Oksefjell Ebeling, Signe and Jarle Ebeling. 2020. “Contrastive Analysis,

Tertium Comparationis and Corpora.” Nordic Journal of English Studies

19(1):97-117.

Contrastive Analysis, Tertium Comparationis and

Corpora

Signe Oksefjell Ebeling and Jarle Ebeling, University of Oslo

investigated based on the two types of corpora. It is concluded that a bidirectional parallel

corpus, relying on both comparable monolingual and bidirectional translation data, may

yield more robust insights into cross-linguistic matters than either of the two on their

own.

Keywords: contrastive analysis; tertium comparationis; comparable corpora; parallel

corpora; English; Norwegian; for NPgen sake

1. Introduction

over the years, including surface form, deep structure and translation, but

no consensus has been reached (see e.g. James 1980; Ebeling & Ebeling

2013a). That form, or surface structure, alone is a poor basis for

comparison if the aim is to study meaning and/or function is

acknowledged by Biber (1995) in a contrastive study of relative

constructions, where he notes that “despite the obvious structural

similarities between relative constructions in Somali and English, the

distribution of these features indicates that they are serving very different

functions in the two languages” (Biber 1995: 75).

Within corpus-based contrastive analysis, the types of available

corpora and parallel corpora of different kinds (e.g. uni-, bi- or

multidirectional). Johansson (2011) strongly believes in the advantages

of using parallel as opposed to comparable corpora:

The special advantage of parallel corpora is that they contain texts which are

intended to express the same meanings and have the same discourse functions in the

relevant languages. Using the source or the target language as a starting-point, we

can establish paradigms of correspondences. […] The most difficult problem in

using comparable corpora is knowing what to compare, i.e. relating forms which

have similar meanings and pragmatic functions. (Johansson 2011:126–127)

instance, says that “[t]ranslations, however good and near-perfect they

may be (but rarely are) cannot but give a distorted picture of the

language they present” (1996: 247).

Against this backdrop, we will conduct the same study based on the

two types of corpora (comparable and parallel) representing different

tertia comparationis. To our knowledge no study has hitherto

systematically carried out the same study on the basis of comparable vs.

bidirectional parallel corpora and juxtaposed the analysis in the way done

here. We hope to be able to demonstrate the impact the type of corpus

might have regarding insights gained about cross-linguistic equivalence.

Norwegian contrastive case study that is used as a testbed for carrying

out the same study on the basis of a comparable vs. a bidirectional

parallel corpus, demonstrating two different tertia comparationis: similar

forms and translation (Sections 5 and 6, respectively). Some concluding

remarks are offered in Section 7.

In this section, we offer a brief discussion of a select few approaches

within modern contrastive analysis (CA). It would be impossible to

describe and discuss the many ways in which to conduct a CA that have

been proposed over the past 60 years or so in this short article. We refer

the interested reader to some important works, and references therein, for

more comprehensive and detailed accounts: Aijmer et al. (1996),

Altenberg & Granger (2002), Johansson (2007), Hasselgård (2010),

tested. Chesterman (1998) advocates such an approach and recommends

the following steps, where the starting point, and first step, is a

(dis)similarity of any kind between phenomenon X in language A and

phenomenon Y in language B. A phenomenon can in principle be almost

anything: a situation, a gesture, a construction, a category or an item.

Based on the similarity, the following question is posed (step 2): what is

the nature of the similarity (form, meaning, function)? The third step

involves the description of the relationship between X and Y in the

compared languages, or as is more often the case, the relationship

between X in language A and Y1, Y2, Y3, etc. in language B (see e.g.

Dyvik (1998) and his semantic mirrors). Based on the outcome of step 3,

point of departure is any observed quantitative difference between

original and translated texts of the same type and size (e.g. Johansson

2007: Ch. 4). Yet another approach is to register zero correspondences in

a parallel corpus (e.g. Ebeling & Ebeling 2013b), that is where an item in

one language does not seem to have a (direct) correspondence in the

other language, and ask what the reasons for these non-correspondences

frame, pattern or construction in language A and record and analyse all

the corresponding items/units in language B, i.e. create a list of

translation paradigms, to describe similarities and differences at the

levels of, for instance, syntax and/or semantics (e.g. Johansson 2001 on

seem and its Norwegian correspondences). In all of these cases, close,

qualitative scrutiny of the differences should be part of the study. Note

that several of the more exploratory approaches require access to a

• the focus on immediate applications is toned down;

• the contrastive study is text-based rather than a comparison

of systems in the abstract;

languages or simply by comparing language systems, as found in e.g.

monolingual grammar books (see James 1980; Mair 2018).

By modern corpus-based contrastive analysis we understand research

which is grounded in empirical (corpus) data in two or more languages,

with the challenges and limitations that this entails, for instance with

regard to the type and amount of data available. This is in fact one of the

main challenges of modern contrastive research. We now turn to these

challenges in a comparison of comparable and parallel corpora.

these two main types of corpora.

Figure 1. Corpora used in CA (inspired by Aijmer 2008; Johansson 2007; McEnery &

Xiao 2007)

As shown in Figure 1, in a comparable corpus, original texts in two or

more languages—L1, L2, L3 etc.—can be compared to each other,

pairwise (L1 vs. L2; L1 vs. L3; L2 vs. L3) or one can compare L1 vs. L2

& L3, L2 vs. L1 & L3, etc.

