves proposed by Swales (1990),
namely, presenting the research topic, overviewing previous research, identifying
a gap and introducing the current project. Hence, the qualitative analysis of lexical
bundles may be based on their use to express one or the other rhetorical move of
an academic text.
The structural and functional classifications of lexical bundles have been used
to analyse academic English used across different language varieties (Biber and
Barbieri, 2007) and disciplines (Cortes, 2004). The lexical bundle approach, as
it is increasingly known among linguists, is also used to investigate non-native

58  CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES

varieties of English (Ädel and Erman, 2012; Chen and Baker, 2010; De Cock, 2004;
Juknevičienė, 2009; 2013; Nekrasova, 2009; Paquot, 2013). Moreover, research into
lexical bundles is also concerned with contrastive studies of different languages
(Granger, 2014) and cross-linguistic (translation) research (Shrefler, 2010). Most
probably, the relative ease of extraction of lexical bundles from a corpus continues
to generate a considerable interest in the lexical bundles approach both among
novice and experienced researchers. The design of a study of lexical bundles,
however, requires a careful consideration of a number of aspects in order to avoid
possible pitfalls in the analysis of those seemingly superficial phraseological units.

4.3 Methodological considerations

One may wonder whether a purely formal or technical approach to corpus data
is at all valid. Apparently, it is. Moreover, it is even seen as advantageous:
The ‘recurrent word combination’ method is an illustration of corpus linguistics
methodology at its most heuristic, i.e. as a raw discovery procedure. The method
does not presuppose any linguistic category or pre-established list of sequences.
(De Cock, 2004)
Hence, it provides the researcher with objectively obtained raw data and, obviously,
is a non-biased starting point. Yet it poses a number of methodological challenges in
order to be fully appreciated.
The first important consideration at the initial stage of any study on lexical bundles
is the overall representativeness of the corpus used, which is perhaps universal in
any corpus-based and corpus-driven research. If a corpus is not representative of
general language use, for instance, it is too small and biased towards one specific
language variety, social group etc., it will yield highly specialised data. So caution
should be taken to obtain an adequate set of lexical bundles and avoid hasty
conclusions when formulating general statements about such data and contrasting
it with material extracted from a different corpus. Frequent lexical bundles
extracted from a corpus of research papers will most probably be different from
those extracted from a corpus of newspaper articles, and it is the responsibility of
the researcher to consider very carefully whether the data obtained lends itself to
any contrastive comparisons or sweeping generalisations. Obviously, the smaller
the corpus, the less generalizable are research findings.
The automatically generated list of lexical bundles is then submitted to
qualitative (structural or functional) analysis. Yet prior to any qualitative analyses, it

CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES  59

is often necessary to carry out a manual revision of the list and weed out all kinds of
irrelevant material. Sometimes the programs fail to recognize certain punctuation
marks and treat words with the apostrophe (don’t or isn’t) as two words, namely,
don t or isn t are counted as two words, which naturally leaves inaccuracies that
have to be manually treated by the researcher. The other frequent issue has to do
with capital letters which might yield two different tokens: there are many and
There are many. Again, it is the researcher’s responsibility to check if capitalisation
could be disregarded and bundles with different capitalisation merged into one
item with their frequencies reported as a sum. Lastly, a similar formal issue has to
do with mistyped words all of which, if repeated, produce irrelevant items. Such
technicalities have to be dealt with before any other steps of analysis are taken.
Another aspect deserving attention is lexical bundles that are made up of words
from (book) titles, proper names, recurring quoted words and the like. In studies of
lexical bundles such items are called topic-related or topic-specific lexical bundles.
For instance, if one is analysing a corpus of film or book reviews, it is inevitable
that the titles of films or books, if not eliminated when compiling the corpus or pre-
treated as unanalysable material, will appear in the list of recurrent lexical bundles.
It might be necessary to remove such topic-related bundles from the sample in
which case it is done manually by checking the whole list and deleting the irrelevant
material. In specific studies, however, the proportion of such ‘quoted’ words might
be useful to consider as it provides evidence on the extent to which production
of language involves reliance of writing input or quotes. But topic-related lexical
bundles present an issue a researcher needs to take into account.
It is also important to remember that corpus linguistics requires a cautious
treatment of frequency statistics which in this case refers to the frequencies of
lexical bundles. In order to be able to compare frequencies of one particular lexical
bundle in two corpora of different sizes, one should first normalise the absolute
frequencies so that the comparison makes sense. Furthermore, some scholars
include a normalised frequency parameter in the operational definition of lexical
bundles which again requires that one has a clear understanding of what is involved
in the normalisation counts. Let us consider the frequency data of a lot of in the BNC
(Figure 4.2). It is difficult to make any claims about the currency of that sequence
in the spoken and written subcorpora simply because the two subcorpora are
different in size.
To be able to make a straightforward comparison, it is necessary to re-calculate
the absolute number of hits per certain number of words. In this case, when the

