A corpus-driven analysis of structural types of lexical bundles in court judgments in English and their translation into Lithuanian

Formulaicity is one of the characteristic features of legal discourse, which manifests itself not only at the level of wording, “but also in the content, structure and layout” of legal texts (Ruusila & Londroos 2016, 123). Formulaic language, which includes phrasal and prepositional verbs, idioms, collocations, lexico-grammatical associations, lexical bundles, etc., are building blocks of legal discourse shaping legal text meanings. However, up to now, far too little attention has been paid to the nature of frequently occurring “sequences of three or more words that show a statistical tendency to co-occur” (Biber & Conrad 1999, 183), i.e. lexical bundles, in different genres of legal texts. Most studies in the field of lexical bundles in legal texts have only been based on one language (e.g. Jablonkai 2009; Goźdź-Roszkowski 2011; Breeze 2013), whereas translation-oriented contrastive studies on lexical bundles are lacking. In respect of the aforementioned gaps, the aim of this pilot study is to analyse structural types of lexical bundles in court judgments of the Court of Justice of the European Union in English and to examine the way these structures are rendered into Lithuanian. To gain insights into the frequency and structure of lexical bundles, the present study uses the methodological guidelines of corpus linguistics. The classification of lexical bundles into structural types is based on the framework suggested by Biber et al. (1999, 2004). For the purpose of this study, a parallel corpus of court judgments was compiled comprising approximately 1 million words of original court judgments in the English language and about 8 hundred thousand words of court judgments translated into Lithuanian. Lexical bundles in this research were identified using the corpus analysis toolkit AntConc 3.4.4 (Anthony 2015). A concordance program AntPConc 1.2.0 (Anthony 2017) was employed to find Lithuanian equivalents of the most frequent lexical bundles identified in the English court judgments. The evidence from this study suggests that different structural types of


