Analysing Parallel Texts with ParaConc

Analysing Parallel Texts with ParaConc
Much of the current research on parallel corpora
concerns the problem of automatic alignment of
two texts that are translations of each other (Gale
and Church 1994, Kay and Roscheisen 1994, Johansson and Hofland 1993). This paper, however,
focusses on the analysis of aligned parallel corpora rather than on the aligning process itself.
In order to analyse a parallel corpus a suitable text
analysis program is needed. ParaConc is a simple
parallel text concordance program available in
Macintosh and Windows versions, which was
created by the author as a tool for linguistic research. This program allows the user to search for
a word or phrase, in the way typical of concordance programs. However, the result of the search is
displayed in two windows rather than one. The
topmost window displays numbered lines containing each instance of the search term in the first
language, along with its context. The lower window displays numbered sentences in the second
language which correspond to the text displayed
in the first language in the upper window. The
results of a search can be sorted, printed or saved.
To obtain a list of words from each text that
correspond, as illustrated below for English line
and French ligne, the results of a search are first
saved as a text-only file and then loaded into a
word-processor for further formatting.
The use of parallel corpora presents very interesting research opportunities in a variety of disciplines including linguistics, literary studies, translation, and language teaching. While these
different areas may be touched upon, the focus of
the present paper is on the use of parallel corpora
in linguistic analysis. This project is similar in
spirit to a variety of parallel corpus projects such
as Intersect, Contragram, ENPC, and TRIPTIC,
among others.
Taking a language to consist of form-meaning
links, what we have in parallel corpora are two sets
of form-meaning linkings, one for each language.
And since the two texts are translations, the meaning part can be assumed to be approximately the
same in both texts. Thus we are able to see how
two different languages encode equivalent meanings. The art of translation is undeniably complex
and involves many different kinds of processes,
but we can consider three main aspects of translation, namely, language particular encodings of (i)
event structure, (ii) discourse structure, and (iii)
lexis. Each of these areas can be profitably analysed using parallel corpora.
1. Event Structure
Event structure simply refers to those actions occurring in the world that are of interest to humans,
such as a transitive event in which one object acts
on another object in some way, which is typically
encoded using a transitive clause. Since we can
assume that the translations are about the same
events, we can use parallel corpora to examine
how languages code events in general, in other
words, how aspects of an event are expressed
grammatically or lexically in different languages.
An objection that could be raised here is that the
particular choices made by a translator will introduce distortions into the data. It is true that some
apparently random choices occur in translations,
but the accretion of motivated translation choices
allow the general patterns to be perceived using a
concordance program.
For example, we can examine the coding of causative events in English and French by searching
for the lemma make and examining patterns such
as "X makes Y do Z" and then observing the
patterns used in French to refer to these same
events. A concordance search reveals that causative make in English covers a wide variety of
situations, for instance, causing a change in state
(make something possible), causing someone or
something to perform an action (make a dog go
away), and causing some kind of transformation
expressed as make followed by two contiguous
noun phrases (make John the president). Having
searched for English causative constructions involving make, we can investigate how these different causative event structures are coded in
French. And, in fact, we find a rather different set
of patterns for French. For the construction of the
type make John president, the equivalent occurs
in French with faire in most cases. However, the
corpus data shows that other causative uses are
often not translated by faire in French. A variety
of constructions are used instead, including verbs
such as rendre, as shown in (1).
(1) a. The American blockage makes life
very difficult for us.
Le blocus rend nos conditions de vie
trés rudes.
On the other hand, uses of make expressing a
causative event in which an agent acts on an animate causee to bring about an event are more
likely to be translated with faire, as exemplified in
(2).
(2) a. It is a behaviour which makes you think
of France...
Un comportement qui fait penser à la
France ...
(2) b. ... their parents had made them lose
their French nationality.
... leurs parents, ...., leur ont fait perdre
la nationalité française.
This example shows how ParaConc can be used to
investigate fairly subtle cross-linguistic distinctions in the expression of causative events.
2. Discourse structure
Parallel corpora can also be used to highlight the
way in which different languages transform the
bare bones of event structure into discourse structures appropriate for each language. There are
many interesting questions related to the structuring of discourse in different languages and the use
of parallel corpora offers one avenue of research
in this area. As an example, we can consider how
English and French discourse signals the fact that
two events occur concurrently (or are alike in
some other way). In English the conjunction while
is used both to link clauses that refer to events that
overlap in time and to indicate that the speaker is
contrasting two events. The two types, the temporal use and the contrastive use, are shown in (3)
and (4).
(3) a. That means that while we’re shooting
one film we can start dreaming about
the next.
(3) b. That’s the way to get the economy
going again while at the same time
discouraging looters.
(4) a. While it never misses an opportunity to
blame the Socialists for the worsening
job situation, the right appears to be just
as helpless in the face of rising unemployment.
(4) b. While saleswomen remain as surly as
ever, shop windows have become
much more attractive.
