Enriching the phraseological coverage of high-frequency adverbs in English-French bilingual dictionaries

Full text

1
Enriching the phraseological coverage of high-frequency
adverbs in English-French bilingual dictionaries
Sylviane Granger and Marie-Aude Lefer
1. Introduction
In many of his publications, from the early descriptions of the English-Norwegian
parallel corpus project (1994) to his magisterial volume published in 2007, Stig
Johansson has explicitly identified bilingual (and multilingual) lexicography as one
of the most obvious applications of cross-linguistic corpus research. Probably due
to the lack of good quality bilingual corpora (especially translation corpora), the
impact of corpora on bilingual dictionaries has however been fairly limited. As
pointed out by Salkie (2008), this situation contrasts sharply with that for
monolingual lexicography: “[t]ranslation (parallel) corpora are standard tools in
several fields, such as translator training, machine translation, contrastive
linguistics, and various language engineering applications. One area where one
might expect such corpora to be widely used is bilingual lexicography, but in fact
such corpora have not been exploited significantly in dictionary compilation –
unlike monolingual lexicography, where it would be unthinkable today not to use
single-language corpora”.
One of the aspects of dictionaries that has benefited most from corpus
analysis is phraseological coverage. However, advances in this area too have been
much slower in bilingual dictionaries than in monolingual ones, a point noted by
Rundell in 1999, and which is still largely true today: “[t]he extraordinary range of
lexical and grammatical information they [monolingual learners’ dictionaries]
include is rarely even approached by the best bilingual dictionaries available”. This
is not to say that phraseology is absent from bilingual dictionaries. Major
improvements have been made in recent years, notably in the treatment of
collocations (Atkins 1996, Lubensky & McShane 2007).
The “huge area of syntagmatic prospection” (Sinclair 2004: 19) opened up
by corpus linguistics covers a wide range of multiword units. In addition to
traditional categories such as idioms, proverbs and collocations, automatic
techniques have uncovered some phraseological patterns that do not fit into any of
the usual classifications. Among them are lexical bundles, which Biber et al.
(1999: 990ff) defined as “simple sequences of word forms that commonly go
together in natural discourse”. These routinised sequences, or “‘preferred’ ways of
saying things” to use Altenberg’s (1998: 122) words, include verbal phrases (e.g.
suffice it to say), nominal phrases (the extent to which), prepositional phrases (in
the case of) and adverbial phrases (as yet). Unlike the units that were focused on in
pre-corpus phraseology, these units tend to be semantically compositional. As a
result, they are less salient and even very advanced users of the language – among
them, translators and lexicographers – may fall into the trap of literal translation, an
option which leads to clumsy, if not downright incorrect, formulation. Lexical
bundles have been largely neglected in bilingual dictionaries. They are sometimes
included as middle- or back-matter sections designed to help users write essays or
letters, but as pointed out by Granger & Paquot (2008), these sections feature some

2
questionable and/or infrequent phrases and would greatly benefit from the
incorporation of more corpus data.
It is therefore urgent to find ways of ‘phrasing up’ the bilingual dictionary.
This is all the more important as several studies have shown that users have a
preference for bilingual dictionaries (see Lew 2004: 18–20). Some recent studies
have convincingly demonstrated the key role that bilingual corpus data can play in
improving the collocational coverage of bilingual dictionaries (e.g. Ferraresi et al.
2010). In this article our focus is on a different category of multiword units, that of
lexical bundles, identified by the n-gram extraction method. We focus on high
frequency words, a category which displays a high rate of phraseological uses. As
adverbs have received scant attention so far, we selected two high-frequency
adverbs: the French adverb encore and the English adverb yet, which are frequent
translation equivalents.
1
In Section 2 we describe the corpora and the bilingual
dictionaries used for the investigation. In Section 3 we examine the coverage of
corpus-derived bundles in bilingual dictionaries both quantitatively (the proportion
of bundles included) and qualitatively (their place in the dictionary microstructure).
Section
4 describes the contribution of corpus data to the translation of lexical
bundles in bilingual dictionaries, and the final section presents some concluding
remarks and suggestions for future research.
2. Data and methodology
Our study relies on both lexicographic and corpus data. The two types of data are
described in this section, together with the methodology used to investigate lexical
bundles and their translation equivalents. For the dictionary analysis, we made use
of the yet and encore entries in three English-French electronic dictionaries: Le
Grand Robert & Collins (2008) (henceforth referred to as RC), Grand Dictionnaire
Hachette-Oxford (2003) (henceforth HO) and Harrap’s Unabridged Pro (2004)
(henceforth HU). HO and RC are corpus-informed: English and French
monolingual data were used to devise and/or revise the bilingual entries. The
situation for HU is less clear: its introduction specifies that the dictionary is based
on “texts in searchable databases” but no further details are provided.
Our approach relies on a two-stage methodology. The first stage consists of
extracting lexical bundles including encore or yet from original French and English
texts respectively. This stage requires the use of monolingual reference corpora,
which should ideally be as large and representative as possible. Translation corpora
are then used in the second stage to identify the translation equivalents of the
chunks uncovered in the first part of the analysis.
The case study of the French adverb encore is a first attempt at
implementing this methodology. French currently suffers from the lack of a
representative corpus along the lines of the British National Corpus (BNC) or the
Corpus of Contemporary American English, and so the study was exclusively
based on the Label France (LF) unidirectional translation corpus. This corpus
1
We selected yet as it is an equivalent of encore that has a high rate of phraseological uses
(43% of phraseological uses in the British National corpus), which is not the case for other
frequent equivalents such as still (8%) or again (15%).

