Abstract

In this study, term variation pertains to the different ways in which specialised knowledge in the environmental domain is expressed in English institutional texts by means of terminological designations. Intralingual variation pertains to term variation within one language, interlingual variation to the possible ways in which a given English term is translated into Dutch and French target texts.

In many descriptive terminology studies, it has been pointed out that terminological variants frequently occur in different types of specialised text genres and that their presence can be motivated on different grounds (Daille 2005; Freixa 2006; Condamines 2010; Tercedor Sánchez 2011). At least two important views result from these observations: a) a view that term variants appearing in source texts should not simply be ignored by translators for the sake of terminological consistency and precision and b) a view that the different terms to express specialised knowledge as well as their possible translations should be represented in special language resources for translators, taking into account different contextual factors that may affect the choice for a specific term or translation.

Based on these views, we will present a new type of translation resource that was compiled on the basis of a corpus of source and target texts. The resource covers a set of English term variants and their French and Dutch equivalents retrieved from a trilingual parallel corpus of institutional texts. Each term occurrence encountered in a source text is combined with its equivalent in the corresponding target text to form a term-based translation unit (TU). Each TU in the resource is marked by a set of semantic and text-related properties. In this article, we will discuss why and how the resource was created, present results and reflect on how it can be used by translators.

Keywords: environmental terminology, institutional translation, term variation, translation resources, translation unit

1. Introduction

In the field of terminology, the conviction reigned that terms should be used unambiguously to refer to clearly delineated concepts (Wüster 1979/1991; Felber 1984; Picht and Draskau 1985) in order to arrive at unambiguous communication. This view dominated terminology theory for several decades, resulting in the prescriptive view on terminology (Cabré 2003).

In prescriptive terminology, the concept – which is the starting-point of terminological analysis – is identified by means of an authorised term. Within this framework, alternative ways of referring to a concept (for instance by means of lexical variants) should be restricted because it is believed that these alternative expressions may hamper the exchange of information in specialised communicative settings (Gerzymisch-Arbogast 2008).

Descriptive terminology approaches have jointly contributed to a more pragmatic view of the relationship between terms and concepts (Gaudin 1993; Cabré 1995; Sager 1998; Condamines 1995; Temmerman 2000; Diki-Kidiri 2001; Collet 2004; Faber 2009). These studies have demonstrated that term variation – i.e. the different ways in which specialised knowledge can be expressed by means of terms – is a natural phenomenon in special languages, which can be motivated on different grounds (Bowker and Hawkins 2006; Freixa 2006).

Translating terminology poses important challenges for translators. On the one hand, translators need to acknowledge the role that terminology fulfils in communicating specialised knowledge in a precise way. On the other hand, they need to be aware of the fact that terminology use is conditioned by several contextual factors. For translators, it is therefore important to know what linguistic options (i.e. terminological variants) are available in languages for expressing specialised knowledge and to understand how these options can be used (i.e. how they function) in specific communicative settings (Hatim and Mason 1990; House 2001).

Translators often consult bi- or multilingual translation resources (e.g. bilingual glossaries, parallel word lists, specialised translation dictionaries, terminological databases, etc.) to find solutions to certain translation problems. However, such structured resources never fully represent the wealth of options available in language (Gerzymisch-Arbogast 2008). Furthermore, by separating terms from their natural environments (i.e. the texts), a lot of valuable information on which translations decisions are based is lost. This is why translators also often resort to unstructured resources: i.e. original texts (written in the source and target languages) or previously translated texts to acquire more insight into the specific uses of terms in various situational contexts.

The present study involves a reflection on how terms and equivalents recorded in multilingual terminological databases (multilingual termbases or MTBs) can be extended with intra- and interlingual variants retrieved from source and target texts. For this purpose, a new type of structured translation resource will be proposed, resulting from a method for identifying intralingual variants and their translations in parallel texts (Kerremans 2011; 2014). The parallel corpus used for this study is comprised of English source texts (related to environmental topics) and their translations into Dutch and French.

In the translation resource, each occurrence of an English term (variant) in the parallel corpus is combined with its French or Dutch equivalent to form a (term-based) translation unit (TU). Each TU is further categorised according to semantic and contextual criteria, which will be discussed in more detail in Section 3.

Throughout this article, several questions will be raised pertaining to the underlying motivation for creating this resource (see Section 2), the content of the translation resource (see Section 3), the methodology for building up the database (see Section 4) and the way the content of the resource could be presented to translators (see Section 5). A conclusion as well as a brief reflection on future work will be presented in Section 6.