60  CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES

 Figure 4.2 Distribution of a lot of in the BNC

size of the corpus reaches millions of words, it is convenient to use a sample of one
million words. So let us calculate the normalised frequency of a lot of in spoken and
written subcorpora of the BNC per one million words (sometimes abbreviated as
‘pmw’):
spoken: 5747*1000000/10409858=552.07 occurrences/pmw
written: 8885*1000000/87903571=101.08 occurrences/pmw
The sample for normalisation need not be one million; its size should be either
meaningful or close in number to corpora under study or set to such value that
would make calculations reasonably convenient, for instance, per 1,000 or 100, 000
words. Once the normalised frequencies are available, it is possible to compare the
use of items in different corpora. In the case of a lot of, one might conclude that this
phrase is 5.5 times more frequent in speech rather than in written language, which,
obviously, explains why it is sometimes considered inappropriate in academic
essays.

Further reading

 Lexical bundles in different registers of English: Biber (2006); Biber & Barbieri
(2007);
 Lexical bundles in different disciplines: Cortes (2004); Forchini & Murphy
(2008); Hyland (2008b);
 Lexical bundles in native and non-native language varieties: Ädel & Erman
(2012); Chen & Baker (2010); De Cock (2004); Juknevičienė (2009; 2013);
Nekrasova (2009);
 Applications of the lexical bundle approach to contrastive research: Granger
(2014);
 Lexical bundles for the teaching of academic English: Martinez & Schmitt (2012).

CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES  61

Study tasks

1. Study the HELP window screenshot from the WordSmith tools (Scott, 2005).
How does the definition of cluster relate to definitions given in Table 4.1 and
other definitions of different types of phraseological units?

2. Study the description of the structural types of lexical bundles in Chapter 13.2
of the Longman Grammar of Spoken and Written English (Biber et al., 1999).
Write an essay (ca. 700 words) entitled ‘Distribution of structural types of
lexical bundles across two English registers: academic prose vs. conversation’.

3. Biber (2006) reports data on the distribution of lexical bundles in university

registers. Study his findings and consider the use of English in the classroom.
In your opinion, in terms of structural types of lexical bundles, is it closer to
conversational English or written academic prose?

4. Figure 4.1 gives a list of top twenty lexical bundles extracted from a corpus
of written English comprising student essays (LICLE corpus). Below (Figure
4.3) there is a list of top twenty lexical bundles from the LINDSEI-LT corpus
which represents EFL learner speech. Analyse lexical bundles in terms of their
structures and, in as much as it is possible, functions. Report the quantitative
findings in a chart and write up the analysis in a short report (ca. 500 words).

62  CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES

 LINDSEI
1 I would like to 130 47
2 I don’t know Ii 42 25
3 er I don’t know 24 19
4 and I don’t know 21 13
5 a lot of time 19 15
6 would like to visit 18 9
7 erm I don’t know 17 9
8 I don’t know what 17 16
9 I would love to 16 8
10 like to talk about 16 15
11 would like to talk 16 14
12 and I would like 15 12
13 I don’t know it’s 15 8
14 and it was very 14 6
15 but I would like 14 12
16 in the first year 14 11
17 and I think that 13 12
18 so I don’t know 13 13
19 at the same time 12 9
20 I think I would 12 10

Figure 4.3 Twenty most frequent 4-word lexical bundles in LINDSEI-LT.