3
contains 1million words and is made up of French magazine articles translated into
English.
2
The LF corpus was used for both stages of the analysis: extraction of the
bundles from the original French texts and identification of their translations in the
English target texts. The wider availability of corpus resources for English allowed
us to refine this methodology for the analysis of the adverb yet. We used the 100-
million-word BNC as a reference corpus to extract bundles including yet. We then
relied on two large bidirectional translation corpora to zoom in on the French
translations of some of these bundles: PLECI
3
, which contains news and fiction,
and the Europarl5 corpus (Koehn 2005), which consists of the proceedings of the
European Parliament. In the yet study, to exploit the full potential of the
methodology, both translation directions were investigated (yet in English source
and target texts).
We used the n-gram method to extract lexical bundles. This method is
employed in a wide range of research fields, notably in English for Academic
Purposes research (see Biber et al. 2004 and Ellis et al. 2008), but remains largely
under-exploited in bilingual lexicography. We extracted 2- to 5-grams with
WordSmith Tools 5 (Scott 2008) and imposed a frequency cut-off of 5 occurrences
per million words for encore in Label France, and 5 occurrences per 10 million
words for yet in BNC. These are relatively low frequency thresholds: Biber and his
colleagues (1999, 2004) used much higher frequencies (between 10 and 40
occurrences per million words). We then manually edited the n-gram lists to weed
out strings which were unlikely to be of lexicographic interest (see Tables 1 and 2
for examples of rejected and selected lexical bundles respectively).
Table 1. Examples of rejected n-grams with yet and encore
English: yet the, yet it, yet at the, yet he had, yet they were, have yet been, is not
yet known
French: est encore, encore le, encore dans, reste encore, encore à la, on peut
encore, il y a encore
Table 2. Examples of selected lexical bundles with yet and encore
English: not yet, and yet, yet another, as yet, yet to be (+ past participle)
French: ou encore, pas encore, encore plus, encore aujourd’hui/aujourd’hui
encore, là encore
Label France and PLECI, which are both sentence-aligned, were analysed with the
ParaConc multilingual concordancer (Barlow 2002), while the Europarl5 sub-
corpora were investigated via the Sketch Engine (Kilgarriff et al. 2004). Note that
2
The Label France and PLECI corpora used in this study were compiled at the Centre for
English Corpus Linguistics (University of Louvain). See http://www.uclouvain.be/en-
258636.html for more information.
3
PLECI (Poitiers-Louvain Échange de Corpus Informatisés) is the result of the
collaboration between the University of Louvain and the University of Poitiers.

4
for Europarl5 we only used texts which had English or French explicitly identified
as the source language. This was done by creating sub-corpora in the Sketch Engine
on the basis of the ‘speaker.language’ criterion (here ‘fr’ and ‘en’).
3. Lexical bundles with encore and yet: dictionaries vs. corpus data
This section analyses lexical bundles including the French adverb encore and the
English adverb yet, and examines how they are treated in the three bilingual
dictionaries reviewed. Section 3.1 looks at the proportion of corpus-extracted
lexical bundles included in the dictionary entries and Section 3.2 takes a closer
look at the lexical bundles that are included and describes their place in the
microstructure of the dictionaries (sub-entry, example, etc.).
3.1 Coverage
In this section, we aim to find out whether lexical bundles are well covered in
current bilingual entries. We will first look at encore (French-to-English half of the
dictionary), before turning to yet (English-to-French half).
3.1.1
E
NCORE
The corpus analysis of the original French texts of Label France shows that lexical
bundles including encore are very frequent: they make up 47% of all encore uses in
the corpus (352 out of 745 occurrences), which is a considerable proportion.
Combining the different chunks found in the corpus and those listed in the
bilingual entries gives us a total of 34 different multiword uses of the adverb. In
terms of coverage, as can be seen in Table 3, the number of different lexical
bundles found in each of the three dictionaries and in the LF corpus is similar
(between 18 and 21 chunks out of 34).
Table 3. Number of different multiword uses of encore in the bilingual dictionaries
and the Label France corpus
HU
RC
HO
LF corpus
Total
18
19
21
21
34
53%
56%
62%
62%
100%
This first impression of similarity is quickly dispelled when we look at which
lexical bundles are present in each resource. In fact, there is relatively little overlap
between the dictionaries and the LF corpus: (1) 41% of the chunks that are found in
the corpus are not well represented in the bilingual dictionaries (they occur in none,
or only one, of the three dictionaries); (2) 26.5% of the lexical bundles that are well
covered by the dictionaries (i.e. present in two or three dictionaries) do not occur at
all in the LF corpus; (3) a meagre 5 chunks out of 34 (14.7%) are included in all
three dictionaries and the corpus. To understand the reasons for these striking
differences, we need to look at the lexical bundles themselves. The two sets of
chunks (those that are only found in dictionaries, and those that are not well-
represented in dictionaries but are found in the corpus), are presented in Table 4. It