2. Why a new type of translation resource is proposed

Our proposal to develop the aforementioned translation resource results from our observations in two descriptive comparative studies of terminological variation applied to European institutional translation (Kerremans 2014). The subject field chosen as a case study is the environmental domain. Within this domain, four main subject areas were selected for our purposes: biodiversity, climate change, environmental pollution and invasive alien species.

The purpose of the first study was to examine how intra- and interlingual variation appears in a corpus of parallel texts. To this end, a multilingual corpus was created based on European institutional source texts (in English) and their translations into French and Dutch.

A general observation in this study was that terminological variation appearing in the source texts of our corpus also tends to be reflected in the translations. In other words: decisions with respect to translations of concepts into the target languages (Dutch and French) were very much influenced by lexical/terminological choices in the English source texts. For a more thorough discussion, we refer to Kerremans (2011; 2014).

In the second study we compared the language data retrieved from the parallel corpus (in the first study) with data retrieved from the EU multilingual terminology base IATE[1]. The purpose of the study was to examine how intra- and interlingual variation is accounted for in the European terminology base and whether the terminological options in the source language as well as the translation options would also be represented in the termbase. Given these objectives, a descriptive study was conducted on the basis of a collection of about 1140 IATE terminological records. This collection contained terms and variants referring to concepts (see Section 3) that were also encountered in the parallel corpus of the first comparative study, allowing for a comparison between the two language resources.

An important outcome of this study was the observation that the different terminological options in which specialised knowledge was expressed in the source texts were not entirely covered by the IATE terminology base. Apart from that, we also observed that the parallel texts featured much more translation options as opposed to the possible equivalents of a given source term that we found in the IATE data collection. For a more detailed discussion of these observations, we refer to Kerremans (2014).

Based on the aforementioned observations, we propose a new type of language resource for translators in which intra- and interlingual variants derived from parallel texts can be used to supplement the information appearing in multilingual terminology bases (see Section 5). As was stated earlier (see Section 1), the translation resource is comprised of term-based translation units (TUs) that are semantically and contextually structured.  

3. How language information is structured in the translation resource

We define a term-based translation unit (TU) as a bilingual text segment, extracted from a source text (ST) and its target text (TT), in which

• the source language (SL) segment is a terminological expression and
• the target language (TL) segment is the translation of the selected SL segment.

Consider the following text sample:

• ST: “Invasive Alien Species” are alien species whose introduction and/or spread threaten biological diversity.
• TT: L’expression “espèce exotique envahissante” s’entend d’une espèce exotique dont l’introduction et/ou la propagation menace la diversité biologique .

In this example, taken from the English and French versions of a (European) Commission staff working document[2], the term invasive alien species in the ST and its translation espèce exotique envahissante are combined to form the term-based translation unit: invasive alien species – espèce exotique envahissante.

Term-based translation units are the primary building blocks of the resource presented in this article. Before entry into the translation resource, each TU is contextually and semantically structured. This means that certain properties are added to the TU to be able to support specific search queries of the language data in a later stage (see Section 5).

The TU is said to be contextually structured when different properties of the bitext have been assigned to it. Examples of such properties are text type (e.g. EC Staff Working Document), text source (e.g. European Commission), language pair (e.g. English-French) and text topic (e.g. invasive alien species). Some properties are used to specify the exact location of the source term and its equivalent in the bitext. Other contextual (or text-related) categories are used to describe the situational context (or register) in which the TU was found.

The TU is said to be semantically structured when the English term in the TU is classified according to the concept to which it refers in the source text. Semantic categorisation is a prerequisite in order to retrieve – at a later stage (see Section 5) – all term variants lexicalising the same concept. This process requires that term variants referring to the same concept are labelled with a unique identification code. This manually created code or label is called a cluster label. An example of such a label for the translation unit above is invasive_alien_species. Terms extracted from the source texts that carry this label appear in the same cluster of terminological variants (Kerremans 2011).    

Before we move on to the question of how the translation resource is created, an important remark has to be made with respect to the way concepts are perceived in the present study because this has important theoretical implications for the treatment of clusters of term variants (cf. supra).

Different views have been expressed in terminology studies on the nature of concepts as well as on their specific relation to terms. In prescriptive approaches, the focus is on establishing clearly delineated concepts which are considered necessary for efficient communication of specialised knowledge (Wüster 1979/1993; Felber 1984; Picht and Draskau 1985). As a result, term variants referring to the same concept are believed to have the same meaning.