5. If lexical bundles are recurrent word sequences in one language (English),

arguably, they should have more or less regular equivalents in other languages.
Using any parallel corpus of English and any other language, for example,
the online multilingual resource GLOSBE (https://glosbe.com/), analyse
translational equivalents of the following English lexical bundles in the language
of your choice:
at the rate of
in relation to
in the case of
in terms of
the amount of
the degree of
the number of

CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES  63

 to a certain extent
with regard to
Write an essay (ca. 700 words) in which you describe your findings and
observations about the translational equivalents of one chosen item.

6. Forensic stylometry is an area where n-grams have been found to be particularly

useful. Combined with several other measures, n-grams were used to uncover
the true identity of Robert Galbraith, the author of The Cuckoo’s Calling. Read the
account of this linguistic detective by Patrick Juola to find out the true identity
of the writer and explain how it was revealed by linguistic analyses (available
from http://languagelog.ldc.upenn.edu/nll/?p=5315).

64  CHAPTER 4. LEXICAL BUNDLES, N-GRAMS, RECURRENT SEQUENCES

POSTSCRIPT

This brief journey through the jungle of English phraseology was inevitably
limited. Owing to the scope of the course for which it was written (one semester) and
two-fold aims of the course, i. e. training both in phraseology and research writing, it
would be difficult to cover all types of phraseological units. Among such regrettable
omissions is the phrase-frame, a further development of the lexical bundle. This is
essentially a lexical bundle with one variable slot, for example, it is * to, as it could *,
is the * of. The variable slot may be realised by a number of different lexemes while
in its basic form a phrase-frame represents a recurrent lexical pattern. Phrase-
frames of English extracted from the BNC are available on a website developed by
Fletcher (n.d.), which the reader is encouraged to explore. Research into phrase-
frames is gaining pace across different linguistic areas: specialised discourses
(Fuster-Marquez, 2014; Grabowski, 2015); different registers of English (Gray and
Biber, 2013); analysis of novice and proficient academic writing and learner English
(Garner, 2016; Juknevičienė & Grabowski, n.d.). The phrase-frame approach reveals
a construction-driven picture of language and more than any other phraseological
unit links lexis and grammar.
The ongoing research into the nature of phraseological units and corpus research
continues to provide evidence that language consists of chunks of words, whatever
the term chosen for the chunks, be it collocations, lexical bundles or, more generally,
formulaic expressions. Research in psycholinguistics demonstrates that chunkiness
of language eases its processing and thus deserves a place not only in the applied
avenues of language such as teaching, learning, translation etc. but also in the
description of the language system. It is thus hardly surprising that phraseology as
a branch of linguistics is gradually developing into a full-fledged research field. In
his Preface to a volume of research papers on phraseology, Ellis aptly described the
role of the field in the world of linguistics:
[P]hraseology pervades theoretical, empirical, and applied linguistics. Like blood
in systemic circulation, it flows through heart and periphery, nourishing all.
(2008, p. 9)

POSTSCRIPT  65
 As an interdisciplinary research field, phraseology attracts scholars from
a variety of research strands and keeps growing. The fact that it does have a
professional association, regular publications and conferences certainly testifies to
a rise of its status. The European Society of Phraseology founded in 1999 and having
its seat in Zurich arranges international conferences and publishes the Yearbook of
Phraseology, all of which points to increasing institutionalisation of phraseology.
Alongside, smaller research communities, for example, Formulaic Language
Research Network, FLaRN, co-ordinated by Alison Wray (University of Cardiff),
continue to provide discussion fora to both novice and seasoned researchers.
Information available on their respective websites might be interesting to anyone
seeking inspiration and recent updates on the latest developments.