5
is clear that the lexical bundles that are found only in dictionaries are characteristic
of speech, whilst those that are found in the corpus but neglected in dictionaries are
typical of writing. Dictionaries seem to favour interactional and attitudinal markers
typical of speech, as illustrated in examples (1) to (6). In the LF corpus, by contrast,
we find many cohesive markers typical of writing that fulfil a range of functions,
such as addition, disjunction, enumeration, concession and emphasis (see examples
7 to 10).
Table 4. Lexical bundles including encore in bilingual dictionaries and in the LF
corpus
Not found in the LF corpus, but listed
in 2 or 3 dictionaries
Found in the LF corpus, but listed in
at most 1 dictionary
encore toi ! / encore vous !
ou (bien) encore
encore une chance que
mais encore
encore heureux
là encore
et encore !
une fois encore
mais encore ?
(et) plus encore
quoi encore ? / (et puis) quoi encore ?
et bien/beaucoup d’autres encore
si encore
plus encore ADJ/ADV que
encore autant
encore davantage
encore pire / pire encore
encore et encore
encore rien
encore et toujours / toujours et encore
encore aujourd’hui / aujourd’hui encore
mieux encore
encore à (+ inf.)
(1) il a dit qu’il avait bien aimé, mais encore ? (HU)
he said he liked it - but what exactly did he say?
(2) encore une chance qu’il n’ait pas été là ! (HU)
thank goodness or it's lucky he wasn't there!
(3) encore vous ! (RC)
(not) you again!
(4) si encore je savais où ça se trouve, j’irais bien (RC)
if only I knew where it was, I would willingly go
(5) c’est tout au plus mangeable, et encore ! (HO)
it's only just edible, if that!

6
(6) et puis quoi encore ! (HO)
what next!
(7) Ce sont quelques-uns de ces artistes que Label France a choisi de vous
présenter dans les domaines renommés et bien établis que sont le vitrail,
la céramique, le verre ou encore la tapisserie. [LF]
For this issue "Label France" has selected a small number of these
artists in the well-known and well established fields of stained glass
window-making, ceramics, glass-blowing and tapestry.
(8) Or, là encore, l’être humain doit coopérer à sa venue, lui permettre
d’obtenir une place. [LF]
Yet here too, the human being has to co-operate with its coming, to
allow it to obtain a place.
(9) Sur scène, Marianne Sergent, Zouc, Sylvie Joly, les Jeanne, pour la
première génération, puis, dans les années 1980-1990, Muriel Robin, les
Vamps, Anne Roumanoff, Valérie Lemercier, Michèle Laroque, et bien
d’autres encore, renoncent à leurs atouts millénaires mais aliénants
(…). [LF]
On stage, Marianne Sergent, Zouc, Sylvie Joly, the Jeannes, for the first
generation, then in the 1980-1990s, Muriel Robin, the Vamps, Anne
Roumanoff, Valérie Lemercier, Michèle Laroque, and many more
besides, are abandoning their age-old but maddening assets (…).
(10) Une fois encore, Sautet transcende le polar en le poussant vers l’étude
de mœurs, dans une ambiance très noire qui surprend. [LF]
Once again, Sautet transcends the whodunit by pushing it towards a
study of moral standards, in a very black and surprising atmosphere.
It is unsurprising to find uses typical of writing in a written corpus like LF; the
preference for speech displayed by the bilingual dictionaries is more remarkable.
The reason for this preference might be that spoken lexical bundles are more
cognitively salient, i.e. they stand out in our minds when we think about language
(Hanks 2000). Our study therefore suggests that lexicographers need corpus data
and corpus-driven methods for the automatic extraction of recurrent sequences that
are less cognitively salient. In the absence of corpus data, they run the risk of
overlooking these chunks when devising or revising bilingual entries.
3.1.2 Y
ET
The English adverb yet, like encore, is frequently found in lexical bundles: 43% of
the occurrences of yet in the BNC are instances of multiword uses (14,724 out of
33,980 occurrences). However, the coverage in bilingual dictionaries is
considerably richer than that for encore. As shown in Table 5, the most frequent
chunks with yet, such as not yet, and yet, yet another and as yet, are included in the
bilingual entries in at least two of the three dictionaries reviewed.

