The contextual behaviour of specialised collocations: typology and lexicographic treatment

1 Introduction

The present article deals with the topic of specialised collocations, with the aim of discussing the distribution of constituents of collocations in technical texts and the way in which this relates to a notion of context that ranges from the immediate surroundings of a node to the inclusion of a much larger portion of text. In order to do so, we will start with a rather broad concept of context as “a frame […] that surrounds a [focal] event being examined and provides resources for its appropriate interpretation” (Goodwin and Duranti 1992: 3; cf. also Goffman 1974 for the notion of frame) but without indulging in further distinctions between context and co-text (cf. Lyons 1995) of collocation constituents. In this way we want to contribute to research on specialised phraseology by presenting an overview of different types of collocational span (Evert 2009), by examining their interplay with contextual properties and by discussing their impact on the (re)presentation of specialised collocations in LSP lexicographic resources.

Despite being a comparatively underrepresented topic in terminology research, the study of specialised phraseology can provide valuable insight into the role of phrasal context in terminology (cf. Sinclair and Carter 2004 on the phrasal nature of language). As pointed out in Giacomini et al. (2020), complex collocations in particular should be the focus of attention in general and specialised lexicography because of their high productivity and key phraseological significance (cf. also Gouws 2015: 184–185).

The contextual behaviour of specialised collocations will be exemplified by English and Italian terms in the domain of photovoltaic technology. This domain has a rich terminology which interfaces with several disciplines, making specialised collocations a varied phenomenon in their denotative content and terminological composition. Section 2 provides an operational, global definition which encompasses specialised collocations and multi-word terms and introduces a typology of specialised collocations, ranging from simple collocations to different forms of complex collocations. The typology reflects the formation processes of shorter or longer collocational sequences in technical texts.

The contribution continues in Section 3 with the description of the English and the Italian corpora and of the behaviour of specialised collocations in context. The corresponding notion of context is inferred from the observation of corpus data and involves all the distinct textual components (e.g. phrase and sentence) in which the constituents of a specialised collocation may appear in a corpus. Accordingly, a typology of contexts is presented and discussed – also from the point of view of their identification in a corpus.

Section 4 explores possibilities of presentation for the different collocational contexts in LSP lexicography and provides an entry draft for a dictionary on photovoltaic terminology. The entry structure covers all available information concerning the contextual behaviour of the collocations of the lemma and proposes a method for systematically ordering this information. The article concludes with some reflections on the relevance of the results of the study and on the need to continue to explore specialised collocations in further domains in order to fully appreciate the role they play in shaping the context of specialised texts.

2 The nature of specialised collocations

In general, an exclusive orientation of LSP collocation theory towards the collocation understanding typical of LGP phraseology appears problematic for a number of reasons, among which is the distribution of idiomatic phrasemes^[1] in specialised texts. The classification of phraseological data based on the principle of ‘idiomaticity’ seems to play a different role in LSP phraseology, where specialised collocations are often assigned to the set of non-idiomatic phrasemes (cf. Gläser 2007: 487). Idiomaticity, i.e. the gradual feature of phraseological expressions that display a discrepancy between their literal and figurative meaning (Burger 2015), has a different status in LSP than in LGP and has often been described as a non-obligatory feature of specialised phrasemes (cf. Cedillo 2004: 91) in the same way as compositionality has been described as a non-central property in terminology (L’Homme and Azoulay 2020: 153). This is supposed to have a direct impact on the flexibility of the context in which phrasemes, collocations in particular, occur in specialised texts. The degree of ‘fixedness’ is also a criterion for distinguishing different types of specialised phrasemes. It certainly affects the contextual behaviour of the least fixed phraseological units, namely specialised collocations, that widely lend themselves to paradigmatic substitutions and syntagmatic modifications. Paradigmatic substitutions are found, for example, in the frequent phenomenon of substantive class formation (Cedillo 2004), i.e. it is often possible to identify, for a given noun in the collocation, a series of alternative nouns that are members of the same semantic field (cf. example a):

Syntagmatic modifiability concerns the ability of a collocation to change its structure by expanding it by means of new elements (cf. example b) and/or adapting it to new inflectional forms (cf. example c).