66  POSTSCRIPT
REFERENCES

I. Corpora
BNC – British National Corpus. Available at: http://bncweb.lancs.ac.uk or http://
www.natcorp.ox.ac.uk/.
COCA – The Corpus of Contemporary American English: 520 million words,
1990-present. Compiled by Mark Davies. Available at: http://corpus.byu.edu/
coca/.
LICLE – Lithuanian component of the International Corpus of Leaner English
(Grigaliūnienė et al., 2008).
LINDSEI-LT – Lithuanian component of the Louvain International Database of
Spoken English Interlanguage (Grigaliūnienė and Juknevičienė, 2011).
LOCNESS – Louvain Corpus of Native English Essays, compiled at the Centre for
English Corpus Linguistics, University of Louvain-la-Neuve, Belgium. 1998.
Online description at: http://www.learnercorpusassociation.org/.

II. Dictionaries
Benson, M., Benson, E., & Ilson, R. (Eds). 1993 [1986]. The BBI Combinatory
Dictionary of English. Amsterdam/Philadelphia: John Benjamins Publishing
Company.
Collins COBUILD English Dictionary for Advanced Learners. 2001. Third edition.
Edited by J. Sinclair. Glasgow: HarperCollins Publishers.
Oxford Collocations Dictionary for Students of English. 2002. Oxford: Oxford
University Press.
Oxford English Dictionary Online. 2017. Oxford: Oxford University Press. Available
at: www.oed.com, accessed on 2017-01-15.

REFERENCES  67
III. Software
Anthony, L. (2015). AntConc (version 3.4.4). Available at: http://www.
laurenceanthony.net/software/antconc/, accessed 2017-02-07.
Cobb, T. Compleat Lexical Tutor. Available at: http://www.lextutor.ca/, accessed
2017-01-10.
Scott, M. (2005). WordSmith Tools (Version 5). Oxford: Oxford University Press.

IV. Literature
Ädel, A. & Erman, B. 2012. Recurrent word combinations in academic writing by
native and non-native speakers of English: A lexical bundles approach. English
for Specific Purposes 31, 81–92.
Altenberg, B. 1998. On the Phraseology of Spoken English: The Evidence of Recurrent
Word-Combinations, in: Cowie, A.P. (Ed.), Phraseology: Theory, Analysis, and
Applications. Oxford: Clarendon Press, pp. 101–122.
Biber, D. 2006. University Language: A Corpus-based Study of Spoken and Written
Registers. Amsterdam/Philadelphia: John Benjamins Publishing.
Biber, D. 2009. A corpus-driven approach to formulaic language in English: Multi-
word patterns in speech and writing. International Journal of Corpus Linguistics
14/3, 275-311.
Biber, D. & Barbieri, F. 2007. Lexical bundles in university spoken and written
registers. English for Specific Purposes 26, 263–286.
Biber, D., Conrad, S. & Cortes, V. 2004. ‘If you look at …’: Lexical Bundles in University
Teaching and Textbooks. Applied Linguistics 25, 371–405.
Biber, D., Johansson, S., Leech, G., Conrad, S. & Finegan, E. 1999. The Longman
Grammar of Spoken and Written English. Harlow: Longman.
Bouma, G. 2009. Normalized (pointwise) mutual information in collocation
extraction, in: Proceedings of the Biennial GSCL Conference. Available at: https://
svn.spraakdata.gu.se/repos/gerlof/pub/www/Docs/npmi-pfd.pdf, accessed
2017-01-20.
Chen, Y.-H. & Baker, P. 2010. Lexical bundles in L1 and L2 academic writing. Language
Learning & Technology 14, 30–49.
Cortes, V. 2004. Lexical bundles in published and student writing in history and
biology. English for Specific Purposes 23, 397–423.

