Applied Linguistics 22/2: 173±195 # Oxford University Press 2001

The Grammatical and Lexical Patterning

of MAKE in Native and Non-native
Student Writing
1
BENGT ALTENBERG AND 2SYLVIANE GRANGER
1
Lund University, Sweden,2Universite Catholique de Louvain, Belgium

This article investigates EFL learner use of high frequency verbs, and in
particular use of the verb MAKE, a major representative of this group. The main
questions addressed are: do learners tend to over- or underuse these verbs? Are
high-frequency verbs error-prone or safe? What part does transfer play in
misuse of these verbs? To answer these questions, authentic learner data has
been compared with native speaker data using computerized corpora and
linguistic software tools to speed up the initial stage of the linguistic analysis.
The article focuses on what proves to be the two most distinctive uses of MAKE,
viz. the delexical and causative uses. Results show that EFL learners, even at an
advanced pro®ciency level, have great diculty with a high frequency verb
such as MAKE. They also demonstrate that some of these problems are shared by
the two groups of learners under consideration (Swedish- and French-speaking
learners) while others seem to be L1-related. In the conclusion, the pedagogical
implications of the study are discussed and suggestions made for using
concordance-based exercises as a way of raising learners' awareness of the
complexity of high-frequency verbs.

By focusing on the grammatical and lexical patterning of the verb MAKE in a

computerized corpus of EFL writing, we aim to throw some light on the use of
high-frequency verbs by EFL learners. As the analysis is computer-aided, we
also aim to assess the usefulness of linguistic software tools for this type of
study.

1. HIGH-FREQUENCY VERBS IN EFL

All languages seem to have some basic verbs that are used again and again in
discourse and consequently turn up early in frequency lists. In English, for
example, we are likely to ®nd the following ®fteen verbs (lexemes) topping
any corpus-based list of high-frequency verbs (disregarding BE and modal
auxiliaries), although their rank order may vary according to medium and
text type (see Svartvik and Ekedahl 1995):

HAVE DO KNOW THINK GET

GO SAY SEE COME MAKE
TAKE LOOK GIVE FIND USE
174 GRAMMATICAL & LEXICAL PATTERNING OF MAKE IN STUDENT WRITING

High-frequency verbs have several characteristics that make them especially

interesting in a cross-linguistic perspective (cf. Viberg 1996):
. they express basic meanings and tend to dominate di€erent semantic ®elds;
. they have high-frequency equivalents in most languages;
. they are characterized by a high degree of polysemy, caused by two kinds of
meaning extension:
± one universal tendency creating more general, abstract, delexicalized or
grammaticalized uses,
± various language-speci®c tendencies resulting in specialized meanings,
collocations, and idiomatic uses;
. they tend to be problematic for foreign language learners.
The last point is undoubtedly a result of the other characteristics. Despite
fundamental semantic similarities across languages, high-frequency verbs
have developed many language-speci®c di€erences which make them
treacherous for foreign language learners.1 The literature on the use of
high-frequency verbs by EFL learners contains two seemingly contradictory
observations. Several studies point to an overuse of these verbs by learners.
Investigating Norwegian EFL learners Hasselgren (1994) notes that core words
are hugely overused by learners who tend to cling on to them like `lexical
teddy bears'. She accounts for this overuse in the following way: `core
wordsÐlearnt early, widely usable, and above all safe (because they do not
show up as errors) are hugely overused, even among learners suciently
advanced to have been weaned o€ them' (1994: 250).
Similar ®ndings have been made by KaÈllkvist (1999: 130±34) for Swedish
learners. The tendency towards overuse may be reinforced where the
learner's mother tongue is a Germanic one, like Norwegian or Swedish,
because of L1 transfer. However, the fact that French-speaking learners also
overuse high-frequency verbs (see Granger 1996) would seem to indicate that
the phenomenon is not entirely L1-related.
However, there also seems to be a tendency pulling in the other direction,
i.e. towards underuse. High-frequency verbs, such as MAKE, TAKE, GIVE, PUT, etc.
are often used as delexical verbs, i.e. they are used `with nouns as their object
to indicate simply that someone performs an action, not that someone a€ects
or creates something. These verbs have very little meaning when they are
used in this way' (Collins Cobuild English Grammar 1990: 147). According to
Sinclair (1991: 79), `many learners avoid the common verbs as much as
possible, and especially where they make up idiomatic phrases. Instead of
using them, they rely on larger, rarer, and clumsier words which make their
language sound stilted and awkward'.
Sinclair does not back up his statement with corpus data, neither does he
give any explanation for this avoidance. One plausible explanation might be
the learners' awareness of the diculty of these verbs. As pointed out by
Allerton (1984: 33), the choice of these verbs is mostly arbitrary: the choice of
take in take a step (rather than make, for instance) is semantically unmotivated
 BENGT ALTENBERG AND SYLVIANE GRANGER 175

and therefore likely to be largely language-speci®c (compare English take/

*make a step and French *prendre/faire un pas).
The error-proneness of these verbs in learner writing and speech has been
highlighted in several recent EFL studies (Lindner 1994; Chi et al. 1994;
Kanatani et al. 1995; KaÈllkvist 1999; Lennon 1996; Howarth 1996). The wide
range of mother tongue backgrounds involved in these studies (German,
Chinese, Japanese, Swedish) shows that the phenomenon is a universal one,
although the error types themselves seem to be partly L1-speci®c.
In this paper we aim to throw some light on EFL use of high-frequency
verbs by focusing on one major representative of this group of verbs, the verb
MAKE. The main questions we will investigate are: do learners tend to over- or
underuse these verbs? What categories of meaning/use di€erentiate learner
from native use? Are high-frequency verbs error-prone or `safe' as Hasselgren
suggests? What part does transfer play in misuse of these verbs? To answer
these questions, we will compare authentic learner data with native speaker
data using computerized corpora to speed up the initial stage of the linguistic
analysis.