7
Table 5. Most frequent lexical bundles with yet in the BNC and their coverage in
bilingual dictionaries
Lexical bundles
Frequency in the
BNC
Number of dictionaries that
include the lexical bundles
not yet
4,357
3
and yet
3,442
2
yet another
1,500
3
as yet
1,419
3
yet to +
INFINITIVE
1,115
3
yet to be +
PAST
PARTICIPLE
648
2
yet again
642
3
yet more
372
3
The encore and yet case studies point to an imbalance between the two halves of
the dictionaries: lexical bundles seem to be relatively well covered in the English-
to-French half, while the coverage of the French-to-English half seems to be much
patchier. One tentative explanation for this difference could be that English and
French are part of radically different ‘corpus cultures’. There are many more
corpora of English than of French. Most English dictionaries are corpus-informed,
while the ‘corpus revolution’ appears to have largely passed French lexicography
by.
3.2 Presentation
As regards presentation, we found that most lexical bundles including encore and
yet are buried in bilingual entries, i.e. very few chunks are listed as sub-entries in
their own right. The boxed bundles in Figure 1 (and yet, not yet, yet to, as yet, yet
more, yet another and yet again) are either only found in contextualised example
sentences (e.g. the campaign has yet to begin) or listed as decontextualised items
within sub-entries (e.g. not yet, yet again) The situation is similar in the other two
dictionaries examined: no lexical bundle with yet is granted sub-entry status (let
alone headword status).

8
Figure 1: yet entry in Hachette-Oxford (2003)
In the French-to-English half of the dictionaries, only five chunks (out of 34 if we
consider the lexicographic and corpus data together) are listed as sub-entries:
encore que (‘even though’) [HO, HU, RC], et encore (‘if that’) [HO, RC], pas
encore (‘not yet’) [RC], encore et encore (‘again and again’) [RC] and si encore
(‘if only’) [RC]. As in the English-to-French half, the other chunks – if included at
all – are buried in the entries, for example in the form of an example.
Clearly, the presentation of lexical bundles could be improved in both parts
of the dictionaries, notably by granting headword status to lexical bundles. As
shown by Tono (2000), phrases are much easier to find if they are listed as
headwords rather than incorporated into the entry itself. Our analysis reveals
another weakness of current bilingual dictionaries, viz. the poor quality of the
examples. The selected examples are often atypical, and this is particularly
problematic in view of the fact that a large proportion of lexical bundles are only
introduced via examples. This is best illustrated with as yet in the entry presented
in Figure 1. The only example mentioned in HO is the as yet unfinished building.
However, the the as yet… structure accounts for only 1.5% of the occurrences of as
yet in the BNC. A corpus-based phraseological analysis can ensure that the
examples included in the dictionary meet the two major requirements of good
examples, i.e. authenticity and prototypicality (Cappeau 2010: 129).

9
4. Enriching the translations with bilingual corpus data
This section is devoted to the translation equivalents of (some of) the bundles
including encore (Section 4.1) and yet (Section 4.2). Our starting-point assumption
is that the most frequent equivalents of the adverbs and the lexical bundles of
which they form part should be included in the bilingual entries.
4.1 The translation of lexical bundles including encore
Table 6 presents the most frequent translations of the encore occurrences found in
LF, in decreasing order of frequency, and their inclusion (or lack thereof) in the
three dictionaries. No distinction is made in the table between the phraseological
and non-phraseological uses of encore.
Table 6. English translations of the most common occurrences (minimum
frequency 5) of encore in the Label France corpus and their appearance in three
bilingual dictionaries
English translations
Freq. in LF corpus
HU
RC
HO
still
234
√
√
√
and
48
×
×
×
not yet
43
√
√
√
or
32
×
×
×
even
ADJ
-er / even more
ADJ
30
√
√
√
further
18
×
×
×
even today
15
×
×
×
or even
15
×
×
×
and even
14
×
×
×
again
13
√
√
√
once again
11
×
√
√
yet
11
×
×
×
here too
10
×
×
×
as yet
9
√
×
×
even
9
×
√
×
or again
8
×
×
×
as well as
7
×
×
×
yet to be +
PAST PARTICIPLE
7
×
×
×
another
6
√
√
√

10
even more
6
√
×
×
still today
6
×
×
×
more ADJ still
5
×
×
×
too
5
√
√
×
TOTAL no. of translations 23
8
8
6
The coverage of the three dictionaries is very similar. But between them the
dictionaries only cover between a quarter and a third of the top 23 most frequent
translation patterns found in the corpus. Regrettably, these are not necessarily even
the most frequent ones. We see, for example, that two frequent translations, and
and or, are not listed. Several translation equivalents extracted from Label France
could be used to enrich the bilingual entry for both phraseological and non-
phraseological uses of encore. We will illustrate this by means of two French-
English pairs: encore - further, and ou encore - and/or.
Encore is a highly polysemous adverb (Mosegaard Hansen 2002). It can,
among other things, mean davantage in French, and a number of corresponding
translations are recorded in the dictionaries (more, even more, another; see
examples 11 to 13). In the LF corpus further is quite a frequent translation of
encore, especially when encore is used in combination with a verb (see examples
14 and 15). Even though further clearly deserves inclusion in the bilingual entry, it
is not mentioned in any of the dictionaries we reviewed.
4
(11) encore un mot, avant de terminer (RC)
(just) one more word before I finish
(12) réduisez-le encore (HU)
Reduce it even more
(13) encore une tasse de café (HU)
another cup of coffee
(14) Or tous les pays n’ont pas ratifié les deux pactes, certains pays
privilégiant l’un ou l’autre, ce qui conduit déjà à une dissociation de
fait, encore accentuée par une autre dissociation qui tient aux
mécanismes de contrôle, beaucoup plus développés en matière de
droits civils et politiques qu’en matière de droits économiques,
culturels et sociaux. [LF]
Yet not all countries have ratified both pacts, some countries
favouring one or the other, which in itself leads to a de facto
dissociation, further emphasized by another dissociation to do with
the mechanisms of control, which are far more developed in the area
4
It should be added, however, that even further is mentioned in HU. This equivalent is not
present in LF.