Introduction
In recent years, there has been an increasing interest in formulaicity in different discourses.Formulaic language, which includes phrasal and prepositional verbs, idioms, collocations, lexico-grammatical associations, lexical bundles, etc., has been an object of analysis in a great variety of linguistic disciplines starting from lexicology, lexicography, discourse analysis, corpus-based and corpus-driven analyses and ending with the first and second language acquisition, foreign language teaching, psycholinguistics, semantics, and stylistics.Such interest in formulaic nature of language was stimulated after it had been established that a great amount of language is not constructed according to wordby-word principle but occurs in sequences of words or strings of linguistic items that are more or less fixed in form (Hunston & Francis 2000, 7).It should be also pointed out that it was the development of computer technology and the introduction of corpus linguistics that made the investigation into longer sequences of words possible.
One of the main categories of formulaic language are lexical bundles, which recently have been studied by many researchers, who have noticed that this type of multi-word expressions constitutes a significant part of spoken as well as written language (Biber & Conrad 1999;Cortes 2004;Biber & Barbieri 2007).The term lexical bundles was introduced by Biber et al. in the Longman Grammar of Spoken and Written English to refer to recurrent expressions or "sequences of word forms that commonly go together in natural discourse" and that "show statistical tendency to co-occur" (e.g. in the case of the, it should be noted that, do you want me to (1999,(989)(990).According to the linguists, lexical bundles can be made of three or more words.Chen and Baker, in their turn, expanded the primary definition by adding that lexical bundles "refer to continuous word sequences retrieved by taking a corpus-driven approach with specified frequency and distribution criteria" (2010,30).
The definitions of lexical bundles above suggest that lexical bundles are identified solely on frequency criterion.To be qualified as a lexical bundle, a sequence of words must meet the criteria of orthographic length, minimum frequency and dispersion range (Biber et al. 1999).The first criterion of orthographic length concerns the number of words that constitute a lexical bundle.Depending on the purpose of the study, different authors choose to analyse lexical bundles containing from three to eight words.It has been observed that shorter lexical bundles are much more common but they are "often incorporated into more than one longer lexical bundle" (Biber et al. 1999, 990).The second criterion, namely, minimum frequency refers to the minimum number of times a word sequence must appear in a corpus in order to be counted as a lexical bundle.Previous research on lexical bundles has applied the normalized frequency threshold between 10-40 occurrences per million words (Biber et al. 1999;Biber et al. 2004;Conrad & Biber 2004;Hyland 2008b;Jablonkai 2009).The higher the frequency cutoff point is set, the shorter the list of lexical bundles is generated.Finally, the third criterion, i.e. dispersion range, should be applied in order to avoid individual speaker or writer idiosyncrasies (Biber et al. 1999, 992-993).In the vast majority of studies on lexical bundles (e.g.Conrad & Biber 2004;Goźdź-Roszkowski 2011;Kopaczyk 2013;Güngör & Uysal 2016;Pan, Reppen & Biber 2016), the dispersion range is set at five, which means that a particular lexical bundle must be used in five or more different texts.However, other studies set the dispersion range at 10% of all texts (e.g.Hyland 2008aHyland , 2008b) ) or even more (e.g.Mackiewicz 2017).
It should be mentioned that despite the fact that most lexical bundles represent incomplete structural units, they have been classified according to their structural correlates and their predominant pragmatic functions.Thus, the classification of lexical bundles follows two dimensions: structural and functional.The structural classification is based on the grammatical structure of lexical bundles.Depending on whether lexical bundles incorporate nouns, prepositional phrases, verbs, or clause fragments, they are divided into (1) lexical bundles that incorporate noun phrase fragments (further also referred to as 'nominal lexical bundles') or prepositional phrase fragments (further also referred to as 'prepositional lexical bundles'), e.g., the end of the, those of you who, at the end of; (2) lexical bundles that incorporate verb phrase fragments (further also referred to as 'verbal lexical bundles'), e.g., I am not going to, take a look at, are you going to; (3) lexical bundles that incorporate dependent clause fragments (further also referred to as 'clausal lexical bundles'), e.g., I want you to, if we look at, what I want to; (3) (Biber et al. 2004, 382).In addition, each main structural type of lexical bundles may be further divided into smaller subtypes.From the functional perspective, lexical bundles have been classified into stance bundles, which express attitudes, discourse organizers, which show relationships between different parts of a text, and referential bundles, which indicate physical or abstract entities or refer to textual context (Biber et al. 2004, 384).Each of the mentioned functional categories include subcategories.
The vast majority of studies on lexical bundles have focused on the English language, except for several works which have attempted to apply lexical bundles approach to the analysis of different texts in other languages, such as Spanish (Tracy-Ventura et al. 2007), Korean (Kim 2009), Polish (Grabowski 2014), and Brazilian Portuguese (Sardinha et al. 2014).Research on lexical bundles in the English language has been carried out in two directions: the analysis of spoken and written language.Different discourses, from academic discourse to legal discourse, as well as different genres of texts have been the main sources for investigations of frequency, structural and functional features of lexical bundles.Different genres of spoken and written academic discourse have been investigated by Biber et al. (1999), Cortes (2004), Biber et al. (2004), Simpson (2004), Biber (2006), Hyland (2008b), Simpson-Vlach and Ellis (2010), Salazar (2010), Jalali and Moini (2014), Pan et al. (2016), andYang (2017).Besides different studies focusing on the use of lexical bundles across different disciplines, many of the previous works on lexical bundles in academic discourse have focused on differences and similarities of the use of lexical bundles between L1 and L2 writers (Salazar 1996;De Cock 1998;Schmitt 2005;Salazar 2011;Juknevičienė 2011;Ädel & Erman 2012;Amirian et al. 2013;Purificación 2013;Pan et al. 2016;Güngör & Uysal 2016;Güngör 2016) as well as between novice and expert writers (Cortes 2004;Chen & Baker 2010).
Besides academic discourse, the lexical bundles approach has been also used to study political (Partington & Morley 2004), religious (Shrefler 2011), literary (Stubbs & Barth 2003), medical (Kopaczyk 2013;Grabowski 2013), and legal (Goźdź-Roszkowski 2004, 2006a, 2006b, 2011;Jablonkai 2009Jablonkai , 2010;;Breeze 2013;Kopaczyk 2012Kopaczyk , 2013;;Tománková 2016) discourse.However, a much greater deal of previous research has been concerned with lexical bundles in academic discourse than in any other discourse.Besides, many of these studies focus on pedagogical perspective of lexical bundles.So far, very little attention has been paid to the role of lexical bundles in translation.Drawing on the idea that combining the fields of phraseology and corpus-based contrastive linguistics is "entering relatively unexplored territory" (Ebeling & Oksefjell Ebeling 2013, 1) and in order to fill the above-mentioned gaps, this pilot study aims to analyse structural types of 4-word lexical bundles prevailing in court judgments of the Court of Justice of the European Union in English and to examine the way they are rendered in the translation of the original court judgments into Lithuanian.For the purpose of this study, the following research questions were formulated: 1. What is the distribution of lexical bundles in English court judgments?
2. What structural types of 4-word lexical bundles prevail in English court judgments?
3. What structures are used to render the prevailing structural types of 4-word lexical bundles in English court judgments into Lithuanian?