2. DATA AND METHODOLOGY

The computer learner corpus used for the study is made up of two corpus
samples from the International Corpus of Learner English (ICLE) database. The
®rst contains c. 170,000 words of essay writing by advanced French-speaking
learners of English (henceforth FR). The second, which is comparable to the
®rst in both size and text type, contains essays written by Swedish learners
(henceforth SW). The learners are all 2nd or 3rd year university students of
English. By selecting learner corpora matching our own mother tongue
backgrounds, we hope to be in a better position to interpret learner errors and
assess the importance of L1 transfer.
To compare EFL use with native English use, it is necessary to have a native
speaker control corpus. For this purpose we used a 170,000 word sample from
the Louvain Corpus of Native English Essays (LOCNESS) which contains
argumentative essays written by native-speaker American students and is
therefore fully comparable to the learner corpus. Table 1 gives the exact size of
the corpora used.2
The learner essays, which are approximately 600 words long, are
argumentative and non-technical in character and cover a variety of topics.
The learners are EFL (not ESL) learners and their pro®ciency level is
Table 1: Learner and native-speaker corpora
SW FR LOCNESS

Number of words 169,608 169,19 168,325

Number of essays 296 285 207
176 GRAMMATICAL & LEXICAL PATTERNING OF MAKE IN STUDENT WRITING

advanced. The native-speaker essays are comparable to the learner essays in

terms of text type and range of topics but are slightly longer (about 800
words). For more information on the ICLE, see Granger (1998).
One of the major advantages of using computer corpora is that the data can
be submitted to text handling software tools, thereby making it possible to
automate part of the linguistic analysis. For this study, we have opted for
WordSmith Tools,3 a user-friendly and powerful package containing several
analytical tools which are useful for phraseological studies. Among those we
have mainly made use of are the lemmatizer, the concordancer, and the
collocation display (see below). The methodology we have used is a
combination of fully automatic analysis and minute manual investigation.

3. OVERALL FREQUENCY OF MAKE

As a ®rst step in our analysis, we wanted to check whether EFL learners had
a tendency to over- or underuse the verb MAKE in comparison with the
native-speaker American students. To compute the frequencies in the three
corporaÐSW, FR, and LOCNESSÐwe used WordSmith Tools' lemmatizing
facility, which enabled us to group all the in¯ectional forms of the lemma
MAKEÐmake, makes, making, made. The advantage of using this facility is that
it is then possible to create a concordance of the lemma MAKE rather than
having to create concordances for each verbal form. The next step consisted
in scrutinizing each concordance line to weed out irrelevant instances (such
as compound usesÐa make-believe problem, hand-made pasta). This done, we
were able to compute normalized frequencies (occurrences per 100,000
words) of the verbal lemma MAKE in the three corpora.4 The results are given
in Table 2.

Table 2: Frequency of MAKE in NNS and NS student writing

Verb SW FR LOCNESS

MAKE 354.3 234.6* 339.8

The table brings out a clear di€erence between the French-speaking and the
Swedish learners. While the former underuse the verb MAKE, the latter use it a
little more than the native-speaker students, though not signi®cantly so. The
term `signi®cantly' here and in the rest of the article is used in its statistical
sense. All frequency di€erences across the samples were tested by means of
the chi-square test, with 1 per cent as the critical level of statistical signi®cance
(p < 0.01). Statistically signi®cant di€erences between each learner group and
the native speaker control corpus (LOCNESS) are marked by an asterisk in the
tables.5 Detailed chi-square values are given in Appendix 1.
While this stage in the analysis yields some interesting quantitative results,
a qualitative approach is necessary to explain them. A high-frequency verb
 BENGT ALTENBERG AND SYLVIANE GRANGER 177

such as MAKE expresses a variety of meanings and enters into a whole range of
structures and it is these di€erences in usage between learners and native
speakers and between the two categories of learners that are of particular
interest.

4. USES OF MAKE: MEANINGS AND PATTERNS

4.1 Eight major categories of use
The verb MAKE has many di€erent meanings and uses which we grouped into
eight major categories. Table 3 lists the categories and illustrates each of them
with one or more typical examples.

Table 3: Major uses of the verb MAKE

1. Produce sth (result of creation) make furniture, make a hole, make a law
2. Delexical uses make a distinction/a decision/a reform
3. Causative uses make sb believe sth, make sth possible
4. Earn (money) make a fortune, a living
5. Link verb uses she will make a good teacher
6. Make it (idiomatic) if we run, we should make it
7. Phrasal/prepositional uses make out, make up, make out of
8. Other conventional uses make good, make one's way

We assigned every instance of MAKE in the three corpora to one of the eight
categories of use. This task was greatly facilitated by WordSmith Tools' powerful
concordance sorting facility, which allows up to three levels of left- and right-
sorting. The sorting order we opted for (1st right, 2nd right and 3rd right)
resulted in clusterings of similar types of sequences such as make a di€erence,
make it possible, make sure that, which speeded up the subsequent
categorization.
The results of the classi®cation are given in Table 4. On the whole, the rank
order of the main uses of MAKE is similar in the three corpora. Category 3Ðthe
causative categoryÐis most common, followed by Category 2Ðthe delexical
category. The other categories are much less common and their rank orders
vary somewhat across the corpora.
However, if instead of looking at rank orders, one considers the absolute
frequencies of the di€erent uses in the three corpora, striking di€erences are
revealed. The delexical categoryÐcategory 2Ðis used signi®cantly less by the
two groups of learners than by the natives. Another striking di€erence
concerns category 3, which is signi®cantly more frequent in SW than in FR
and LOCNESS. In fact, there are nearly twice as many occurrences of
causative structures in SW as in FR. Among the other categories, one can also
178 GRAMMATICAL & LEXICAL PATTERNING OF MAKE IN STUDENT WRITING