Using parallel texts, it is possible to search for
English while and investigate how temporal and
contrastive structures are represented in French
discourse. Searching for while produces a variety
of equivalent items in French including: tout en,
alors que, tandis que, pendant que, contre, and si,
among others. To provide a complete analysis of
these conjunctions it is necessary to examine the
results of the search in some detail and also to
examine the translations in English of the different
expressions: tout en, alors que, etc. One result that
we can identify is that French si is used to indicate
a contrastive meaning. Thus the equivalent sentences to (4) are those given in (5), which use si for
while.
(5) a. Si elle ne manque aucune occasion de
verser au débit des socialistes la détérioration de la situation de l’emploi, la
droite paraît tout aussi désarmée devant
la montée du chômage.
(5) b. Si les vendeuses sont toujours aussi peu
aménes, les vitrines, en revanche, se
font plus alléchantes.
3. Lexis
ParaConc has several uses in investigating the
meaning of lexical items and collocations in two
languages. The program can take advantage of the
information-on-demand aspect of concordance searching and provide equivalences that may be
incompletely captured or not captured at all in
bilingual dictionaries. For example, a parallel corpus based on computer texts will allow the user to
see how modern computer terms such as information highway, email, and home shopping are being
translated. Thus a parallel corpus can be used to
reveal both the latest usage and also the variation
in usage that occurs.
Rather than explore these lexicographic uses of
ParaConc, in this final section I will again pursue
a linguistic investigation and indicate how ParaConc can reveal metaphorical and other extensions of a concept occurring in two languages. The
word line in English, for example, has a variety of
uses, some of which are based on extensions of the
prototypical meaning. Tied in with these extensions is the existence of certain collocations such
as hard line, firm line, etc. Some of these extensions are also present in French; others are not. We
find, for instance that the in line with uses do not
appear to have a French equivalent based on ligne.
In (6) a small sample of correspondences is given.
(Examples of ligne and their equivalents in English are omitted from this abstract for reasons of
space.)
(6)
a line une réplique
communication line ligne de communication
cultural line ligne culturelle
dedicated line ligne spécialisée
26
a democratic line une ligne démocratique
dividing line ligne de partage
dividing line ligne de fracture
drain line canalisation d’écoulement
took a firm line apporté un soutien d’une
fermeté
the following lines cette formule
the front line au front
front line front
hard-liners l’intransigeance des
in line with á l’image de
kept in line on encadre
not in line with pas correspondre au
In line with Comme l’indiquait
in line with s’inscrit dans
into line with en accord avec
line positions
our line of conduct notre conduite
our line notre principe
the poverty line le seuil de pauvreté
Given these sets of data, it is possible to map out
how the semantic domain of line and ligne resemble each other and how they differ in terms of
semantic extensions and usages. The undertaking
of this kind of investigation can play a part in
linguistic investigations of grammaticalisation
(Hopper and Traugott 1993) and of the study of
general constraints on form-meaning mappings
(Barlow and Kemmer 1994).
4. Conclusion
These analyses provide an illustration of how the
common content of parallel corpora can be exploited to gain linguistic insights into the structure and
function of languages. The technique of investigating pairs of languages is promising for a variety
of research areas. One advantage is that a two-way
analysis of a domain, from language A to language
B, and from language B to language A provides
clues to the different meanings/uses of each
language form.
In sum: in this paper I describe the analysis of
parallel texts using ParaConc, a parallel concordancer, and outline some fruitful areas of corpusbased research that are opened up by the use of
such a program.
References
Barlow, M. To appear. Parallel Texts for Linguistic Analysis. In M. Barlow and S. Kemmer
(eds) Usage-Based Models of Language.
Barlow, M. 1995. A Guide to ParaConc. Athelstan: Houston.
Barlow, M. and S. Kemmer. 1994. A Schema-based Approach to Grammatical Description. In
S. Lima, R. Corrigan and G. Iverson (eds) The
Reality of Linguistic Rules. Amsterdam: Benjamins.
Gale, W. and K. Church. 1994. A program for
Aligning Sentences in Bilingual Corpora. In S.
Armstrong (ed) Using Large Corpora. MIT
Press: Cambridge.
Hopper, P. and E. Closs Traugott. 1993. Grammaticalization. Cambridge: CUP.
Johansson, S. and K. Hofland. 1993. Towards an
English-Norwegian parallel corpus. Paper
from the Fourteenth International Conference
on English Language Research on Computerized Corpora, Zürich, May 19-23, 1993. In U.
Fries, G. Tottie, and P. Schneider (eds.), Creating and Using English Language Corpora.
Rodopi: Amsterdam.
Kay, M. and M. Roscheisen. 1994. Text-Translation Alignment. In S. Armstrong (ed) Using
Large Corpora. MIT Press: Cambridge.
Moon, R. 1987. The Analysis of Meaning. In J.M.
Sinclair (ed) Looking Up. Collins: London.
Noel, Jacques. 1992. Collocation and Bilingual
Text. In G. Leitner (ed) New Directions in
English Language Corpora. Mouton de Gruyter: Berlin.