11
of civil and political rights than in that of economic, cultural and
social rights.
(15) Nous devons aussi améliorer encore la sécurité des transports et
renforcer la lutte contre le financement du terrorisme. [LF]
We also have to further improve transport security and step up the
fight against the financing of terrorism.
The usefulness of translation corpus data can also be illustrated with the lexical
bundle ou encore (and its variant ou bien encore), which is used in French to
convey enumeration. Ou encore is the most frequent chunk with encore in LF. It
accounts for 20% of all encore uses in the corpus but is only recorded in one
dictionary (HO). Table 7 contains the English translations of ou encore found in LF,
in decreasing order of frequency. The most frequent translations are and and or
(58%), as illustrated in examples (16) and (17), while the only translation found in
HO is or else (example 18). Or else is not present in the LF corpus. It can be used
to enumerate verbs (see Example 18: swim, go scuba diving or else learn to sail)
but it would be an incorrect translation in the majority of the corpus examples
examined here, where ou encore is used to enumerate nouns or prepositional
phrases (introduced by à, contre, en, par or pour) rather than verbs. In other words,
the example listed in HO is atypical.
Table 7: English translations of ou encore in the Label France corpus
English translation
Frequency in
LF corpus
%
and
47
34.6
or
32
23.5
or even
15
11.0
and even
14
10.3
or again
8
5.9
as well as
7
5.1
and … too
4
2.9
and also
2
1.5
and again
1
0.7
both
1
0.7
or then again
1
0.7
then
1
0.7

12
no translation
5
3
2.2
Total
136
100
(16) Des groupes financiers comme Dexia Asset Management, BNP
Paribas ou encore la Macif Gestion sont engagés dans ce processus.
[LF]
Some financial groups such as Dexia Asset Management, BNP
Paribas and Macif Gestion are committed to this process.
(17) Car si l’humour anglais - le terme est même de leur invention -, la
fantaisie slave ou encore l’autodérision des Italiens sont, par exemple,
réputés, la France, malgré des succès parfois même internationaux
dans ce domaine, n’est pas automatiquement associée à l’humour.
[LF]
Whereas English humour - they even invented the term - Slavonic
whimsy or the self-mockery of Italians, for instance, are well-known,
France, in spite of successes in this field sometimes even at
international level, is not automatically associated with humour.
(18) vous pouvez pratiquer la natation, la plongée sous-marine ou encore
vous initier à la voile [HO]
you can swim, go scuba diving, or else learn to sail
A caveat is in order, however. In spite of the undeniable usefulness of translation
corpus data, such data should not be used indiscriminately in lexicographic projects.
The corpus contains a non-negligible number of incorrect translations of ou encore,
most of which are due to source text interference. In several cases the translator
seems to have been unaware of the meaning of ou encore and mistranslated it into
or even, and even and or again (see examples 19 to 21). These cases account for
27% of the translation corpus data, which is a substantial proportion.
(19) Lille renferme l’une des plus prestigieuses collections françaises de
peintures, dessins, sculptures, faïences et objets d’art: celle- ci n’a-t-
elle d’ailleurs pas fait l’objet d’expositions à New York, Londres ou
encore au Japon au cours de ces cinq dernières années ? [LF]
The Lille museum houses one of the most prestigious of France’s
collections of paintings, drawings, sculptures, ceramics and objets
d’art, all of which have been exhibited in New York, London and
even Japan in the course of the last five years.
5
The category ‘no translation’ includes cases of zero translation and cases where a whole
sentence or paragraph containing ou encore in the source text has not been translated.