68  REFERENCES
Cortes, V. 2013. ‘The purpose of this study is to’: Connecting lexical bundles and
moves in research article introductions. Journal of English for Academic Purposes
12, 33–43.
Cortes, V. 2015. Situating lexical bundles in the formulaic language spectrum.
Origins and functional analysis developments, in: Cortes, V. & Csomay, E. (Eds)
Corpus-based Research in Applied Linguistics: Studies in Honor of Doug Biber.
Amsterdam: John Benjamins Publishing Company, pp. 197–216.
Cowie, A. P. 1981. The treatment of collocations and idioms in learners’ dictionaries.
Applied Linguistics 2, 223–235.
De Cock, S. 2004. Preferred sequences of words in NS and NNS speech. Belgian
Journal of English Language and Literatures (BELL) 2, 225–46.
Durrant, P. 2014. Corpus frequency and second language learners’ knowledge of
collocations: A meta-analysis. International Journal of Corpus Linguistics 19,
443–477.
Economist Books (Ed.). 2005. Style Guide, 9th ed. London: Profile Books.
Ellis, N.C. 2008. Phraseology: The periphery and the heart of language, in: Meunier, F.
& Granger, S. (Eds), Phraseology in Foreign Language Learning and Teaching.
Amsterdam & Philadelphia: John Benjamins Publishing Company, pp. 1–13.
Erman, B. & Warren, B. 2000. The idiom principle and the open choice principle.
Text-Interdisciplinary Journal for the Study of Discourse 20, 29–62.
Fernando, C. 1996. Idioms and Idiomaticity. Oxford: Oxford University Press.
Firth, J. R. 1957. Papers in Linguistics. 1934–1951. London: Oxford University Press.
Fletcher, W. (n.d.) Phrases in English. Online resource. Available at: http://
phrasesinenglish.org/, accessed 2017-01-20.
Forchini, P. & Murphy, A. 2008. N-grams in comparable specialized corpora:
Perspectives on phraseology, translation, and pedagogy. International Journal of
Corpus Linguistics 13, 351–367.
Fuster-Marquez, M. 2014. Lexical bundles and phrase-frames in the language of
hotel websites. English Text Construction 7, 84–121.
Garner, J. 2016. A phrase-frame approach to investigating phraseology in learner
writing across proficiency levels. International Journal of Learner Corpus
Research 2, 31–67.
Gibbs, R. W., Jr. 1994. The Poetics of Mind. Figurative Thought, Language, and
Understanding. Cambridge: Cambridge University Press.

REFERENCES  69
Gläser, R. 1998. The stylistic potential of phraseological units in the light of genre
analysis, in: Cowie, A. P. (Ed.), Phraseology: Theory, Analysis and Applications.
Oxford: Oxfors University Press, pp. 125–143.
Grabowski, Ł. 2015. Phrase-frames in English pharmaceutical discourse: a corpus-
driven study of intra-disciplinary register variation. Research in Language 3,
266–291.
Granger, S. 2014. A lexical bundle approach to comparing languages. Genre- and
register-related discourse features in contrast. Languages in Contrast 14:1,
58–72.
Granger, S. & Paquot, M. 2008. Disentagling the phraseological web, in: Granger, S.
& Meunier, F. (Eds), Phraseology: An Interdisciplinary Perspective. Amsterdam &
Philadelphia: John Benjamins Publishing Company, pp. 27–49.
Gray, B. & Biber, D. 2013. Lexical frames in academic prose and conversation.
International Journal of Corpus Linguistics 18, 109–135.
Gries, S.T. 2008. Phraseology and linguistic theory: A brief survey, in: Granger, S. &
Meunier, F. (Eds), Phraseology: An Interdisciplinary Perspective. Amsterdam &
Philadelphia: John Benjamins Publishing Company, pp. 2–26.
Gries, S. T. 2015. Quantitative designs and statistical techniques, in Biber, D. &
Reppen, R. (Eds), The Cambridge Handbook of English Corpus Linguistics.
Cambridge: Cambrigde University Press, pp. 50-72.
Grigaliūnienė, J., Bikelienė, L. & Juknevičienė, R. 2008. The Lithuanian component
of the International Corpus of Learner English (LICLE): a resource of English
language learning, teaching and research at Lithuanian institutions of higher
education. Žmogus ir žodis 10, 62–66.
Halliday, M. A. K. & Hasan, R. 1976. Cohesion in English. London: Pearson Education.
Grigaliūnienė, J. & Juknevičienė, R. 2011. Formulaic language, learner speech and
the spoken corpus of learner English LINDSEI-LITH. Žmogus ir žodis 13, 12–18.
Howarth, P. 1998. Phraseology and second language proficiency. Applied linguistics
19, 24–44.
Hyland, K. 2008a. Academic clusters: Text patterning in published and postgraduate
writing. International Journal of Applied Linguistics 18, 41–62.
Hyland, K. 2008b. As can be seen: Lexical bundles and disciplinary variation. English
for Specific Purposes 27, 4–21.

