Testing Out Translation Universals in Legal Translation: Quantitative Insights From A Parallel Corpus of Spanish Constitutional Court’s Judgments Translated Into English

Abstract Research into ‘translation universals’ in legal translation is a relatively new field, which still needs to be expanded with further empirical studies. The few studies conducted so far fall into two main categories: a) analyses that explore the typical features of European legalese as translated language against national legal language; b) studies based on corpora of national legal language translated into other national languages. The present paper is framed within the second category and aims at contributing to the academic debate on translation universals applied to legal language; more specifically, it aims at testing the methodology adopted to study translation universals on a bilingual parallel corpus of judgments delivered by the Spanish Constitutional Court (Tribunal Constitucional, TC) translated for informative purposes into English. The corpus-based analysis, carried out mainly quantitatively, includes the comparison with a larger corpus of original judgments delivered by the UK Supreme Court (UKSC) with the final objective of testing some indicators of simplification, explicitation, normalisation, levelling out, interference, untypical collocation. Preliminary results are promising, even though it is not possible to identify robust and homogeneous trends.


Introduction
The present paper 1 is part of an ongoing research project which is being carried out by the author at the University of Trieste, aimed at applying the methods usually used to test translation universals (see Zanettin 2012: 12-13) to the field of legal translation (Pontrandolfo 2019b(Pontrandolfo , 2020. It presents the preliminary results of a small pilot study conducted in the Spanish-English combination (see Franceschini 2020). More specifically, it follows the suggestion made in Pontrandolfo (2019a: 22): Empirical research on [legal translation universals] is extremely needed both in training and professional settings. The use of legal corpora effectively helps scholars to isolate descriptive features of translations that actually give insights into the complex dynamics of legal translation.
One of the major advantages of adopting a corpus linguistics perspective is that, prior to corpus studies, many of the suggestions regarding translation universals were an often stated but unproved hypothesis (Zanettin 2012: 17). Corpora can effectively demonstrate trends based on empirical findings. Table 1 lists and defines the universals identified in literature (see Zanettin 2012: 12-25) together with some indicators 3 of how to test these regularities with corpora. Simplification "the idea that translators subconsciously simplify the language or message or both" (Baker 1996: 176) Type-Token Ratio (TTR), lexical density, high frequent vs. low frequent words, mean sentence length Explicitation "an overall tendency to spell things out rather than leave them implicit in translation" (Baker 1996: 180-181) mean sentence length, discourse markers/linking adverbials, punctuation Normalisation "in translation, ST relations are often modified in favour of habitual options offered by a target repertoire" (Toury 1995: 268); "the language of translation tends to conform to typical patterns of the TL and translators tend to adhere to conventional expressions at the expense of more creative ones" (Zanettin 2012: 19) Conservatism/ standardisation in the choice of language patterns (e.g. grammar, punctuation, collocations, etc.) (Baker 1996: 183).

Levelling out
"the tendency of TT to gravitate around the centre of any continuum rather than move towards the fringes" (Baker 1996:177) (Baker 1996: 184) Translation of unique items "TL specific lexical items, i.e. elements which tend to be 'untranslatable' because they lack a straightforward lexicalised equivalent in other languages, are proportionally underrepresented in translation" (Tirkkonen-Condit 2004, Zanettin 2012 specific, languagebound elements (such as clitics, see Tirkkonen-Condit (2004) Untypical collocations "translations tend to favour word combinations that 'although possible in the TL system, are rare or absent from actual TL texts" (Mauranen 2008: 44).

phraseology (especially collocations)
Interference "features of the SL are transferred to the TT in the process of translation" (Toury 1995).
negative interference (deviation from the typical features of the target language: such as syntactic calques); positive interference (overrepresentation of linguistic features which already exist and are acceptable in the TL) Toury (1995: 275) Following Chesterman (2004: 39), a distinction could be made between S-Universal (where S is for Source) and T-Universal (T for Target): the former are "characteristics of the way in which translators process the source text" whereas the latter are "characteristics of the way translators use the target language". The former generally require bilingual parallel corpora whereas the latter are usually tested with monolingual comparable corpora (e.g. translated English vs. original English).
It is difficult to operate a clear-cut distinction among these universals due to an inevitable overlapping, as in the case of simplification and explicitation (simplifying a text means explicitating it on certain levels) or simplification and normalisation (where dividing a long sentence, for example, can be either the result of a simplification process or a normalisation one). This is the reason why they are generally considered together and, for the purposes of this study, they have been used exclusively as an effective method to isolate trends/regular patterns in the corpus under investigation.
As far as legal translation studies (see Prieto Ramos 2014) are concerned, translation universals have not been tested extensively, possibly due to the absence of large legal corpora that could be used as testbeds to confirm or disconfirm such patterns (see Biel 2014: 96-110). A recent overview of the main studies conducted in this area is provided in Pontrandolfo (2019a: 20-22).
The aim of the present study is to contribute to filling this gap by presenting a case study on the application of the methods used to test some translation universals to a parallel corpus of judgments delivered by the Spanish Constitutional Court (Tribunal Constitucional, TC) and translated for informative purposed into English.
Indicators of six translation universals have been applied to the TC corpus, mainly from a quantitative perspective: 1) simplification, tested by analysing quantitative elements pertaining to textual complexity (lexical variety, medium number and length of period, lexical density); 2) explicitation, tested by analysing the frequency and types of discourse markers that clarify the message in the corpus; 3) normalisation, tested by using complex prepositions, which are typical features of legal texts (Pontrandolfo 2020) as a testbed, thus establishing whether translated texts adhere to textual conventions and expectancy norms; 4) levelling out, tested against the background of lexical variation and medium number and length of period; 5) untypical collocation hypothesis, empirically tested by looking at one of the most frequent collocation patterns in the corpus, i.e. Verb + Noun of the ten most common nouns in the corpus; 6) interference, tested by replicating a study on antinormative gerunds (Pontrandolfo 2019b) as a source of potential negative interference.
As far as the expected results and the hypotheses guiding the study are concerned, translated texts would display a lower lexical variety, a higher number of (shorter) sentences and a lower lexical density (simplification); a higher frequency of explicative discourse markers in translated texts (explicitation); similar complex prepositions in the translated corpus (normalisation); a higher consistency and uniformity in translated texts (levelling out); rare or infrequent collocation patterns (untypical collocation hypothesis); negative interference and therefore instances of anti-normative gerunds also in the target texts (interference).