Lexical bundles in legal discourse and translation
According to Goźdź-Roszkowski and Pontrandolfo, corpus research into legal phraseology has focused on five major areas: (1) research into lexico-syntactic combinations, focusing on specialised collocations; (2) research into routine formulae; (3) lexicographic studies with a focus on terminology and terminography; (4) contrastive studies of phraseology, including translation; and (5) semantics of legal patterns (2015,(133)(134).Investigations of uninterrupted recurrent word sequences, i.e. lexical bundles, belong to the second area.It should be noted that, despite the fact that "law itself is an inherently formulaic discipline and formulaicity an integral quality of law" (Ruusila & Lindroos 2016, 126), phraseology in general and lexical bundles in particular have received relatively scant attention in legal language studies for some time.However, recently, there has been a renewed interest in formulaicity in legal discourse.So far, there have been several noteworthy studies contributing to the investigation of lexical bundles in legal discourse, namely, works by Goźdź-Roszkowski (2006b, 2011), Jablonkai (2010), Kopaczyk (2013) and Breeze (2013).The studies approach lexical bundles from different perspectives: synchronic (Goźdź-Roszkowski 2006b), variation in legal discourse (Goźdź-Roszkowski 2011; Breeze 2013) and standardisation of early legal discourse (Kopaczyk 2013) (Goźdź-Roszkowski & Pontrandolfo 2015, 133-134).
What is more, the investigations focus on different legal genres belonging to four different legal systems, such as in the EU, the US, England, and Scotland.For instance, Goźdź-Roszkowski bases his study on American legal English and analyses such genres of legal texts as legislation, briefs, contracts, court opinions, legal academic journals, legal professional articles, and textbooks.The study suggests that it is legislation and contracts that demonstrate the highest degree of formulaicity.Breeze also focuses on lexical bundles in different English legal genres, including academic law articles, law reports, court opinions, legislation (Companies Acts), and legal documents (e.g.contracts, merger agreements and others).Breeze's analysis shows that legislation and documents might be regarded as the most formulaic genres of legal texts.Variation of lexical bundles in a mix of EU genres of legal texts compiled into one corpus against the British National Corpus is analysed in Jablonkai' s work.This study demonstrates that there are many more lexical bundles in EU legal discourse than in different sections of the BNC.Unlike the overviewed studies, the work by Kopaczyk takes a different approach to lexical bundles and analyses them from a historical perspective.The results of her study reveal the highly formulaic nature of Scots legal texts (burgh records, notarial protocols, statutes and other official and administrative texts), which manifests itself through long 7-word and 8-word lexical bundles.In brief, all the above reviewed studies are based on a cross-generic investigation of lexical bundles.
In addition to cross-generic investigations into lexical bundles in legal discourse, one would also expect works analysing lexical bundles from a cross-linguistic, translational perspective.However, there are scarcely any studies approaching the translation of different genres of legal texts from the perspective of lexical bundles.There exist probably one noteworthy work related to the topic of lexical bundles, translation and multilingualism, i.e. the study by Biel The impact of translation process on the patterning of legal language (2017).In her article, Biel focuses on 4-word lexical bundles in English and Polish.However, instead of applying cross-generic or crosslinguistic perspective towards lexical bundles, the scholar investigates lexical bundles from a new angle, i.e. she focuses her analysis on "internal variation of a single genre of legislation -translator-mediate multilingual legislation and domestic legislation" (i.e.Polish legislation (Biel 2017, 13).For this purpose, Biel uses Polish Eurolect corpus, which is compared to English Eurolect corpus and the Polish Domestic Law corpus.
Biel holds the position that "if we are interested in cross-linguistic comparisons, which lie at the heart of translation, a question should be asked to what extent a 4-gram reflects the same level of formulaicity across languages; in other words, how we can compare 4-grams across languages" (Biel 2017, 16).The scholar presupposes that, as Polish is an inflectional language coding grammatical information morphologically and not marking (in)definiteness through articles, 4-word English lexical bundles may correspond to shorter lexical bundles in Polish.In her opinion, "we are at too early a stage to solve the problem of cross-linguistic comparisons of bundles and more field work is required in this area" (ibid.).Biel's study refutes the hypothesis that translations tend to be "less patterned and less formulaic than nontranslations" (Biel 2017, 23).What is more, it is demonstrated that translations share very few bundles with non-translations, which suggests that translations "resort to their own n-grams prompted by source texts rather than prime bundles which are natural and expected in target-language legal texts" (ibid.).