13
(20) Un apprentissage que les candidats pourront poursuivre, en France,
dans l’un des nombreux centres d’enseignement du FLE, les
universités, à l’Alliance française ou encore dans les chambres de
commerce et d’industrie. [LF]
Applicants will be able to continue the learning process in France at
one of the many FLE teaching centres, universities, Alliances
Françaises or even at chambers of commerce and industry.
(21) Milan Kundera, après avoir fui sa Tchécoslovaquie natale, a lui aussi
fini par adopter notre langue, comme Cioran, l’écrivain roumain,
Vassilis Alexakis, le Grec, ou encore François Cheng, le plus
francophone des écrivains chinois. [LF]
Milan Kundera, having fled his native Czechoslovakia, ended up
adopting our tongue, as did Cioran, the Romanian writer, Vassilis
Alexakis, the Greek writer, or again François Cheng, the most
accomplished of the Chinese writers writing in French.
Some researchers might be tempted to use the imperfection of translation corpora
as an argument against using translation corpus data in bilingual lexicography. We
believe, however, that lexicographers, who are highly-skilled bilinguals, are
perfectly capable of separating the wheat from the chaff. One added bonus of these
mistranslations is that they draw lexicographers’ attention to frequent errors and
provide them with useful material to design corpus-informed warning boxes such
as ‘do not literally translate ou encore into or even/and even or or again’ which
could be included in the bilingual entries.
4.2 The translation of lexical bundles with yet
This section focuses on the French translation of two frequent lexical bundles
including yet: as yet (fourth most frequent chunk) and yet another (third most
frequent chunk). In this part of the study, we have relied on data extracted from
PLECI and Europarl5 in both translation directions (as yet and yet another in
English source and target texts), which places us in a good position to assess the
added value of bidirectional corpus data.
We examined 135 bilingual concordances of as yet and found more than a
dozen French equivalents. These are listed in decreasing order of frequency in
Table 8 and illustrated in examples (22) to (24). Only two of these equivalents
(encore and déjà) are recorded in bilingual entries. This once again demonstrates
the need for translation corpus data to improve the translations included in
bilingual dictionary entries. Interestingly, the results show that it would be
advisable to look at both translation directions as some of the French equivalents
can only be uncovered by examining the French-to-English direction (i.e. as yet in
English target texts rather than in source texts). For example, examining cases
where as yet is found in English target texts makes it possible to unearth
equivalents such as aujourd’hui (see example 25) and jusqu’alors/jusque là, which
would be overlooked if only the English-to-French translation direction were
investigated.

14
Table 8. French equivalents of as yet in PLECI and Europarl5
Corpus equivalents
Original English,
translated French
Original French,
translated English
Total
encore
28
41
69
à ce jour/à ce stade/à la
date d’aujourd’hui
7
4
11
Pour l’heure/pour
l’instant/pour le moment
8
2
10
jusqu’à présent/jusqu’ici
8
1
9
aujourd’hui
0
8
8
toujours
5
2
7
encore … aujourd’hui/
aujourd’hui … encore
0
2
2
jusqu’alors/jusque-là
0
2
2
déjà
1
0
1
dorénavant
1
0
1
en l’état
1
0
1
mais encore
0
1
1
Other
6
8
5
13
Total
67
68
135
(22) In my view, only a very small number of somewhat equivocal issues
remain as yet unresolved. [Europarl5]
À mon avis, seul un nombre très restreint de questions quelque peu
équivoques restent à ce jour non résolu.
(23) We note that as yet there is no evidence to link depleted uranium and
ill health suffered either by troops or civilians. [Europarl5]
Nous notons qu'il n'y a jusqu'à présent aucune preuve du lien entre
l'uranium appauvri et les problèmes de santé dont ont souffert soit des
soldats, soit des civils.
6
The ‘other’ category includes cases of modulation, zero translation (in the English-to-
French translation direction, when as yet is not translated into French in the target text) and
addition (in the French-to-English translation direction, when as yet does not correspond to
any French item in the source text).

15
(24) As yet, western broadcasters are under little political pressure to cut
back. [PLECI news]
Pour l'instant, peu de pressions politiques poussent les stations
occidentales à réduire leur budget.
(25) Nous savons très bien qu'il n'existe aujourd'hui aucune étude de
toxicologie pour évaluer les conséquences de la dissémination d'OGM
dans l'environnement. [Europarl5]
We know full well that as yet there is no toxicology study that
evaluates the consequences of releasing GMOs into the environment.
The examination of yet another and its French equivalents brings to light another
interesting feature of bilingual entries. Yet another is listed in all three dictionaries
and is systematically translated into an adverb or an adverbial phrase (encore, de
plus), as shown in examples (26) to (28). However, the corpus data reveal many
excellent translations with other parts of speech. These include translations into
adjectives such as nouveau (new), énième (nth, umpteenth), supplémentaire
(supplementary) (see examples 29 to 31). Langlois (1996) provides similar
examples: while bilingual dictionaries translate the adverb automatically into the
French adverb automatiquement, the translation corpus shows that in 25% of cases
automatically is translated into something else, such as a phrase including the
adjective automatique (e.g. to pay dues automatically: payer les primes par
versement automatique). It can therefore be concluded that another benefit of
translation corpora is that they can help lexicographers free themselves from the
categorial bias, i.e. the tendency to translate a given source item exclusively into a
word of the same grammatical category in the target language (translate an
adjective into an adjective, an adverb into an adverb, etc.).
(26) she was yet another victim of racism [RC]
c'était une victime de plus du racisme
(27) yet another attack/question [HO]
encore une autre attaque/question
(28) yet another bomb [HU]
encore une bombe
(29) Nous n’avons pas besoin d’un énième rapport dirigiste. [Europarl5]
We don’t need yet another interventionist report.
(30) I heard the other day of yet another attack on the British farmer,
namely that the Government is about to impose a tax on fertilizer -
that is just as an aside. [Europarl5]
J'ai entendu parler l'autre jour d'une nouvelle attaque à l'encontre des
agriculteurs britanniques, à savoir que le gouvernement est sur le point
de prélever une taxe sur les engrais.