In descriptive terminology studies, this prescriptive view on concepts has been criticised for being unrealistic. As an alternative to the prescriptive view, several proposals have been formulated emphasising embodied, situated and dynamic aspects of cognition (Zawada and Swanepoel 1994; Cabré 1995; Condamines and Rebeyrolle 1997; Kageura 1997; Temmerman 2000; Diki-Kidiri 2001; Faber 2009).

As a result, concepts are believed to be fuzzy categories which are given shape and constantly change in contexts of social interactions. On the level of discourse, which features multiple perspectives and opinions, the concept is an amalgam of diverse conceptualisations. These different conceptualisations contain overlapping properties and can therefore be attributed to a prototypically structured unit of understanding (Temmerman 2000). We believe that in a context of translation, it makes sense to talk about prototypically structured units of understanding rather than concepts. This is because transferring meaning from one language into the other is determined by the translator’s understanding and interpretation of the message in the source language.

Starting from the unit of understanding (instead of a clearly defined concept) has important implications for the way terminological variants are perceived in cognitive studies of terminology (see e.g. Temmerman 2000; Faber 2009). In our own study of term variation, we do not claim that the English terms appearing in the same cluster of variants share exactly the same meaning. Obviously, their meaning is conditioned by the situational contexts in which they appear. What these terms have in common is the fact that they are used to refer to the same (prototypically structured) unit of understanding in a specific situational context. How we collected these terms will be explained in the next section.

4. How the translation resource was developed

The translation resource resulting from our study of term variation in institutional texts, was developed on the basis of predefined set of units of understanding (concepts) in the environmental domain. This list of units of understanding was created on the basis of the Environmental Terminology and Discovery Service or ETDS glossary, a large multilingual terminology base maintained by the European Environment Agency (EEA). This resource contains almost 10,000 English terms and their translations into several other languages.[3]

By first automatically determining the number of occurrences of each English ETDS term in the English source texts of our corpus, we were able to filter out terms appearing several times in our corpus. Based on this list of terms, we randomly created a set of 240 units of understanding that formed the basis for the descriptive studies of term variation (see Section 2).

For each unit of understanding, a unique label (i.e. the so-called ‘cluster label’) was created in order to manually identify and mark term variants in the English source texts. Terms appearing in the source texts that are annotated according to the same label appear in the same cluster of terminological variants (see previous section).

Our method for identifying term variants in the source texts, was presented in Kerremans (2011). The method is based on analyses of lexical chains. According to Rogers (2007: 17), a lexical chain consists of “cohesive ties sharing the same referent, lexically rather than grammatically expressed”. In other words: terms appearing in the same lexical chain were marked with the same cluster label.

This is illustrated by means of the following text sample taken from the Commission’s staff working document that we also mentioned in an earlier section of this article (see Section 3). In this example, the term variants appearing in the same lexical chain (referring to the unit of understanding invasive_alien_species) are marked in bold:

“Invasive Alien Species” are alien species whose introduction and/or spread threaten biological diversity [...]. The Millennium Ecosystem Assessment revealed that IAS impact on all ecosystems [...]. The problem of biological invasions is growing rapidly as a result of increased trade activities. Invasive species (IS) [...] negatively affect biodiversity [...]. IS can cause congestion in waterways, damage to forestry, crops and buildings and damage in urban areas. The costs of preventing, controlling and/or eradicating IS and the environmental and economic damage are significant. The costs of control, although lower than the costs of continued damage by the invader, are often high.

The result of this analysis is a lexical chain composed of seven lexicalisations. Five of these occurrences are unique terminological units:

invasive_alien_species =>

1. Invasive Alien Species
2. IAS
3. Invasive species
4. IS
5. IS
6. IS
7. invader

Once the entire source text was analysed, we manually retrieved for each selected term in the source text its corresponding lexicalisation in the target text. For more details on this procedure, we refer to Kerremans (2011).

To be able to carry out this kind of analysis in an efficient way for about 240 cluster labels, we developed several automated procedures in Perl.[4] The module for supporting the analysis of the source texts, for instance, relies on three external resources to support the process of identifying SL terms and adding these to the proper clusters: a dataset of clusters that continuously grows as more variants are retrieved from texts, a list of filtering rules and a dictionary of lemmatised forms. For more details on these automated routines, we refer to Kerremans (2014).

5. How the translation resource can be exploited

The translation resource resulting from our comparative studies (see Section 1) contains about 18,600 semantically and contextually structured translation units for the language pairs English-French and English-Dutch. In this section, we tentatively describe how the resource might be used by translators (see Section 5.2). But before we provide such a description, it is important to first clarify in what respect our translation resource differs from multilingual terminology bases (in particular: IATE), on the one hand, and translation memories, on the other hand (see Section 5.1).