70  REFERENCES
Ishida, P. 2008. Contrastive idiom analysis: The case of Japanese and English idioms
of anger, in: Granger, S. & Meunier, F. (Eds), Phraseology: An Interdisciplinary
Perspective. Amsterdam & Philadelphia: John Benjamins Publishing Company,
pp. 275–291.
Juknevičienė, R. 2013. Recurrent word sequences in written learner English, in:
Šeškauskienė, I. & Grigaliūnienė, J. (Eds), Anglistics in Lithuania: Cross-Linguistic
and Cross-Cultural Aspects of Study. Newcastle upon Tyne: Cambridge Scholars
Publishing, pp. 178–197.
Juknevičienė, R. 2011. Lexical Bundles in Non-Native Speaker and Native Speaker
Written English. Summary of doctoral dissertation in English. Vilnius: Vilnius
University.
Juknevičienė, R. 2009. Lexical bundles in learner language: Lithuanian learners vs.
native speakers. Kalbotyra 61, 61–72.
Juknevičienė, R. & Grabowski, Ł. (n.d., submitted) Comparing formulaicity of learner
writing through phrase-frames: a corpus-driven study of Lithuanian and Polish
EFL student writing.
Kjellmer, G. 1991. A mint of phrases, in: Aijmer, K. & Altenberg, B. (Eds), English
Corpus Linguistics. Studies in Honour of Jan Svartvik. London: Longman, pp. 111–
127.
Levshina, N. 2015. How to do Linguistics with R. Data Exploration and Statistical
Analysis. Amsterdam & Philadelphia: John Benjamins Publishing Company.
Lewis, M. 2000. Teaching Collocation. Further Developments in the Lexical Approach.
London: Language Teaching Publications.
Lewis, M. 1993. The Lexical Approach. The State of ELT and the Way Forward. Hove:
Language Teaching Publications.
Marcinkevičienė, R. 2010. Lietuvių kalbos kolokacijos. Kaunas: Vytauto Didžiojo
universitetas.
Martinez, R. & Schmitt, N. 2012. A Phrasal Expressions List. Applied Linguistics 33,
299–320.
McEnery, T. & Wilson, A. 2001. Corpus Linguistics. 2nd ed. Edinburgh: Edinburgh
University Press.
Molina-Plaza, S. & de Gregorio-Godeo, E. 2010. Stretched verb collocations with
give: their use and translation into Spanish using the BNC and CREA corpora.
ReCALL 22, 191–211.