16
(31) Finally, I would make a plea on behalf of my group: let us not set up
yet another agency. [Europarl5]
Pour conclure, je voudrais introduire une demande au nom de mon
groupe : abstenons-nous de créer une agence supplémentaire.
Table 9 presents the complete inventory of the French equivalents of yet another
found in PLECI and Europarl5. As in the as yet dataset, we can see that some
equivalents, such as énième (and its variant nième), are more frequent in the
French-to-English translation direction. In fact, looking at the French components
of the Europarl5 sub-corpora used in this study (original French and translated
French), we find that énième is five times more frequent in French source texts
than in French target texts translated from English. Many of the studies described
in Stig Johansson’s 2007 volume point to similar differences between original and
translated texts. Clearly, looking at both translation directions holds great potential
for lexicographers: it can help them uncover all possible cross-linguistic
equivalents and avoid overlooking potentially interesting contrasts.
Table 9. French equivalents of yet another in English found in PLECI and
Europarl5
Corpus equivalents
Original
English,
translated
French
Original
French,
translated
English
Total
nouveau
35
8
43
de plus
13
18
31
autre
23
6
29
énième
6
12
18
encore
4
14
18
supplémentaire
7
4
11
Equivalents with less
than 5 occurrences
11
15
26
Other
7
9
6
15
Total
108
83
191
7
The ‘other’ category includes cases of modulation, zero translation (in the English-to-
French translation direction, when yet another is not translated into French in the target
text) and addition (in the French-to-English translation direction, when yet another does not
correspond to any French item in the source text).

17
5. Conclusion
Our study lends support to Stig Johansson’s observation that “dictionaries fall short
in the light of the evidence from bilingual corpora” (2007: 308). The use of corpus
data can help (1) improve the number of lexical bundles included in the bilingual
entry; (2) improve the number and accuracy of translation equivalents, especially
when the two translation directions are examined; (3) improve the prototypicality
and authenticity of the examples; and (4) avoid categorial bias in translations. Our
study has also brought to light an interesting difference between the two dictionary
halves: the phraseological coverage of the French-to-English part of the
dictionaries is quite limited as regards the number of bundles with encore listed in
the entries, while the most frequent bundles with yet are included in the entries in
the English-to-French part of the dictionary. In both directions, however, the
phraseological units, when included, can be poorly translated, and would benefit
from corpus-derived insights.
One of the encouraging results of our study is that, in the absence of large
balanced translation corpora, the systematic use of monolingual corpora can go a
long way towards ‘phrasing up’ bilingual dictionaries and even small translation
corpora bring their share of lexicographically relevant translation equivalents.
However, there is ample scope for improvement. As noted by Moon (2008: 333),
phraseological patterning varies according to genre. A corpus such as Label France
will only produce lexical bundles typical of the genre represented in the corpus (in
the same way as the Hansard Corpus was shown to be biased towards words and
lexical patterns that are typical of parliamentary debates (Langlois 1996)). In
addition, corpora like Label France or the BNC are static and cannot capture new
patterns of use. Those two caveats argue in favour of collecting and analysing
automatically generated web-based bilingual corpora, whose usefulness has been
convincingly demonstrated by Ferraresi et al. (2010). It is also important to
remember that corpus evidence is not everything. Moon (2008: 334) reminds us
that “lexicographers still have to use intuition and judgement in selecting,
interpreting, and setting out the evidence, rather than simply relaying it to the user
as quasi-scientific truth.” The combined use of bilingual corpus resources and
lexicographers’ bilingual expertise should ensure that phraseology, which has only
just begun to show its promise in bilingual lexicography, is at the root of major
lexicographic developments in the coming years. These advances would be a fitting
tribute to Stig Johansson’s inspiring legacy.
References
Corpora
The British National Corpus, version 3 (BNC XML Edition). 2007.
Distributed by Oxford University Computing Services on behalf of the
BNC Consortium. http://www.natcorp.ox.ac.uk/
Europarl corpus: http://www.statmt.org/europarl/
Label France corpus: http://www.uclouvain.be/en-cecl-labelfrance.html
Poitiers-Louvain Échange de Corpus Informatisés: http://www.uclouvain.be/en-
cecl-pleci.html