5.1. How our resource differs from termbases and translation memories

The purpose of this first subsection is to explain the potential advantage that our resource offers to translators, when compared to the information that can be retrieved from termbases and translation memories.

In conventional multilingual terminology bases, intralingual terminological variants and their translations appear on concept-oriented records. This particular way of structuring terminological data has its roots in prescriptive terminology in which terms are merely viewed as labels assigned to clearly delineated concepts (see Section 3). The number of variants that are displayed on these records is usually restricted and never fully covers the wealth of terminological options that can be found in authentic texts. This is because multilingual termbases tend to be limited to a representation of “the ‘langue’ or ‘norm’ level of language and not the individual actual text level” (Gerzymisch-Arbogast 2008: 20). Moreover, these resources very seldom provide information concerning the specific origin or the specific use of each recorded term. Our descriptive analysis of IATE terminological records (see Section 2) revealed, for instance, that in many cases this pragmatic information about terms is left unspecified.

The consequence is that when translators consult the IATE terminological database (and actually find corresponding matches for specific search terms) they do not necessarily get to see a solution in the target language that is directly applicable to the specific text that they are translating. This is for instance because the equivalent found in the database was taken from a document that belongs to another ‘register’ (e.g. a legal document vs. a Commission’s fact sheet) or source (e.g. a term entered into IATE by a translator/terminologist from the Commission vs. a term entered by someone from another EU institution) as the text being translated.

In this case, translators need to see examples of actual term use in order to check the appropriateness of the given term in the translation context. To this end, they can rely on multilingual corpora comprised of original texts in the source and target languages as well as on parallel corpora or translation memories.

A translation memory is basically a corpus of bitexts, usually aligned on the level of the sentence. The corpus is converted to a standardised exchange format (TMX)[5]  so that it can become searchable by means of a Translation Memory System (TMS). In translation memories, translators can find actual occurrences of terminology and translation choices because these are comprised of authentic texts.

The advantage of translation memories is that translators can actually search for translation options in texts that are similar to the ones they are translating. An important disadvantage is the fact that the search facilities provided by a TMS still do not go beyond finding simple text string matches. This means that only concordances (i.e. linguistic contexts featuring the search term) are shown in which the search string literally or partially appears.

A weakness of this type of searches is that no variants can be taken into consideration when looking for potentially interesting contexts. Suppose the search term alien invasive species appearing in a source text does not occur so often in the translation memory, in comparison to other terminological variants such as invasive exotics or pest organisms, the search results would turn out poor for this term in comparison to the many solutions that could have been found in the TM if the search was based on clusters of term variants.

Taking into account the strengths and weaknesses of termbases and translation memories, we propose a translation resource that stands in between these two language resources, as it combines features of both. Termbases allow us to indicate whether several terms refer to the same unit of understanding (and should therefore be treated as term variants). This important feature is also present in our translation resource in which term variants are marked by the same cluster label. The advantage of using translation memories is that these resources show translators how terms were actually translated in authentic texts. But the lack of a semantic specification of the translation segments do not yet allow these repositories to be exploited in more advanced ways (Moorkens 2012; Reinke 2013). Adding a semantic specification to the term-based translation units will enable users to filter out those contexts that are semantically related.

5.2. How the language data can be visualised

As an alternative to traditional ways for representing language data in dictionaries (e.g. on the basis of alphabetically sorted lists, tables or matrices), we propose to present the language data in graphs, allowing for a flexible and dynamic visualisation of data that may be connected to one another in several ways (see further).

The motivation for this particular visualisation is based on the Hallidayan premise that each choice (variant) in the language network acquires its meaning against the background of other choices which could have been made (Eggins 2004). The choices are perceived as functional: i.e. they can be motivated against the backdrop of a complex set of linguistic, situational and cognitive factors (Kerremans 2014). Changing the contextual conditions causes direct changes in the network of terminological options that are shown to the user.

An additional interesting aspect of the graph representation is that it supports the theoretical model of prototypically structured units of understanding (see Section 3). The translation resource holds information about the frequency of occurrence of each term variant, translation equivalent and translation unit retrieved from the parallel corpus. That information can be used in a visualisation module to make a distinction between typical and uncommon patterns for a given context configuration.