REFERENCES  71
Moon, R. 2008. Dictionaries and collocation, in: Granger, S. & Meunier, F. (Eds),
Phraseology: An Interdisciplinary Perspective. Amsterdam & Philadelphia: John
Benjamins Publishing Company, pp. 313–336.
Moon, R. 1998. Fixed Expressions and Idioms in English: a Corpus-Based Approach.
Oxford: Clarendon Press.
Naciscione, A. 2010. Stylistic Use of Phraseological Units in Discourse. Amsterdam &
Philadelphia: John Benjamins Publishing Company.
Nattinger, J. R. & DeCarrico, J. S. 1992. Lexical Phrases and Language Teaching.
Oxford: Oxford University Press.
Nekrasova, T. M. 2009. English L1 and L2 speakers’ knowledge of lexical bundles.
Language Learning 59, 647–686.
Nesselhauf, N. 2005. Collocations in a Learner Corpus. Amsterdam & Philadelphia:
John Benjamins Publishing Company.
O’Keeffe, A., McCarthy, M. & Carter, R. 2007. From Corpus to Classroom: Language
Use and Language Teaching. Cambridge: Cambridge University Press.
Paquot, M. 2013. Lexical bundles and L1 transfer effects. International Journal of
Corpus Linguistics 18, 391–417.
Pawley, A. & Syder, F. H. 1983. Two puzzles for linguistic theory: nativelike selection
and nativelike fluency, in: Richards, J.C. & Schmidt, R.W. (Eds), Language and
Communication. London & New York: Longman, pp. 191–226.
Piirainen, E. 2008. Phraseology in a European framework, in: Granger, S. & Meunier, F.
(Eds), Phraseology: An Interdisciplinary Perspective. Amsterdam & Philadelphia:
John Benjamins Publishing Company, pp. 234–258.
Römer, U. 2009a. The inseparability of lexis and grammar: Corpus linguistic
perspectives. Annual Review of Cognitive Linguistics 7, 140–162.
Römer, U. 2009b. English in academia: Does nativeness matter. Anglistik:
International Journal of English Studies 20, 89–100.
Shrefler, N. 2011. Lexical bundles and German bibles. Literary and Linguistic
Computing 26(1), 89–106.
Sinclair, J. 1991. Corpus, Concordance, Collocation. Oxford: Oxford University Press.
Sinclair, J. 1996. The search for units of meaning. Textus 9, 75–106.
Sinclair, J. 2004. Trust the Text. London & New York: Routledge / Taylor & Francis
Group.

72  REFERENCES
Sinclair, J. 2008. The phrase, the whole phrase and nothing but the phrase, in:
Granger, S. & Meunier, F. (Eds), Phraseology: An Interdisciplinary Perspective.
Amsterdam & Philadelphia: John Benjamins Publishing Company, pp. 407–410.
Stubbs, M. 2007. Notes on the history of corpus linguistics and empirical semantics,
in: Nenonen, M. & Niemi, S. (Eds), Collocations and Idioms. Joensuu: Joensuun
Yliopisto, pp. 317–329.
Stulpinaitė, M., Horbačauskienė, J. & Kasperavičienė, R. 2016. Issues in translation
of linguistic collocations. Kalbų studijos/Studies about Languages 29, 31–41.
Swales, J.M. 1990. Genre Analysis: English in Academic and Research Settings.
Cambridge & New York: Cambridge University Press.
Tognini-Bonelli, E. 2001. Corpus Linguistics at Work. Amsterdam & Philadelphia:
John Benjamins Publishing.
Vilkaitė, L. 2016. Formulaic language is not all the same: comparing the frequency
of idiomatic phrases, collocations, lexical bundles, and phrasal verbs. Taikomoji
kalbotyra 8, 28–54.
Wang, Y. 2016. The Idiom Principle and L1 Influence: A contrastive learner-corpus
study of delexical verb + noun collocations. Amsterdam & Philadelphia: John
Benjamins Publishing Company.
Watcyn-Jones, P. 2002. Test Your Idioms. Harlow: Pearson Education Limited.
Williams, G. C. 2008. The Good Lord and his works: a corpus-driven study of
collocational resonance, in: Granger, S. & Meunier, F. (Eds), Phraseology: An
Interdisciplinary Perspective. Amsterdam & Philadelphia: John Benjamins
Publishing Company, pp. 159–174.
Wray, A. 2002. Formulaic Language and the Lexicon. Cambridge & New York:
Cambridge University Press.
Wray, A. 2000. Formulaic sequences in second language teaching: principle and
practice. Applied Linguistics 21, 463–489.
Xiao, R. & McEnery, T. 2006. Collocation, semantic prosody, and near synonymy: a
cross-linguistic perspective. Applied Linguistics 27, 103–129.
Zareva, A. & Wolter, B. 2012. The “promise”of three methods of word association
analysis to L2 lexical research. Second Language Research 28, 41–67.

REFERENCES  73
 Rita Juknevičienė

ENGLISH PHRASEOLOGY AND CORPORA

An introduction to corpus-based
and corpus-driven phraseology

ISBN 978-609-459-820-3
4,00 aut. l.
Išleido Vilniaus universitetas,
Vilniaus universiteto leidykla
Universiteto g. 3, LT-01513 Vilnius