Despite the fact, that there remains a paucity of evidence on the role of lexical bundles in legal translation, several cross-linguistic studies have been published on other categories of formulaic language in legal discourse.These include the article by Biel (2015), who compares translated and non-translated English and Polish legal language focusing on complex prepositions; a translation-oriented lexicographical study conducted by Castro and Faber (2015), whose work analyses the phraseological units with the head word trial in four legal English-Spanish histories, and the investigation of different phraseological units (e.g.complex nominal phrases, extended lexical collocations, etc.) in the language of contracts in English and Croatian by Basaneže (2015).The most recent research on formulaic language in legal discourse includes the work by Nebot (2017), who focuses on the structure of binomials and multinomials in the International Bill of Human Rights in English, French, and Spanish; and the comparative analysis of noun binomials in UK and Scottish legislation by Kopaczyk (2017).In addition, the study by Salkie (2017) focuses on a contrastive analysis of the English phrase the fact that and its German counterparts; whereas Goźdź-Roszkowski (2017) investigates the same phrase the fact that in American and Polish judicial discourse revealing a general similarity between the usage of the mentioned phrase in both languages.All these works give valuable insights not only into different legal languages but also into different legal cultures.
In the light of the overview of cross-linguistic and translational research on lexical bundles, it should be mentioned that, in general, most of works on the translation of legal discourse have focused on terminology.For this reason, there exists a number of bilingual lexical resources that are mainly restricted to single words and compounds, whereas there is a general lack of bilingual terminological resources including multiword units, which play a significant role in the process of translation.As Parra and ten Hacken state (2008, 1), "[w]hereas multi-word units are linguistically heterogeneous, in translation they raise a very similar set of problems.In order to translate them, they first have to be recognised as belonging together".Thus, the translation of multi-word units poses challenges not only to human translators but also to Machine Translation, as explained by Monti et al. (2013, 8): In spite of the recent positive developments in translation technologies, multi-word units still present unexpected obstacles to Machine Translation and translation technologies in general, because of intrinsic ambiguities, structural and lexical asymmetries between languages, and cultural differences.Multi-word unit identification and translation problems are far from being solved and there is still considerable room for improvement.
Referring to the above-mentioned problems, Grabar and Lefer (2015) offer a lexical bundle approach to the identification of multi-word units by applying this method for the extraction of lexical bundles from English and French comparable and parallel corpora.They focus on 3-word lexical bundles and longer lexical bundles containing them.The methodology offered by Grabar and Lefer involve extraction of lexical bundles, partial lemmatization of French lexical bundles, manual selection of structurally complete bundles, automatic extraction of target language equivalents, and manual validation of target language equivalents.It should be noted that this work approaches lexical bundles in two languages from the perspective of functional taxonomy of lexical bundles suggested by Biber et al. (2004).Altogether, 400 identified English discourse organizers and stance expressions were analysed in respect to their French equivalents.By limiting the scope of the analysis to French equivalents with minimum frequency of two and discarding hapaxes, the final list of over 4000 of French equivalents was generated.
Only 32 out of 400 English lexical bundles were found to have no equivalents in the French language.This study highlights that it is polyfunctional bundles and categorial changes when translating them from one language into another that pose challenges for human translators, Computer Assisted Translation and Machine Translation.Although this research focuses more on political than on legal discourse, its insights can be successfully applied to the analysis of lexical bundles in legal language from a crosslinguistic perspective.