What we propose is that term variants, translation equivalents and/or translation units that frequently occur in a particular register would move to the centre of the graph, while uncommon patterns would move to the periphery. In this way, translators will immediately be able to distinguish between core variants (i.e. variants that are frequently encountered within the selected collection of texts) and peripheral variants (i.e. variants that were only sporadically encountered). They will also be able to immediately infer the typical ways of translating a given term (in a given register).

An example illustrating this idea is shown in Figure 1. The example shows a network of linguistic (terminological) options for the unit of understanding greenhouse_gas_emission_reduction. Apart from the English term variants, the network also shows one particular Dutch equivalent found in the corpus data: i.e. emissiereductie.

Figure 1. A network of terminological ‘options’ for the unit of understanding greenhouse_gas_emission_reduction

The network of English term variants visualises the coreferential ties between the language data in the English source texts. The structure of the network is influenced by the configuration of specific semantic and text-related parameters (see Section 3).  Selecting or deselecting certain parameters can potentially cause term variants to move from the centre to the periphery or vice versa, allowing for dynamic, ‘customised’ visualisations of semantically and contextually structured term variants and their equivalents.

Additional information such as definitions, notes or co-texts (i.e. linguistic contexts) may immediately be linked to language data in the network. This is illustrated by means of the example in Figure 2. In this example, the Dutch term emissiereductie (see also Figure 1) is linked to a Note field in the IATE termbase, a definition in an another external source and, finally, a bilingual concordance marking the position of the translation and its English source term in a specific bitext in the parallel corpus.

Figure 2. Definitions, notes and co-texts are connected to the selected translation equivalent

The aforementioned examples illustrate how the translation resource of semantically and contextually structured translation units can be used as an extension to existing termbases. Moreover, by enriching language data extracted from parallel corpora with semantic and text-related criteria, the translation resource can be used to establish connections between terms in terminology bases and their actual occurrences in texts.

6. Conclusion and next steps

In this article we presented a new type of translation resource comprised of English term variants and their Dutch and French equivalents, manually harvested from a collection of source and target texts. The motivation for doing so was stated in the beginning of this study (see Section 1) and can be summarised as follows: given the fact that term variants frequently appear in special languages and can be motivated on different grounds (Bowker and Hawkins 2006; Freixa 2006), it is important for translators to know the different variants (i.e. linguistic options) available in source and target languages for expressing units of understanding and to know how these can be used (i.e. how they function) in specific communicative settings (Hatim and Mason 1990; House 2001). This involves a careful study of concrete instances of language use. The translation resource of semantically and contextually structured translation units can be perceived as the outcome of such a study. It can be used to connect (and enrich) language data in multilingual terminological databases to actual occurrences in parallel texts (or translation memories).

Not a language-independent concept but a context-bound unit of understanding forms the point of departure in the methodology (see Section 3). In order to visualise the data appearing in the translation resource, a graph representation is proposed in which intra- and interlingual variants derived from texts are directly linked to one another in a contextually conditioned network.

The resource can be perceived as an additional tool to be integrated in a computer-assisted translation (CAT) environment or workflow together with bi- or multilingual termbases and translation memories. Implementations of the prototype resource will need to be tested by translators in order to estimate the relevance of the resource as well as the usefulness of graph representations of intra- and interlingual variants in combination with existing multilingual terminology bases and translation corpora.

Apart from extending the visualisation module, we will also examine how certain natural language processing technologies for identifying intralingual variants and translations can be used to better support the creation of future translation resources comprised of semantically and contextually structured translation units.

About the author(s)

Koen Kerremans obtained his Master's degree in Germanic Philology (Dutch-English) at Universiteit Antwerpen in 2001, his Master's degree in Language Sciences - with a major in computational linguistics - at Universiteit Gent in 2002 and his PhD degree in Applied Linguistics at Vrije Universiteit Brussel in 2014. In his research, several topics pertaining to translation and communication in specialised registers are addressed: e.g. studies of language use in specialised (mono-/multilingual) communicative contexts, specialised translation, terminology and terminological resources (mainly for translators/interpreters) as well as the use of technologies in translation and in other contexts of multilingual communication. His research has been published as peer-reviewed articles in international scientific journals, as book chapters or as conference papers. He currently holds a position as a post-doctoral research and teaching assistant at the department of Applied Linguistics (Faculty of Arts and Philosophy) of Vrije Universiteit Brussel (VUB) where he teaches courses on research methodologies in applied linguistics, terminology and special languages, technical and scientific translation (English-Dutch), Dutch proficiency and software localisation. At VUB, Koen Kerremans is the programme and work placement coordinator of the Master of Translation.

Email: [please login or register to view author's email address]

•  Previous article
• Next article