Table 4: Uses of MAKE by EFL and NS students

Category SW FR LOCNESS

1. Produce 59* 19 27
2. Delexical 134* 134* 187
3. Causative 327* 174* 236
4. Money 25* 9* 56
5. Link uses 1 10 7
6. Make it Ð 1 6
7. Phrasal 25 29 22
8. Other 30 21 26
Total 601 397* 567

note an overuse of category 1 (Produce) in SW and an underuse of category 4

(Money) in both SW and FR.
In the following sections, we will concentrate on the two most frequent
categoriesÐdelexical and causative usesÐwhich display striking di€erences
across the corpora.

4.2 Delexical uses

The learner corpus data con®rms Sinclair's underuse hypothesis: both the
Swedish and French-speaking learners underuse delexical structures. This
shows that the two observations noted in Section 1 are not really paradoxical.
Learners may at the same time overuse a high-frequency verb and underuse
its delexical structures.
To go a bit deeper into the use of this structure by native and non-native
students, we examined all the collocates that occurred at least twice in each
corpus. They are listed in decreasing order of frequency in Table 5.
One of the striking di€erences between the native speaker corpus and the
two learner corpora lies in the frequency of `speech' or `verbal commun-
ication' collocates, i.e. words like argument, claim, point, statement, case,
comment, observation, reference (printed in bold in the table). According to
Collins Cobuild English Grammar (1990: 150±1), the verb MAKE is often used
with nouns expressing speech actions, many of which have related reporting
verbs (to make a remarkÐto remark that). The number of speech collocates
(tokens) in the three corpora is given in Table 6.
The table shows that almost a third of the native speaker tokens belong to
the category of speech nouns, while only 9±13 per cent of the learner
instances do.
However, learners do not only underuse delexical structures, they also
 BENGT ALTENBERG AND SYLVIANE GRANGER 179

Table 5: Main collocates of delexical MAKE in SW, FR, and LOCNESS

Swedish FR LOCNESS

e€ort(s) 13 progress 16 decision 31

mistake(s) 10 use 11 mistake 16
change(s) 9 e€ort(s) 10 choice 10
choice(s) 8 di€erence 9 argument 9
sacri®ces 7 choice(s) 8 claim 9
decision(s) 6 mistake(s) 7 point 8
impression 6 distinction 6 statement 8
progress 5 decision(s) 6 case 5
comparison 3 step 6 error 5
improvements 3 discoveries 6 e€ort 4
promises 3 sacri®ces 4 assumption 3
statement(s) 3 poll 4 attempt 3
achievement(s) 2 point 3 contribution 3
agreements 2 comparison(s) 2 discovery 3
alterations 2 contribution 2 impact 3
attempt(s) 2 attempt 2 judgement 3
deals 2 contacts 2 advance(s) 3
discoveries 2 statement 2 appointment 2
generalization(s) 2 impression 2 attack 2
investment 2 calculation 2
notes 2 call 2
Research(s) 2 change 2
trip(s) 2 comment 2
improvement 2
love 2
observation 2
reference 2

misuse them. In fact, it is this category that accounts for the majority of
learner errors with MAKE in the corpus. The following examples illustrate some
typical errors in SW and FR.
[1] We have to make a balance between material comfort and pleasure (FR)
(correct form: strike/®nd)
180 GRAMMATICAL & LEXICAL PATTERNING OF MAKE IN STUDENT WRITING

Table 6: Number of `speech' collocates of delexical MAKE

SW FR LOCNESS

n % n % n %

18* 13 12* 9 55 30

[2] If I made a poll among the Belgian population . . . (FR)

(correct form: carry out/conduct)
[3] When man makes a step, he wants to go further . . . (FR)
(correct form: take)
[4] The idea rose of making an end to them . . . (FR)
(correct form: put)
[5] we started using nature as a source of pro®t without considering the
serious harm we were making (SW)
(correct form: do)
[6] we've got to . . . use and develop technology and make scienti®c research
(SW)
(correct form: do)
[7] Sweden has to ful®l the requirements made by EU (SW)
(correct form: lay down)
All these examples illustrate cases where the verb MAKE should be replaced by
another verb.
The errors are partly interlingual and partly intralingual. Examples [2] and
[3] have corresponding structures with FAIRE in French, but examples [1] and
[4] do not. Similarly, examples [5] and [6] re¯ect corresponding structures
with GoÈRA in Swedish, while [7] does not. In addition, both [5] and [6]
illustrate the dicult choice between MAKE and DO facing EFL learners.
The following examples illustrate other types of error involving delexical
structures:
[8] . . . to attract the audience mainly in order to make bene®ts . . . (FR)
(correct form: make pro®ts)
[9] . . . spacecraft which really makes impression on our Johnny . . . (FR)
(correct form: make an impression)
[10] Because there is so little communication over the border, so few living in
the south that make visits to the north . . . (SW)
(correct form: visit/explore the north)
In [8] it is the noun collocate which is incorrect. In [9] the article is missing.
Both structures are directly transferred from French [faire des beÂne®ces and faire
impression sur]. In [10] the Swedish learner has not understood the ®ne
di€erence in meaning between the delexical construction make/pay a visit and
the simple verb visit.
 BENGT ALTENBERG AND SYLVIANE GRANGER 181

4.3 Causative uses

While the delexical use of MAKE creates similar problems for the Swedish and
French-speaking learners, with underuse and collocational misuse as a result,
the causative use of MAKE brings out a clear di€erence between the two
learner groups.
Grammatically, causative MAKE is a complex-transitive verb involving three
types of object + complement construction: adjective structures (make
something possible), verb structures (make somebody realize something) and
noun structures (make somebody a star). The distribution of these types in
the three corpora is given in Table 7.