Data and methods
In the course of the investigation of structural properties of lexical bundles in English and Lithuanian, methodological guidelines of corpus linguistics have been followed.
For the purpose of this study, a parallel corpus of judgments delivered by the Court of Justice of the European Union has been compiled.The corpus contains 104 different court judgments in English delivered between June 2015-November 2016 together with their translations into Lithuanian.The size of the corpus is 1 million words of the original texts and 730,000 words of the translated texts.Obviously, the translations of the court judgments are shorter than their English counterparts.
A list of English lexical bundles was extracted with the software program AntConc, version 3.4.4(Anthony 2015), using the N-Grams tool.In this study, the frequency cutoff point was set at 50 times per million words to make the quantity sample of lexical bundles manageable for the analysis.The distribution threshold was set as at 10, meaning that, in order to be considered as a lexical bundle, the multi-word sequence had to appear in at least 10 different English court judgments.To follow the prevailing choice in other studies, only 4-word lexical bundles were considered.
The extracted list of English lexical bundles was further refined manually.During the process of the manual identification and selection of lexical bundles to be included in the final list, all lexical bundles that contained titles of cases (e.g. and others v council), numbers (e.g.regulation no.207 2009), meaningless fragments (e.g.p eer eu e), or overlaps (e.g. the meaning of article and within the meaning of is a part of 5-word bundle within the meaning of article) were excluded from the count.In total, 164 cases of 4-word lexical bundles incorporated in 5-word lexical bundles were deleted.As a result, the total number of lexical bundles after the refinement of the initial list decreased from 590 to 245 lexical bundles.All 245 lexical bundles were further grouped into structural types according to their grammatical structure.The structural analysis of English lexical bundles drew on the previous work by Biber at al. (1999) and Biber et al. (2004).
In order to analyse how different structural types of English lexical bundles are rendered into Lithuanian, the software program AntPConc 1.2.0 (Anthony 2017) was used, and translational equivalents of the identified structural types of English lexical bundles were manually retrieved from the Lithuanian corpus.It should be pointed out that the investigation of translational equivalents of lexical bundles focusing on their structure was limited to 10 most frequent English lexical bundles of each structural subtype.
Besides, to make the analysis manageable, only 10 translational cases of each lexical bundle distributed evenly throughout all the texts were taken into consideration.Based on the translational equivalents in the Lithuanian court judgments, the patterns of structural correspondences were established.