A further issue is the controversial status of complex terms, in particular multi-word terms (as opposed to specialised collocations) and compounds. Some authors trace a boundary between terminology and phraseology, ascribing to multi-word terms a purely naming function (“La terminologie désigne des objets et concepts alors que la phraséologie formule des relations.”, Gouadec 1994: 173; cf. also Gläser 2007: 494), whereas the function of specialised collocations would be the description of relations. This kind of strict duality also agrees with the view of multi-word terms and specialised collocations as covering complementary syntactic patterns, with the former as typical noun phrases (e.g. fault current, or PV array mounting structure) and the latter as typical verb phrases (e.g. installing a PV array or to route cables). However, as pointed out by Cedillo (2004) and Giacomini (2021), separating word combinations based on their syntactic features leads to inconsistencies when trying to explain the semantic equivalence of variants of the kind V+(P)+N (protect against overcurrent) and N+(P)+N (overcurrent protection or protection against overcurrent). Among complex terms, compounds are, according to some researchers (cf. Burger 2015: 16), equally problematic from a phraseological standpoint since they do not meet the fundamental criterion of polylexicality of phrasemes. A broader definition of ‘polylexicality’ as a combinatorial property of lexical morphemes, not exclusively of words, would help overcome this obstacle and reflect the natural semantic equivalence of some compounds, e.g. (sth is) roof-mounted, and collocations, e.g. mount (sth) on the roof.

In the context of this study, we will start from an operational definition of specialised collocation aimed at a lexicographic description of the phenomenon.

By specialised collocation we mean a combination of two or more words that is typical of a specialised language, with unitary phraseological meaning and terminological character. The terminological character of the collocation is independent of the terminological or non-terminological character of its individual constituents. The degree of idiomaticity of the specialised collocation is variable, as is the degree of its fixedness.

This is a comprehensive definition, which, relying on a phraseological but also empirical notion of collocation (cf., among others, Evert 2009 and Bartsch 2004), is aimed at providing a framework for the treatment of specialised collocations in a lexicographic resource. It allows considerable flexibility in the treatment of data extracted from corpora without precluding further restrictions, e.g. of multi-word terms in the narrow sense.

3 Constituents of specialised collocations and their contextual behaviour

The present study is corpus-based, since the context types of specialised collocations are verified in a corpus according to the previously defined typology. Two small comparable corpora, Photovoltaics2021_en and Fotovoltaico2021_it, have been created for English and Italian, each comprising around 800,000 words and made up of handbooks and guidelines concerning the field of photovoltaic technology. The texts, which have been manually collected among online resources, are addressed to technicians and prospective technicians and cover the topics of design and installation of photovoltaic systems. From a terminological point of view, corpus texts are characterized by a high degree of specialisation in both languages.

The language of the photovoltaics domain, as part of the language related to renewable energies (e.g. hydroelectricity and wind power) is quite rich and combines terminology from base disciplines (e.g. physics, photochemistry, electrochemistry), as well as from sister disciplines (e.g. the construction industry). As a result, the collocations themselves are quite varied in their denotative content and terminological composition. The English corpus shows a considerable amount of nominalization, verb forms are often passivised, and synthetic structures like N-V or N-A compounds (e.g. roof-mounted, air-permeable) are frequent, as is fairly typical in technical texts.

The two corpora have been compiled in Sketch Engine (Kilgarriff et al. 2014) using the previously selected texts. Sketch Engine was also been used for the extraction of specialised collocations and the analysis of their contexts. Data are collected and examined first for English, and the overall result is then compared to the one obtained by analysing the Italian data. The aim is not to identify exact correspondences between the two languages at the level of distribution of different specialised collocation types and corresponding contexts, but to test the applicability of the typology and method to different languages.

The first step of the applied procedure involves the extraction of simple and complex terms using the Keywords tool. Candidate term lists are obtained based on the comparison between the two specialised corpora and the general language reference corpora English Web 2020 (enTenTen20) for English and Italian Web 2016 (itTenTen16) for Italian. They are then assigned a keyness score by the Simple maths method.

4 Presenting collocational context in LSP dictionaries

In this section we will focus on possible ways of presenting the different collocational contexts in LSP dictionaries, providing guidelines for implementing observations made on corpus data. As pointed out by Gouws (2015: 184), “the inclusion of complex collocations remains important and lexicographers should negotiate the best possible way of presenting them and of making users aware of their existence”. As a consequence, this need also involves the presentation of collocations in different contexts. In existing LSP resources the focus of presentation generally falls on predominantly binary specialised collocations, for which usually no context is given or, at most, some usage examples are provided.

The variety of contexts brought to light by our analysis makes us reflect on the need to give these phenomena greater weight in lexicography. Providing the dictionary user with detailed data on the contexts of use of specialised collocations supports with high probability the textual production function of the dictionary. These data can be located within the microstructure of the dictionary in a dedicated section or be systematically substituted for generic usage examples.

Based on the example of PV system, a very frequent collocation from the English corpus, an entry draft will now be presented in which the lexicographic items related to contextual knowledge will be highlighted (Table 5). PV system serves in this entry as a lemma, although the term could alternatively be presented as a collocation of the lemma PV together with other collocations such as PV cell, PV module and PV array.