Table 7: Causative uses of MAKE

Complement SW FR LOCNESS

adjective 179* 98(*) 130

verb 125* 67 80
noun 23 10* 26
Total 327* 174* 236

The ®gures point to clear di€erences in the use of these types by the
Swedish and French-speaking learners. While the Swedish learners show a
signi®cant overuse of adjective and verb structures, the French-speaking
learners reveal a consistent underuse of causative MAKE, especially noun and
adjective structures. As the noun complement structures are comparatively
infrequent in the material, we will concentrate our analysis on the ®rst two
types of structure.

Adjective structures
The Swedish learners' overuse of causative MAKE with adjective complements
can partly be explained as a result of positive transfer from the corresponding
Swedish construction with GOÈRA (cf. Eng make X happy = Sw goÈra X lycklig).
However, it can also be seen as an e€ect of overgeneralization of the target
pattern. Indeed, the parallelism between the two languages in this respect is
so obvious, and the English construction so easy to use, that it is natural to
suspect that the Swedish learners are enticed to employ the semantically
`decomposed' MAKE + sb/sth + Adj pattern even in cases where a native writer
would prefer a more `synthetic' causative verb alternative. This learner
strategy is supported by the fact that there are very few clear Swedish errors in
this category but a number of rather clumsy constructions. This is illustrated
in the following examples, where causative verb constructions were put
forward by native speaker informants as more natural alternatives.
182 GRAMMATICAL & LEXICAL PATTERNING OF MAKE IN STUDENT WRITING

[11] The use of the plastic wrap not only increases the garbage mountain, it
also makes the air polluted . . . (pollutes the air)
[12] . . . technology will never make imagination and dreams unnecessary . . .
(replace)
[13] I love the way the di€erences between men and women are blurred, or
even made non-existing (eliminated)
It seems then that the Swedish overuse of MAKE + Adj constructions is a
combined e€ect of interlingual and intralingual forces: both languages have
a dominant pattern with equivalent high-frequency verbs (MAKE/GOÈRA), but
they also have alternative constructions with single causative verbs that are
often lexically speci®c and consequently more dicult for learners. The
importance of such alternatives is clearly borne out in translations where
Swedish GOÈRA + adj is rendered by synthetic English verbs in nearly a
quarter of the cases (for example, goÈra moÈjlig `make possible' ? enable,
allow, let, permit; goÈra foÈrvaÊnad `make surprised' ? surprise, astonish; goÈra
generad `make embarrassed' ? embarrass; goÈra ren `make clean' ? clean) (see
Altenberg, forthcoming). Learners who are unfamiliar with such alter-
natives are likely to overuse the dominant target pattern and treat it as a
lexico-grammatical `teddy bear', especially if it is easy to transfer from their
native language.
Another indication of this strategy is the Swedish learners' fondness for the
constructions make it (im)possible (for sb) to and make it easy/easier (for sb) to.
Both have exact equivalents in Swedish and both can be replaced by causative
verb constructions in English (for example, enable sb to, prevent sb from, facilitate
sth). These patterns are greatly overused in SW and can almost be said to have
the status of prefabricated structures.
In sharp contrast to the Swedish learners' overuse of the causative MAKE +
adjective construction, there is a striking underuse of the construction by
French learners, for which there are two likely explanations. The ®rst is that
MAKE does not in this case correspond to its prototypical equivalent faire but to
another causative verb, rendre (make possible = rendre/*faire possible) and the
second is that the rendre + adjective construction seems to be less dominant
than its English counterpart with MAKE. In fact bilingual data indicate that
MAKE + adj is only likely to be translated by rendre + adj in c. 20 per cent of the
cases, other alternatives being to use synthetic verbs (make clean: assainir; make
stronger: renforcer) or verb + noun patterns (make accessible: faciliter l'acceÁs; make
more transparent: accroõÃtre la transparence).
While the high degree of congruence between the English and the Swedish
causative + adjective structures accounts for Swedish learners' overuse of this
construction, the much more blurred correspondence between the English
and French structures helps explain French learners' underuse.
 BENGT ALTENBERG AND SYLVIANE GRANGER 183

Verb structures
As to causative MAKE with a verb complement, a semantic analysis of the verb
used after MAKE proves enlightening. The verbs can be subcategorized into
three main semantic categories: relational (seem, appear, become), mental
(think, realize, understand) and actional (work, pay, change). The distribution of
these categories in the three corpora is given in Table 8.

Table 8: Causative MAKE with verb complements: semantic types of verb

Types SW FR LOCNESS

n % n % n %

Relational 4* 3 5* 7 25 31
Mental 65* 52 31 46 28 35
Actional 56* 45 31 46 27 34
Total 125* 100 67 100 80 100

The table indicates both similarities and di€erences between the learner
groups in comparison with the native writers. While the three verb categories
are about equally common in the native students' writing, both learner
groups show a signi®cant underuse of relational verbs. Thus, while a third of
the verb structures in the native corpus involve relational verbs, only 3±7 per
cent of the learner structures do. Examples such as the following are
conspicuously missing in the learner corpora:

[14] . . . they are lying to their consumers by making their product seem more
intriguing.
[15] This turnabout is an e€ort to make animal research appear to be more
ethical to animal rights activists.
On the other hand, the learner groups di€er clearly in their use of mental and
actional verbs. While the French-speaking learners use these categories to the
same extent as the native writers, the Swedish learners signi®cantly overuse
them. Their treatment of these categories is strongly reminiscent of their use
of the adjective structures illustrated above: the learners give the impression
of painstakingly trying to convey their meanings using the process of semantic
primitives as building blocks instead of using single verbs or more appropriate
causative verbs. The following examples illustrate what seems to be an almost
mechanical use of the MAKE + verb pattern in cases where a single causative
verb would be more appropriate. They all re¯ect a corresponding Swedish
structure with the causative verb FAÊ, which can be used with practically any
kind of verb complement:
184 GRAMMATICAL & LEXICAL PATTERNING OF MAKE IN STUDENT WRITING

to make its inhabitants open their eyes (? to open its inhabitants' eyes)
make people come closer to each other (? bring people closer)
the di€erences are made to vanish (? are eliminated)
pressure from the outside to make us change (? change us)
the larger part of the population of the earth are made to do with less (? are
forced to do with less/have to do with less)
Although the French-speaking learners do not signi®cantly overuse these
categories, their treatment of MAKE with actional and mental verbs often
seems mechanical and clumsy, and many of the examples give the impression
of the same `decompositional' strategy as used by the Swedes. The following
examples from FR illustrate this phenomenon:
earn money and make the family live (? and support the family)
this makes your fears diminish (? this reduces/allays your fears)
each movement makes him su€er (? causes him to su€er)
music can make you earn a lot of money (? can help you to earn a lot of
money)
Nearly all the French structures are directly translatable into a French
causative structure with FAIRE, which can be used with the full range of
actional verbs.
The frequent use of this type of structure by learners may be partly
teaching-induced. Most grammar courses focus on the verb MAKE in the
section on causative structures. Little is said about the other causative verbs:
CAUSE, HELP, ALLOW, ENABLE, HAVE, GET, etc. It is very dicult to ®nd a good
description of the usage di€erences between these verbs.
To conclude, the Swedish and French-speaking learners' use of causative
MAKE presents a rather complex picture. Unlike delexical MAKE, the causative
uses of MAKE bring out several striking di€erences between the two learner
groups. We have tried to explain the learners' performance as the result of
several factorsÐinterlingual, intralingual, and inadequate teaching. Broadly
speaking, we have suggested that cases where both learner groups deviate
signi®cantly from the native writers may re¯ect intralingual diculties, while
cases where only one of the learner groups deviates from the native writers
may indicate interlingual problems, with positive or negative transfer from L1
as a result. However, with learner data limited to only two national groups
and in the absence of good contrastive descriptions of causative constructions
in the languages involved, these explanations can only be very tentative. This
is especially obvious in cases where alternative `synthetic' single-verb
constructions compete with the dominant `analytic' pattern, in L1 as well as
in L2. Hence, one important conclusion to be drawn from this part of the
study is that there is a need for a wider range of learner data and detailed
corpus-based contrastive studies to supplement the learner data examined
here (see Granger 1998).
 BENGT ALTENBERG AND SYLVIANE GRANGER 185

5. WORDSMITH TOOLS' COLLOCATE DISPLAY

Our analysis of the collocability of MAKE has been largely manual. This
approach was possibleÐthough time-consumingÐbecause our corpora were
quite small. A manual collocate analysis of larger learner corpora would be
impracticable. In this section we will compare the results of our manual study
with results based on WordSmith Tools' Collocate Display. This tool displays the
words adjacent to a search word within a collocation `horizon' (or span)
which can go up to 25 words on either side. The default is 5 to the left and 5 to
the right. Table 9 gives part of the collocate display for MAKE in the French
learner corpus, with the default setting of ®ve words in either direction. The
collocates of MAKE are listed in frequency order in the second column,
preceded by their rank number and with the search word MAKE itself at the
top of the list. The third column gives the total number of times the collocate
co-occurred with MAKE in the corpus. The following columns (6±16) give a
breakdown of the total frequency: ®rst, the sum of all occurrences to the left
and right of MAKE (columns 4±5), then the frequencies for each position to the
left and to the right of MAKEÐfrom L5±L1 (5th±1st position to the left) to R1±
R5 (1st±5th position to the right). The centre position, representing the search
word MAKE itself, is indicated by a star.
The tool is fast and very user-friendly and is therefore a good starting-point
for identifying collocates in computer corpora. However, it has some
limitations that researchers ought to be aware of.
First of all, it is not possible to use the collocate display with lemmatized
entries. As we really wanted to use the tool, we found a way of going round
this diculty, not very scienti®c perhaps, but it works!6 We used the `search
and replace' tool in MS Word to replace all instances of makes/making/made by
make and then ran the collocate display on the new text ®le.
Other diculties arise from the fact that (a) the Collocate Display only
identi®es word forms as collocates, not lemmas (cf. Table 9: the plural form
e€orts is separate from the singular form e€ort); (b) it does not separate
homographs (cf. Table 9: order could be a verb, a noun, or part of the
complex conjunction in order to); (c) it displays potential collocates, i.e. words
which occur frequently in the span of MAKE, not all of which are
`constructional' collocates of MAKE (cf. the function words and the topic-
dependent noun Europe in Table 9). This is easy to demonstrate. In Table 10
some items selected from the LOCNESS collocate display are contrasted with
their constructional (i.e. object) uses in the corpus. Column 2 shows the
WordSmith Tools Collocate Display ®gures, column 3 shows the frequencies of
the constructional collocates of MAKE identi®ed in a manual analysis of the
corpus. The di€erences between these are sometimes considerable. For
instance, a fully automatic collocate search brings up many more instances of
argument than a manual analysis because it does not di€erentiate between
cases such as [16] where argument is a constructional collocate of MAKE and
[17] where it is not.
 186 GRAMMATICAL & LEXICAL PATTERNING OF MAKE IN STUDENT WRITING
Table 9: WordSmith Tools' Collocate Display for MAKE