Research findings
The extraction of lexical bundles from a one-million-word corpus of English court judgments revealed that even with a relatively high frequency cut-off point and distribution threshold set, the list of 4-word lexical bundles is rather long.In addition, many of the extracted bundles occur much more frequently than 50 times per one million words.
Actually, there are lexical bundles which are very common and repeated in the corpus more than 500 times.These findings support the view that formulaicity is a characteristic feature of legal language (Kopaczyk 2013;Ruusila & Lindroos 2016;Trklja 2017).
Table 1 below shows the first 20 most frequent lexical bundles in the corpus:

1.
of the basic regulation 2.
of the contested decision 3.
in the present case 4.
the board of appeal 5.
the fact that the 6.
in accordance with the 7.
in accordance with article 8.
on the basis of 9.
in the context of 10.
for the purpose of 11.
for the purposes of 12.
with regard to the 13.
in relation to the 14.
laid down in article 15.
in the contested decision 16.
see to that effect 17.
that the general court 18.
in respect of the 19.
referred to in article 20.
in so far as

Structural classification of English lexical bundles
After the refinement of the extracted initial list of lexical bundles, all 245 bundles were categorized structurally.It should be noted that even though the majority of the selected lexical bundles are not complete structural units, they have certain structural associations, which serve as a basis of the classification.The distribution of the four major structural categories of lexical bundles is illustrated in Figure 1: As can be seen in Figure 1 above, lexical bundles in court judgments are primarily prepositional.Lexical bundles that incorporate prepositional phrase and noun phrase fragments account for almost 70% of all the bundles, whereas verbal and clausal lexical bundles tend to be infrequent.These results are in line with those of the previous studies conducted on lexical bundles in legal discourse by Goźdź-Roszkowski (2011), Breeze (2013) and Tománková (2016).Similar tendencies were observed by Conrad and Biber (2004)  in academic prose.This is in contrast to the findings in conversation, where even 90% of all lexical bundles incorporate verb phrases (Conrad & Biber 2004, 63).
Several subtypes were identified in the major four categories of lexical bundles.The structural classification and distribution of the lexical bundles across the structural subtypes are presented in It can be seen from the data in Table 2 that lexical bundles that incorporate noun phrase fragments can be divided into two subtypes.The first group of nominal lexical bundles, which accounts for 72% of all lexical bundles of this type, include lexical bundles that consist of a noun phrase followed by a post-modifying of-phrase fragment.The second group of nominal lexical bundles is made of noun phrases with other post-modifier fragments.This group covers 28% of all nominal lexical bundles.Together these subtypes are used to specify particular aspects of information in court judgments (e.g. an infringement of the, the nature of the, the general court in).
Even though, as illustrated in Figure 1, verbal lexical bundles constitute only 23% of all lexical bundles in court judgments, they are more varied in structural subtypes.Overall, four subtypes of this category have been distinguished.The largest subtype, i.e. 34% of all verbal lexical bundles, is made of lexical bundles that incorporate a verb phrase fragment with a verb in the active or passive voice, which are used to denote some kind of an action (e.g. must be rejected as, see to that effect).The table above shows that each of the other two subtypes, namely, lexical bundles that incorporate a noun phrase with a verb phrase fragment and a past participle with a prepositional phrase fragment covers 23% of all verbal lexical bundles.Lexical bundles that incorporate a noun phrase with a verb phrase fragment are mainly used to refer to actions of the parties to the case (e.g. the applicant claims that), whereas lexical bundles representing the subtype of lexical bundles made of past participles with prepositional phrase fragments are used to specify a particular part of the legal text or an institution (e.g.laid down in article, provided for in article).20% of verbal lexical bundles are covered by the subtype that incorporates it with verb phrase fragments, which serves to report the stance of the author such as necessity (e.g. it should be noted), logical consequence (e.g. it follows that the, it is clear that), or appropriateness (e.g. it is appropriate to).
The last, rather infrequent type of lexical bundles in court judgments, i.e. clausal lexical bundles, can be divided into three subtypes, as shown in the table above.Lexical bundles that incorporate that-clause fragment are the most common structure of clausal lexical bundles as it accounts for almost 70% of all lexical bundles of the clausal type.These multi-word sequences appear mainly either with the parties to the case as the subject (e.g. that the general court, that the applicant had, that the commission was), with an extraposed clause embedded in that-clause, with it as the subject and the copula is/was as the verb (e.g. that it is not, that it was not), or that with prepositional phrase fragments (e.g. that in order to, that in accordance to).Lexical bundles with the to-clause fragment covers 26% of all clausal lexical bundles and are used to refer to particular actions (e.g. to pay the costs, to ensure that the, to determine whether the).Only one case of the lexical bundle with -ing form clause fragment (having regard to the) was found in the corpus, which covers only 5% of all clausal bundles.
Taking into consideration the analysis of structural types of lexical bundles in court judgments it can be noted that lexical bundles mainly incorporate prepositional and nominal phrases.As explained by Goźdź-Roszkowski, "such distribution of structural types contributes to the perception of legal genres as linguistically highly nominal" (2011,117).However, court judgments have a small proportion of lexical bundles incorporating verb phrase and dependent clause fragments, which are more characteristic of spoken language.All four main structural types of lexical bundles are expressed by different structural subtypes, which help to shape the meanings in court judgments.The following subsection of the paper focuses on the way the discussed structures of lexical bundles in court judgments are rendered into Lithuanian.

Rendering of the structure of English lexical bundles into Lithuanian
The main goal of the analysis of the translation of English lexical bundles into Lithuanian is to examine what happens with the structure of 4-word English lexical bundles in translation and to establish the translational patterns of different structural types of lexical bundles.
First of all, the largest group, i.e. prepositional lexical bundles, was taken into consideration.The contrastive analysis of English prepositional lexical bundles and their Lithuanian equivalents revealed that prepositional lexical bundles incorporating in + noun phrase fragments are rendered into Lithuanian by five different structures, as illustrated in Table 3.The table above shows that the same structural pattern in English may be translated using prepositions, participles, conjunctions, adverbs, or noun phrases.Although all prepositional lexical bundles with the preposition in incorporate a noun (except for in so far as) in English, the noun element is not rendered in most of the translational equivalents.This pattern is used when translating lexical bundles that incorporate in + a noun referring to an abstract entity (e.g. in accordance to, in support to), whereas if this structural subtype of lexical bundles incorporates noun phrases referring to legal domain (e.g. in the contested decision, in the present case), the noun element of these lexical bundles is transferred in the translation.What is more, all lexical bundles of this structural subtype are translated into Lithuanian by one to two words, since functional words such as the, of, to, with are not transferred.

Structural patterns English lexical bundles
The table below illustrates some of the main structures used to render English lexical bundles incorporating of + noun phrase fragments into Lithuanian.Besides 2-word noun phrases, these lexical bundles are also translated simply by single nouns or dependant clause fragments, as shown in Table 4.In most cases, noun phrase fragments present in this structural subtype of lexical bundles are retained in translational equivalents, only these nouns are used in the Genitive case in the Lithuanian language.What is more, the majority of English lexical bundles incorporating of + NP is translated directly into Lithuanian; thus, this structural subtype of lexical bundles should be least challenging to translators.