This study has shown that specialised collocations form a continuum of constituents that fit into contexts of varying length (cf. also Wahl and Gries 2018 for a study of multi-word expressions of increasing length). If the collocational range (McIntosh 1966) of a specialised term is exhausted after a certain number of constituents, as can be inferred from the results presented in Section 3.2, it is necessary to establish, early in the lexicographic process, what the spatial limit in the representation of the phraseological continuum in question can or should be.

From the perspective of textual production both in the mother tongue and in the foreign language, as well as of ‘active’ translation, the availability in the dictionary entry of typical contexts, more or less extended depending on the location, can be crucial. It is reasonable to assume, therefore, that flexibility in the coverage of such contexts is beneficial. Moreover, the typicality of the contexts can be measured in terms of frequency and strength of association in the corpus, obviously adapting the statistical validation thresholds of the candidate collocations as one gradually moves on to more extensive and thus per se (much) less frequent combinations.

The proposed microstructure contains a specific search zone dedicated to the different contexts of use of specialised collocations. Thinking of the ideal user of the LSP dictionary as a translator or technical writer with good metalinguistic skills, we have chosen to mark these contexts with syntactic labels, as shown in the abstract microstructure of the entry:

We have chosen to indicate the various types of context by means of syntactic tags, without resorting to the corresponding terminology (e.g. embedded, argument-related, remote constituents) used in this study, which might not be particularly user-friendly in a lexicographic environment. At the phrase level, collocates of the lemma are highlighted and accompanied by less frequent collocates indicated in round brackets (e.g. install... on a roof). Nominalisations of verbs (e.g. components of a PV system) are referenced to the corresponding verb form (PV system component) and vice versa. Equivalences between different syntactic structures are also indicated, e.g. between a noun phrase with post-modifier and a compound (e.g. design of a PV system and PV system design).

At the discourse level, the emphasis is on the ways in which the collocative context is typically constructed in certain textual genres. Here, it is important to highlight paradigmatic cases of explicit (or implicit) anaphora by means of term repetition or pro-form (both underlined) with indication of the textual genre and source (in parentheses). These context examples are very broad but not generic, as they also focus on the collocative behaviour of terms. Each zone of the entry should be integrated with further (linked) corpus examples. All examples in the entry are followed by the indication of their source as well as of the textual genre.

The presented microstructural model can be varied in many ways, also depending on the mode of publication. Nevertheless, it introduces elements that are essential for the description of the possible context of the specialised collocations of a certain domain, such as

• the subdivision of specialised collocations not on the basis of each individual syntactic structure, but of classes of contexts valid for both SSCs and CSCs;

• the possibility of expanding collocations on the basis of the concrete behaviour of the terms in the corpus, without imposing a predefined scope.

5 Conclusions

This paper has focused on the role of specialised phraseology, in particular collocations, in determining a significant part of the contexts in which domain terminology is used. It contributes to corpus-based research on terminology and phraseology by providing new insights into the formation and behaviour of n-ary collocations in technical texts. Different contexts of simple and complex specialised collocations have been described. It is precisely the complex collocations that turn out to be extremely interesting from this point of view, since they are formed in accordance with different contexts. The possible contexts range from the area of the phrase to that of the sentence (around a predicate) until they cross the border of the sentence to develop in the text discourse.

A notion of collocation context has been directly inferred from the analysis of corpus data: it is the portion of the text in which specialised collocation occurs while still being perceived as a phraseological unit. The context varies both with the manner of collocation expansion from a simple to a complex collocation, and with the anaphoric resumption of collocation constituents from one sentence to subsequent sentences.

Apart from single phrases, in which collocations occur and expand in virtue of syntactic-semantic restrictions and typicality, various factors seem to intervene in the distribution of collocations above the sentence level, for example textual, communicative and pragmatic factors, such as the structural coding conventions of a textual genre already mentioned in Section 3, or possibly further functional or discursive causes (cf. terminology employed by Freixa (2013) for describing causes of term variation).

The limitations of the analysis carried out lie in the restricted possibilities of detecting and thus automatically extracting complex collocations as well as identifying remote components of collocations beyond the level of the individual sentence. In future work, new possibilities for extracting terms related to discourse analysis (Widdowson 2008; Brezina 2018b; Loureda et al. 2019) could be explored, including the contextual role of genuinely pragmatic aspects. From a genuinely computational point of view, the application of existing methodologies for collocation identification such as finite state transducers associated with metagraphs (Tutin 2017) as well as the analysis of word and sentence embeddings (cf., among others, Goldberg 2017, Reimers and Gurevych 2019), might complement the current method by building the ground for quantitative analysis.

Further experiments in data collection and processing should be carried out in new specialist areas to assess the general applicability of the model. Likewise, the ordering strategies for lexicographic data should be further investigated, varying the structure presented in this contribution according to the specific dictionary function and ideal user group, but also taking into account the possibility of covering context information concerning bilingual or multilingual data.