N Word Total Left Right L5 L4 L3 L2 L1 * R1 R2 R3 R4 R5

1 MAKE 425 6 7 0 1 5 0 0 412 0 0 5 1 1

2 THE 205 82 123 17 30 18 16 1 0 44 13 14 31 21
3 AND 97 50 47 8 11 9 8 14 0 1 5 15 11 15
4 THAT 82 48 34 10 12 11 10 5 0 1 7 15 8 3
5 THEM 57 12 45 4 0 2 6 0 0 28 0 2 10 5
6 PEOPLE 47 18 29 3 3 3 4 5 0 17 6 2 2 2
19 EUROPE 23 23 11 12 4 3 1 3 0 2 2 1 5 2
34 ORDER 14 8 6 0 0 0 8 0 0 0 2 2 1 1
48 EFFORTS 10 7 3 1 1 2 1 2 0 0 2 1 0 0
 BENGT ALTENBERG AND SYLVIANE GRANGER 187

Table 10: Automatic vs manual collocate analysis of MAKE (LOCNESS)

Collocate Automatic Manual

argument 5 9
point 12 8
statement(s) 12 8
pro®t 6 7
attempt 5 3

[16] Yet, to make an even stronger argument, these proponents for adoptive
parents should state why this is a bad court decision and ®nd other
examples.
[17] The goal in this type of argument is to make the public aware of the truths
of nuclear power.
However, these diculties do not detract from the general usefulness of the
tool. On the basis of contrasted collocate displays such as those shown in
Table 11, which lists the 50 most frequent collocates in the three corpora, it is
possible to ®nd some of the relevant collocate di€erences between the three
corpora.7 The collocate possible, for instance, occurs much more frequently in
both learner corpora than in the NS data. Decision(s), on the other hand, occurs
much more frequently in the NS corpus (44 times) and is used both in the
singular and plural, whereas in the French learner list only the singular
features (5 occurrences) and in the Swedish learner list both singular and
plural forms are completely absent.
However, as the collocates are not ordered in any systematic way except by
frequency, the analyst will ®nd it very dicult to see the wood for the trees.
Also, as rightly pointed out by Stubbs (1995: 249), the `occurrence of any
single word form may be quite low, and will be missed in a simple list of
descending frequencies. What is signi®cant is the summed frequency of
semantically related items; and this can be spotted only by a human being.'
Using a computer-aided yet largely manual approach, we were able to classify
the occurrences of MAKE into grammatical and semantic categories. This in
turn enabled us to make some generalizations about the collocates which are
potentially very useful for teaching purposes, such asÐfor delexical
structuresÐthe frequency of speech noun collocates andÐfor causative
structuresÐthe frequency of relational verb collocates and the restrictions
on actional verb collocates. This description in terms of `semantic prosody'8
would be much harder to achieve with a fully automated method.9
188 GRAMMATICAL & LEXICAL PATTERNING OF MAKE IN STUDENT WRITING

Table 11: Top 50 collocates of MAKE in decreasing order of frequency

SW FR LOCNESS

people 43 people 47 people 42

possible 28 Europe 23 money 39
immigrants 25 progress 21 decision 23
easier 21 possible 20 life 20
money 20 think 16 mistake 16
world 20 world 14 easier 16
society 19 order 14 argument 15
way 17 use 13 better 15
feel 16 new 12 seem 14
think 16 di€erence 12 students 13
Swedish 16 like 11 society 13
life 16 money 11 other 12
want 15 television 10 feel 12
technology 14 e€orts 10 point 12
e€ort 13 dream 10 right 11
new 13 aware 9 decisions 11
change 13 European 9 lives 11
dreams 13 easier 9 year 11
time 13 life 9 good 11
like 11 man 9 others 10
try 11 other 8 work 10
mistake 10 good 8 di€erence 10
things 10 ®ction 8 family 9
man 10 mistake 8 world 9
aware 10 pro®t 8 like 9
di€erent 9 believe 8 claim 9
changes 9 clear 8 women 8
thing 9 important 7 important 8
part 9 choice 7 use 8
living 9 try 7 think 7
important 9 work 7 person 7
use 8 countries 7 states 7
better 8 imagination 7 possible 7
good 8 country 7 statement 7
realize 8 discoveries 6 computers 7
order 8 men 6 drug 7
 BENGT ALTENBERG AND SYLVIANE GRANGER 189

SW FR LOCNESS

place 7 live 6 day 7

sure 7 understand 6 television 7
reader 7 women 6 available 7
able 7 uni®cation 5 case 7
di€erence 7 united 5 aware 7
®lm 7 programmes 5 want 7
lives 7 draw 5 strong 7
love 7 decision 5 e€ort 7
example 7 community 5 system 6
impression 6 part 5 things 6
Sweden 6 mind 5 take 6
hard 6 situation 5 choices 6
country 6 keep 5 di€erent 6
friends 6 children 5 marijuana 6