When it comes to parallel corpora, these can be further sub-divided

comparable, such a corpus gives the researchers the possibility of starting

the analysis in either language and in either original or translated texts

and, more importantly, to control for translation effects and source

language shining through (Johansson 2007; Teich 2003).

Before we present the case study, it is useful to keep the pros and

cons of the main types of contrastive corpora in mind. These are outlined

in Table 1, which is inspired by Altenberg & Granger (2002), Johansson

items in original and translation, that is, a sound common ground on

which to build a direct comparison is present. This common ground has

been claimed to become even firmer and more objective if you have a bi-

or multidirectional parallel corpus, which is also in effect a comparable

corpus in that it contains original language data in all the languages

compared (cf. Johansson & Hofland 1994; Johansson 2007; Aijmer

2008).

the same phenomenon in the two different types of corpora.

The case study takes a previous contrastive investigation as its starting

point. Ebeling & Ebeling (2014) present a corpus-based contrastive

analysis of two similar-looking patterns in English and Norwegian,

namely for * sake and for * skyld, where the asterisk stands for a genitive

noun.

The present study is in many ways an experiment in the sense that an

attempt will be made to carve up the original study in a different way.

bidirectional parallel study” (Section 6).

The corpus used for the occasion is the English-Norwegian Parallel

Corpus+ (ENPC+) (Ebeling & Ebeling 2013a), which is structured

according to Johansson’s parallel corpus model; thus it contains both

comparable and bidirectional translation data in one (Johansson &

Hofland 1994). The ENPC+ is an extended version of the fiction part of

the English-Norwegian Parallel Corpus (ibid.) and contains around 1.3

million running words of contemporary English and Norwegian fiction

texts and their respective translations into the other language.

In other

words, it is a balanced corpus amounting to rougly 5.2 million words in

total.

Ebeling & Ebeling (2014: 192) start by outlining the meanings of the for

* sake pattern according to Oxford Dictionaries Online (now

Lexico.com): (a) for the purpose of, (b) out of consideration for or in

order to help or please someone, and (c) to express impatience,

annoyance, urgency, or desperation. It is argued that the Norwegian

pattern incorporates the same meanings, as attested in examples (1)–(3)

from Norwegian original texts with their translations into English.

(2) En gang sa hun at hun gjorde det for pappas skyld og for oss, så vi

(3) Si noe, for Guds skyld! (LSC2)

Say something, for Godʼs sake! (LSC2T)

fairly mild expletives.

The other overlap in the correspondences is arguably more curious,

as the two English expletives—one from the sexual domain (for fuck’s

sake) and one from the celestial (for God’s sake)—both are found to

correspond to the diabolical faen ‘the devil’. This is in fact the most

common Norwegian correspondence of the English expletive overall.

Of the two remaining, relatively common non-congruent

6.1 Summary of the Bidirectional Parallel Study

In addition to uncovering different preferences of use in English and

Norwegian (including different extended-units-of-meaning) on the basis

of both comparable and parallel data, we have been able to demonstrate

that access to, and analysis of, bidirectional translation data means that

we, through translation as TC, can offer corresponding, and arguably

more equivalent, expressions of the for * sake / for * skyld pattern in the

other language. For example, typical Norwegian expletives

the English expletive (for svingende) may be considered less equivalent

than the other correspondences since it was only attested in the translated

material. Moreover, the observation from the comparable study that the

patterns seem to be cross-linguistically most similar to each other in the

consideration use was not only confirmed on the basis of the

the large proportion of congruent correspondences.

We believe that we gain a deeper understanding of the items

compared when using a bidirectional parallel technique as it offers a

more comprehensive contrastive account of the patterns compared. It

should also be mentioned that a bidirectional parallel study like this one

could, and perhaps even should, serve as a starting point for further

investigations that draw on much larger comparable monolingual

investigate cross-linguistic correspondence and equivalence on the basis

of a comparable corpus, researchers typically take predefined items or

categories as their starting point, emerging from a perceived similarity

(or even dissimilarity in some cases) in two or more languages (e.g.

causative constructions). The CA in these cases may be based on the

researcher’s bilingual knowledge, dictionaries, grammars, and earlier

research of previously identified causative constructions.

Parallel bidirectional corpora, on the other hand, add the dimension

of having access to a pool of translators’ bilingual competence, and what

choices the translators make in similar linguistic situations. This enables

investigations of both predefined and undefined items through perceived

methods by performing two versions of the “same” cross-linguistic

study. The rationale behind this exercise was to see to what extent it

would lend support to the view held by some contrastive linguists that

correspondence in translation is a good tertium comparationis if applied

to carefully structured bidirectional parallel corpora. It was found that a

bidirectional study arguably yielded more robust cross-linguistic results

bidirectionality of such corpora also ensures that the researcher can

control for translation effects, i.e. potential translation-specific features.

It should be stressed that the idea of using bidirectional data for

contrastive analysis is not our own. We are indebted to Stig Johansson, in

particular, for devising a corpus integrating comparable monolingual and

bilingual translation data within the same model, as first seen in the

English-Norwegian Parallel Corpus (ENPC) (Johansson & Hofland

also as providers of more extensive data sets. Indeed, in a recent

publication by Granger (2018: 183), a methodological framework termed

Contrastive Translation Analysis is introduced, where a “multi-corpus

empirical basis for corpus-based crosslinguistic studies” is called for. In

this framework, both comparable and parallel corpora have a place and,

in addition, learner corpora of the languages being compared (ibid. 190).