Structural patterns English lexical bundles
Table 5 shows that other prepositional lexical bundles starting with such prepositions as on, for, with, before, and to are rendered into Lithuanian by participles, conjunctions, noun phrases, prepositions, or prepositional constructions.It can be seen from the data in Table 6 that lexical bundles incorporating noun phrases with other post-modifier fragments are rendered into Lithuanian either by noun phrases or nouns + dependant clause fragments.In general, in most cases of the translation of nominal lexical bundles, the noun element is retained in Lithuanian.

Structural patterns English lexical bundles
Table 7 provides the summary of translational equivalents of verbal lexical bundles.It has been observed that lexical bundles incorporating only verb phrases in an active or passive voice are rendered into Lithuanian by infinitive phrases.Further, it has been found that the structural pattern of a noun phrase + a verb phrase is retained in the translation, as all of these lexical bundles are translated by the structure made of a noun and a verb.The table provides clear evidence that, when translating both lexical bundles incorporating that-clause fragments and lexical bundles incorporating to-clause fragments, the original structural patterns are retained, although translational equivalents are shorter, which may be explained by differences between the two languages.

Conclusions
The present pilot study was designed to investigate lexical bundles from structural and translational perspectives.In contrast to the majority of the previous studies on lexical bundles, which have been based on the English language, this is the first study to investigate lexical bundles in legal discourse by taking into consideration two languages, i.e.English and Lithuanian, and conduct a contrastive analysis.The findings of this research suggest that lexical bundle approach could be adopted not only to language learning and teaching but also to translation.
This study has shown that the analysis of lexical bundles provides insights into formulaic nature of legal discourse.Taken together, the results of the paper have revealed that four different structural categories, which are characteristic of this particular genre of legal texts, prevail in English court judgments.The great frequency of pre-formulated multiword sequences in English court judgements reflects a high level of formality of legal texts, which leaves little room for creativity.The most obvious findings to emerge from the analysis are that prepositional and nominal lexical bundles are the most frequent multi-word expressions in court judgments, whereas verbal and clausal bundles are not prevailing in court judgments.This tendency confirms the observation made by many researchers that verbal lexical bundles are more characteristic of spoken language.The four major types of lexical bundles were further classified into smaller structural groups of lexical bundles, identifying 12 different structural subtypes of lexical bundles found in English court judgments.
The analysis of the translation of the identified subtypes of English lexical bundles showed that prepositional lexical bundles are rendered from English into Lithuanian by a great variety of different structures, i.e. conjunctions, adverbs, participles, prepositions, and noun phrases.Nominal lexical bundles are rendered from English in Lithuanian by noun phrases, nouns, participles, nouns + dependant clause fragments.Verbal lexical bundles are translated from English into Lithuanian mainly by infinitive phrases, nouns + verbs, nouns + past participles, verbs, and verb phrases.Both in the translation of nominal and verbal lexical bundles, the noun element and verb element are retained respectively.Finally, clausal lexical bundles are rendered from English into Lithuanian by that-clause fragments and infinites.As the paper demonstrates, a lexical bundle can be successfully treated as a single cognitive unit for the purposes of establishing a translational equivalent.The identification of translational equivalents in the Lithuanian corpus is in line with the fact that English and Lithuanian are typologically different languages.
There are several noteworthy limitations of this pilot study.First of all, this pilot study is limited to the analysis of only 4-word lexical bundles.What is more, this paper does not provide a complete picture of translation tendencies followed when translating different English lexical bundles into Lithuanian since not all 245 cases of the identified lexical bundles were taken into consideration in respect of translational equivalents.Moreover, even structurally incomplete lexical bundles were taken into consideration when identifying translational equivalents in the parallel corpus.Finally, the translation of different structural types of lexical bundles from English into Lithuanian was approached only from the structural perspective.Despite the limitations, this study is considered to contribute to the existing knowledge of lexical bundles in legal discourse in general and the application of lexical bundle approach to legal translation in particular.
There is still room for further research to provide better understanding of the application of this approach to the translation of different genres of legal texts.Firstly, the investigation into longer lexical bundles may provide better insights into the formulaic nature of legal discourse.What is more, a research focusing on Lithuanian lexical bundles in translated court judgments and court judgments originally written in Lithuanian would offer some important insights into the phenomenon of formulaicity in legal discourse in general and the role of lexical bundles in both languages in particular.Finally, the contrastive study of translational equivalents of lexical bundles based on the functional perspective could reveal more promising results.