18
Dictionaries
Grand Dictionnaire Hachette-Oxford Français-Anglais English-French on CD-
ROM. 2003. Oxford University Press/Hachette Multimédia.
Grand Robert & Collins électronique français-anglais / anglais-français. DVD-
Rom Version 2.0. 2008. Le Robert/HarperCollins.
Harrap’s English-French Unabridged PRO Dictionary on CD-ROM. 2004.
Chambers Harrap Publishers.
Secondary sources
Altenberg, B. 1998. On the phraseology of spoken English: the evidence of
recurrent word combinations. In Phraseology: Theory, Analysis and
Applications, A.P. Cowie (ed.), 101-122. Oxford: Oxford University Press.
Atkins, B.T.S. 1996. Bilingual dictionaries: Past, present, future. In EURALEX
1996 Proceedings. Available online: http://www.euralex.org/
Barlow, M. 2002. ParaConc: Concordance software for multilingual parallel
corpora. In LREC 2002 Proceedings, Las Palmas, Gran Canaria, Spain.
http://www.mt-archive.info/LREC-2002-Barlow.pdf
Biber, D., S. Conrad & V. Cortes. 2004. If you look at…: Lexical bundles in
university teaching and textbooks. Applied Linguistics 25(3): 371-405.
Biber, D., S. Johansson, G. Leech, S. Conrad & E. Finegan. 1999. Longman
Grammar of Spoken and Written English. London: Longman.
Cappeau, P. 2010. Qu’est-ce qu’un bon exemple (oral)? In L’exemple et le corpus.
Quel statut? P. Cappeau, H. Chuquet & F. Valetopoulos (eds), 119-132.
Travaux linguistiques du CerLiCO 23. Rennes: Presses universitaires de
Rennes.
Ellis, N. C., R. Simpson-Vlach & C. Maynard. 2008. Formulaic language in native
and second-language speakers: Psycholinguistics, corpus linguistics, and
TESOL. TESOL Quarterly 42(3): 375-396.
Ferraresi A., S. Bernardini, G. Picci & M. Baroni. 2010. Web corpora for bilingual
lexicography: A pilot study of English/French collocation extraction and
translation. In Using Corpora in Contrastive and Translation Studies, R. Xiao
(ed.), 337-359. Newcastle upon Tyne: Cambridge Scholars Publishing.
Granger, S. & M. Paquot. 2008. From dictionary to phrasebook? In Proceedings of
the XIII EURALEX International Congress Barcelona, Spain, 15-19 July, E.
Bernal and J. DeCesaris (eds), 1345-1355.
Hanks, P. 2000. Contributions of lexicography and corpus linguistics to a theory of
language performance. In Proceedings of Ninth Euralex International
Congress, U. Heid, S. Evert, E. Lehmann & C. Rohrer (eds), I: 3-13. Stuttgart:
IMS Stuttgart University.
Hansen, M. M.-B. 2002. La polysémie de l'adverbe encore. Travaux de linguistique
44: 143-166.
Johansson, S. 1994. Towards an English-Norwegian parallel corpus. In Creating
and Using English Language Corpora, U. Fries, G. Tottie & P. Schneider
(eds), 25-37. Amsterdam: Rodopi.
Johansson, S. 2007. Seeing through Multilingual Corpora. On the Use of Corpora
in Constrastive Studies. Amsterdam & Philadelphia: Benjamins.
Kilgarriff, A., P. Rychly, P. Smrz & D. Tugwell. 2004. The Sketch Engine. In
Proceedings of Euralex 2004, Lorient (France), 105-116.
http://www.sketchengine.co.uk/

19
Koehn, P. 2005. Europarl: A parallel corpus for statistical machine translation. MT
Summit X, Phuket, Thailand, September 13-15, 2005, Conference
Proceedings: the tenth Machine Translation Summit, 79-86.
Langlois, L. 1996. Bilingual concordancers: A new tool for bilingual
lexicographers. Expanding MT horizons: Proceedings of the Second
Conference of the Association for Machine Translation in the Americas, 2-5
October 1996, Montreal, Quebec, Canada (Washington, DC: AMTA), 34-42.
Lew, R. 2004. Which Dictionary for Whom? Receptive Use of Bilingual,
Monolingual and Semi-Bilingual Dictionaries by Polish Learners of English.
Poznań: Motivex.
Lubensky S. & M. McShane. 2007. Bilingual phraseological dictionaries. In
Phraseology. An International Handbook of Contemporary Research, H.
Burger, D. Dobrovol’skij, P. Kühn & N.R. Norrick (eds), 919-928. Berlin &
New York: de Gruyter.
Moon, R. 2008. Dictionaries and collocations. In Phraseology: An
Interdisciplinary Perspective, S. Granger & F. Meunier, (eds), 313-336.
Amsterdam & Philadelphia: Benjamins.
Rundell, M. 1999. Dictionary use in production. International Journal of
Lexicography 12(1): 35-53.
Salkie, R. 2008. How can lexicographers use a translation corpus? In Proceedings
of The International Symposium on Using Corpora in Contrastive and
Translation Studies. Zhejiang University, Hangzhou, 25-27 September, R.
Xiao, L. He & M. Yue (eds). Available online:
http://www.lancs.ac.uk/fass/projects/corpus/UCCTS2008Proceedings/papers/
Salkie.pdf
Scott, M. 2008. WordSmith Tools version 5. Liverpool: Lexical Analysis Software.
http://www.lexically.net/wordsmith/
Sinclair, J. 2004. Trust the Text: Language, Corpus and Discourse. London:
Routledge.
Tono, Y. 2000. On the effects of different types of electronic dictionary interfaces
on L2 learners’ reference behaviour in productive/receptive tasks. In
Proceedings of EURALEX 2000.
Granger, S. & Lefer, M.-A. (2013). Enriching the phraseological coverage of high
frequency adverbs in English-French bilingual dictionaries. In Aijmer, K. &
Altenberg, B. (eds.) Advances in Corpus-based Contrastive Linguistics, 157-176.
Amsterdam & Philadelphia: Benjamins.