6. CONCLUSION
Second language acquisition (SLA) research makes use of a variety of data
types. Ellis (1994: 670) classi®es them into the following categories: natural
language use data, elicited language use data, metalingual judgements, and
self-report data. Recent SLA studies have clearly favoured the last three data
types to the detriment of natural language use data. While more experimental
data o€ers the obvious advantage of allowing greater control over the
variables that in¯uence performance, it also has a series of disadvantages, the
most serious of which is the unnaturalness of learner productions due to the
arti®ciality of the experimental situation. Learner corpus data gives SLA
researchers an ideal way of redressing the balance. It constitutes a new type of
performance data which, unlike the decontextualized catalogues of errors
from the Error Analysis era, o€ers a valuable view of learners' interlanguage
(both errors and non-errors).
In addition, the fact that learner corpus data is in machine-readable form
makes it possible to analyse much more data than before. For the ®rst time
ever, it is now possible to conduct large-scale comparative analyses of the
interlanguage of di€erent learner populations and native speaker groups and
thus uncover their distinguishing features.
This is exactly what this study has aimed to do with regard to the use of a
high-frequency verb like MAKE. Results show that EFL learners, even at an
advanced pro®ciency level, have great diculty with a high-frequency verb
such as MAKE. Delexical uses prove to be particularly treacherous, but even
190 GRAMMATICAL & LEXICAL PATTERNING OF MAKE IN STUDENT WRITING

causative structures, which might at ®rst appear pretty `safe', turn out to
cause problems as well.
These results have interesting pedagogical implications because although
high-frequency verbs are encountered very early in instructional pro-
grammes, once they have been taught, they tend to be neglected. This is
particularly unfortunate because these verbs are extremely complex and
learners are at a risk of having only a very crude knowledge of their
grammatical and lexical patterning. We can only agree with Lennon's (1996:
23) suggestion that `teaching at the advanced level should aim not only to
increase the word store but also to ¯esh out the incomplete or `skeleton'
entries which even advanced learners may have for high-frequency verbs.' In
view of the in¯uence of transfer on the learners' use of these verbs, we also
agree with Lennon that learners would bene®t from `consciousness raising as
to areas in which lexico-semantic divisions do not correspond in L1 and L2'
(Lennon 1996: 35).
Concordance-based exercises extracted from native corpora are a useful
resource for raising advanced learners' awareness of the structural and
collocational complexity of high-frequency verbs. For instance, learners could
be asked to scan the concordance lines reproduced in Appendix 2 to draw up a
list of the major collocates of delexical MAKE. This awareness exercise would
then be followed by a consolidation exercise in which learners are asked to ®ll
in the blanks in corpus excerpts from which common collocates of MAKE
(decision, mistake, claim, argument, e€ort, etc.) have been removed. Such
exercises would increase the learners' `depth of processing' and hence
hopefully their degree of retention of the collocates.
(Revised version received July 2000)

ACKNOWLEDGEMENTS
We would like to thank Martin Bygate and two anonymous reviewers for their insightful
comments on an earlier version of the article.

APPENDIX 1: CHI-SQUARE VALUES

All frequency di€erences across the samples have been tested by means of the chi-
square test, with 1 per cent (p < 0.01) as the critical level of signi®cance. Absolute
frequencies of MAKE were related to the total number of words in each sample. Non-
signi®cant values are enclosed within parentheses.

Table 2A
SW vs LOCNESS FR vs LOCNESS SW vs FR

MAKE (0.74) 30.77 41.07

 BENGT ALTENBERG AND SYLVIANE GRANGER 191

Table 4A
SW vs LOCNESS FR vs LOCNESS SW vs FR

Causative MAKE 14.03 9.68 46.28

Delexical MAKE 9.14 9.01 (0.004)
Produce MAKE 11.66 (1.43) 20.40
Money MAKE 12.09 34.21 7.48

Table 6A
SW vs LOCNESS FR vs LOCNESS SW vs FR

Speech collocates 19.03 27.91 (1.18)

Table 7A
Causative MAKE SW vs LOCNESS FR vs LOCNESS SW vs FR

adjective 7.39 (4.65)* 23.46

verb 9.53 (1.21) 17.36
noun (0.20) 7.19 5.08

*Signi®cant at the 5% level (p < 0.005)

Table 8A
Semantic type SW vs LOCNESS FR vs LOCNESS SW vs FR

relational 15.36 13.43 (0.11)

mental 14.43 (1.03) 11.95
actional 9.91 (0.25) 7.12
192 GRAMMATICAL & LEXICAL PATTERNING OF MAKE IN STUDENT WRITING