Figure 1 .
Figure 1.Distribution of structural types of lexical bundles in English court judgments in academic prose.According to Conrad and Biber, lexical bundles that are parts of noun phrases and prepositional phrases account for about 60% of all lexical bundles Corpus

Table 1 .
The first 20 most frequent English lexical bundles in the corpusThe list of 20 most frequent lexical bundles in the corpus suggests that most of the bundles belong to shared or unmarked vocabulary, to use Varo and Hughes terms(2002,  16) (e.g. in respect of the, in relation to the, in accordance with the).The other group of lexical bundles in the list include lexical bundles related to legal domain (e.g. of the basic regulation, laid down in article, of the contested decision).Besides, this short list reveals the prevalent structural type of lexical bundles in court judgments.It is prepositional lexical bundles that constitute the major part of all the bundles in the list.They are used to express abstract, logical relations (e.g. on the basis of, in relation to the), or specify legal context (e.g. of the basic regulation, in respect of the).

Table 2 .
For the purpose of this study, lexical bundles that incorporate noun phrases are taken separately from the ones that incorporate prepositional phrase fragments, although they are included under one category inBiber et al.'s (2004)classification, since it is presumed that different structures might be used in Lithuanian to render prepositional lexical bundles and nominal lexical bundles.

Table 2 .
Structural classification and distribution of lexical bundles in English court judgmentsIt is apparent from Table2that, in total, twelve different subtypes of lexical bundles have been identified.The table above shows that lexical bundles that incorporate prepositional phrase fragments have been found to contain the following three structural subtypes: in + noun phrase fragment, of + noun phrase fragment, and other prepositions(on, for,  to, at, according to, under, by, from, with)+ noun phrase fragment.In + noun phrase fragment is, undoubtedly, the most numerous category of lexical bundles incorporating prepositional phrase fragments; it makes 37% of all prepositional lexical bundles.Many of these bundles are used to interpret the prior or forthcoming text, to indicate relations in the text (e.g. in accordance with article, in the preamble of), or abstract relations (in relation to the, in the event of the, in the form of). Lexical bundles with the construction of + noun phrase fragment account for 30% of all lexical bundles of this type.These bundles help to specify a particular text, legal context, values, and principles (e.g. of the basic regulation, of the principle of, of the value of).33% of all prepositional lexical bundles are covered by lexical bundles that start with other prepositions.

Table 3 .
Rendering of lexical bundles that incorporate in + NP into Lithuanian

Table 4 .
Rendering of lexical bundles that incorporate of + NP into Lithuanian

Table 5 .
Rendering of lexical bundles that incorporate other prepositions + NP into LithuanianThe translational profile of lexical bundles incorporating other prepositions followed by noun phrase fragments shows that they are rendered into Lithuanian by the structures of one or two words omitting functional words of the source language.Besides, the same translational pattern identified in the case of the translation of the structural subtype in + noun phrase is retained when translating lexical bundles which incorporate different prepositions + noun phrases.To be more specific, lexical bundles which are made of prepositions and nouns referring to abstract entities (e.g. on the other hand, with regard to the) are translated by different parts of speech, except for nouns, whereas lexical bundles incorporating prepositions and noun phrases related to legal domain (e.g. to the contested decision, before the general court) are rendered into Lithuanian by noun phrases.The analysis of translational equivalents of nominal lexical bundles has revealed that this type of lexical bundles is mainly rendered into Lithuanian by noun phrases or simply by nouns.Lexical bundles with noun phrase + of-phrase fragment tend to be translated by noun phrases either in Nominative or Genitive case.In addition, this structural subtype of lexical bundles is rendered into Lithuanian by participles, which refer to the actions expressed by nouns in the English language (e.g.adoption -'the act of adopting').

Table 6 .
Rendering of lexical bundles that incorporate nominal lexical bundles into Lithuanian

Table 7 .
Rendering of verbal lexical bundlesThe investigation of the parallel corpus has also revealed that past participle + prepositional phrase constructions are rendered into Lithuanian using past participle phrases; thus, the main elements of the original structure are retained.The last subtype of verbal lexical bundles, it + verb phrases, are translated by single verbs or verb phrases with infinitives, as illustrated in the table above.The data in the table also indicate that many synonyms are used to render the structure of English verbal lexical bundles into Lithuanian.What is more, the translation of some lexical bundles (e.g.see to that effect, see by analogy judgment) by equivalents incorporating abbreviations (e.g.šiuo klausimu žr.) reveals some of the stylistic conventions in the Lithuanian language.The data in the table below gives an overview of the translational correspondences of clausal lexical bundles in the Lithuanian language.

Table 8 .
Rendering of clausal lexical bundles