APPENDIX 2: CONCORDANCE OF DELEXICAL MAKE

 BENGT ALTENBERG AND SYLVIANE GRANGER 193

NOTES
1 The deceptive cross-linguistic similarity of puted the frequencies of MAKE in three larger
high-frequency verbs is also revealed in native standard corpora: the 1 million word
translations. This can be illustrated by a LOB and Brown corpora and the 200 million
comparison of three common verb pairs in word British National Corpus (BNC). The
English and Swedish whose members are normalized frequencies (occurrences per
usually regarded as translation equivalents: 1,000 words) of MAKE in these corpora were
the cognates GO/GAÊ and GIVE/GE, and the 247 (LOB), 221 (Brown), and 213 (BNC), i.e.
prototypical `creation' verbs MAKE/GOÈRA. An much lower than in LOCNESS. This di€er-
examination of their `mutual translatabil- ence is probably due to the fact that MAKE is a
ity'Ði.e. the tendency of the members of style-sensitive verb and that stylistic di€er-
each pair to be translated into each otherÐ ences are ironed out in so-called balanced
in the English±Swedish Parallel Corpus (see corpora such as LOB, Brown, and BNC,
Aijmer et al. 1996), shows that they corre- which contain a variety of genres. Although
spond in only about a third of the cases: LOCNESS may not be an ideal standard of
English /? Swedish comparison in every respect, it was chosen
GO GAÊ 35% as control corpus for this study because it
GIVE GE 32% ensures full comparability in terms of text
MAKE GOÈ RA 30% type.
In other words, although the members of 5 When the level of statistical signi®cance is
each pair are generally regarded as cross- 5 per cent (p < 0.05), the asterisk is enclosed
linguistic equivalents, they are in fact ren- in parentheses.
dered by other verbs in the majority of cases 6 We would like to thank Fanny Meunier who
in translations between the languages. (The provided us with this brilliant idea.
®gures for the ®rst two pairs are from Viberg 7 Function words have been excluded from
(1996: 161); the ®gures for MAKE/GOÈRA the lists.
represent their `delexical' usesÐsee section 8 The term `semantic prosody' refers to the
4.2.) tendency for words to collocate with other
2 It could be argued that the corpora used are words from a de®nable semantic set. Stubbs
too small to produce stable results in fre- (1995), for instance, shows that the lemma
quency terms. It is true that each essay CAUSE is overwhelmingly used with negative
contains a limited number of occurrences of events such as problems, trouble, damage,
the variable under investigation, viz. MAKE. It death, pain, and disease.
should be borne in mind, however, that our 9 In fact WordSmith Tools comes with a warn-
focus of interest is not on the individual ing against using computer tools to replace
variation between the learners but on their the human researcher: `The computer is an
behaviour as groups and on the di€erences awful device for recognizing patterns. It is
between the total frequencies of MAKE, and good at addition, sorting, etc. It has a
certain uses of MAKE, across the groups (SW vs memory but it does not know or understand
FR vs LOCNESS). Within that perspective, anything . . . Nevertheless, the computer is a
corpora of c. 300 essays of 500+ words should good device for helping humans to spot
be large enough to give a reliable picture of patterns and trends. That is why it is import-
the uses of MAKE in each student population. ant to see computer tools such as WordSmith
3 For a review of the software, see Berber Tools in their true light. A tool helps you to
Sardinha (1996). do your job, it doesn't do your job for you'
4 For the sake of comparison we also com- (Scott 1997: 7).

REFERENCES
Aijmer, K., B. Altenberg, and M. Johansson. English. Presentation of a project' in
1996. `Text-based contrastive studies in K. Aijmer, B. Altenberg, and M. Johansson
194 GRAMMATICAL & LEXICAL PATTERNING OF MAKE IN STUDENT WRITING

(eds): Languages in Contrast. Lund University Academic Writing. Max Niemeyer Verlag:
Press: Lund, pp. 73±85. TuÈbingen.
Allerton, D. J. 1984. `Three (or four) levels of KaÈllkvist, M. 1999. Form-Class and Task-Type
cooccurrence restriction.' Lingua 63, 17±40. Effects in Learner English. A Study of Advanced
Altenberg, B. (forthcoming) `Advanced Swed- Swedish Learners. Lund University Press:
ish learners' use of causative make. A con- Lund.
trastive background study' to appear in Kanatani, K., K. Itoh, T. Noda, Y. Tono, and
S. Granger, J. Hung, and S. Petch-Tyson K. Oikawa. 1995. `Teacher correction and
(eds): Computer Learner Corpora, Second lexical errors in EFL compositions.' Depart-
Language Acquisition and Foreign Language ment of English Education, Tokyo Gakugei
Teaching. University: Tokyo.
Berber Sardinha, A. P. 1996. `Review: Word- Lennon, P. 1996. `Getting `easy' verbs wrong
Smith Tools.' Computers and Texts 12. at the advanced level.' IRAL 34/1: 23±36.
Chi, A. M. L., K. Wong, and M. Wong. 1994. Lindner, C. 1994. `Unnaturalness in advanced
`Collocational problems amongst ESL lear- learners' English: A corpus-based feasibility
ners: a corpus-based study' in L. Flowerdew study.' Unpublished MA Dissertation,
and A. Tong (eds): Entering Text. The Hong Albert-Ludwigs-UniversitaÈt Freiburg.
Kong University of Science and Technology: Scott, M. 1997. WordSmith Tools Manual.
Hong Kong, pp. 157±65. Version 0.2. Oxford University Press. http: //
Collins Cobuild English Grammar. 1990. Collins: liv.ac.uk/~ms2928/wordsmit.htm
London and Glasgow. Sinclair, J. 1991. Corpus, Concordance, Colloca-
Ellis, R. 1994. The Study of Second Language tion. Oxford University Press: Oxford.
Acquisition. Oxford University Press: Oxford. Stubbs, M. 1995. `Collocations and semantic
Granger, S. 1996. `Romance words in English: profiles. On the cause of the trouble in
From history to pedagogy' in J. Svartvik quantitative studies.' Functions of Language
(ed.): Words. Proceedings of an International 2/1: 23±55.
Symposium. Almqvist and Wiksell Interna- Svartvik, J. and O. Ekedahl. 1995. `Verbs in
tional: Stockholm, pp. 105±21. private and public speaking' in B. Aarts
Granger, S. 1998. `The computerized learner and C. Meyer (eds): The Verb in Contem-
corpus: a versatile new source of data for porary English. Theory and Description. Cam-
SLA research' in S. Granger (ed.): Learner bridge University Press: Cambridge, pp. 273±
English on Computer. Addison Wesley Long- 89.
man: London and New York, pp. 3±18. Viberg, A Ê . 1996. `Cross-linguistic lexicology.
Hasselgren, A. 1994. `Lexical teddy bears and The case of English go and Swedish gaÊ' in
advanced learners: a study into the ways K. Aijmer, B. Altenberg and M. Johansson
Norwegian students cope with English voca- (eds): Languages in Contrast. Lund University
bulary.' International Journal of Applied Lin- Press: Lund, pp. 151±82.
guistics 4: 237±60. WordSmith Tools. 1996. Oxford University Press:
Howarth, P. 1996. Phraseology in English Oxford.