The Spanish Constitutional Court
The Tribunal Consitucional (TC) (see art. 1, Ley Orgánica 2/1979) is the supreme interpreter of the Spanish Constitution; as such, it is an independent body, which does not pertain to the other judicial courts of the country. It is the authority guaranteeing the respect and application of fundamental rights, public freedoms and supremacy of the Constitution.
The TC's case-law affects the whole national territory. The decisions of the TC are not appealable because they are considered as final judgments. Its powers comprise action or question of unconstitutionality, preliminary appeals of unconstitutionality, complains regarding tax issues, recursos de amparo for violation of public rights and freedoms, conflicts between constitutional bodies, conflicts in defence of local or foral autonomies, declaration on the constitutionality of international treaties, etc.
Article 5 of LO 2/1979 establishes the composition of the TC: eleven members (called Magistrados) appointed by the King of Spain and by the Spanish Parliament, the Government and the General Council of the Judiciary.
The Court exercises its judicial functions through the bodies in which it is composed, that act in several constitutional processes: Plenary, Chambers and Sections (see arts. 6-10 of LO 2/1979).
As the website of the TC clearly states: The Constitutional Court of Spain offers a selectionthat will be progressively increasedof the grounds of its most relevant decisions translated into English. These are not official translations of the Judgments: the texts are provided to allow the consultation of legal grounds and the knowledge of the Court's case law.
It is thanks to these translations that the present pilot study has been carried out. Before delving into the composition of the corpus, some information on the legal translation activities at the TC is needed. These data have been gathered by means of a questionnaire directly addressed to the body (more specifically to the General Secretary of the body, from which the Translation unit of the TC depends).
The TC established its own small translation unit in 2016. Before that, all the translations were outsourced. It is made of two translators working from Spanish into English and French respectively. Both of them have a double degree in law and translation and are native speakers of the two foreign languages. They have access to some training opportunities in the field of legal translation at the Constitutional Court. In their translation activities, the two translators have recently been trained to use CAT-Tools (Déjà Vu) and the texts they produce are always reviewed by a legal expert working in the TC. The TC library offers them many resources and tools (including databases such as Eur-Lex, Hudoc, etc.), as well as terminological records in the language combinations. They also refer to the TC Style manual, which contributes to guaranteeing internal consistency.

The TC corpus
The Tribunal Constitutional corpus (TC corpus) is a bilingual parallel corpus containing both judgments delivered by the Spanish TC (STC_ES) and their translations in English (STC_EN). The texts were extracted from the TC website 4 . Each subcorpus contains 31 judgments dealing with different topics such as right to strike, discrimination based on sex, education rights, rights related to religion, access to justice, rights of foreign citizens, freedom of information, same-sex marriage. As far as the textual composition of the judgments is concerned, only two sections of the judgment were included: fundamentos de derecho (egal reasoning) and fallo (final decision); this is because some of the texts in the TC website were incomplete and therefore, for consistency reasons, only the text parts that were available for the whole dataset were included in the TC corpus.
A larger reference corpus has been built to test some of the translation universals: UKSC. Partially based on COSPenSup (see Pontrandolfo 2016: 84-85), it is a monolingual ad-hoc corpus containing 178 judgments delivered by the Supreme Court of the United Kingdom (2009-2019), whose original texts have been extracted from the British and Irish Legal Information Institute (BAILI) database 5 .
Even though the UK does not have a full-blown Constitutional Court, some of the powers of the Spanish TC are similar to the UK Supreme Court; hence the reference corpus was created based on the criterion of thematic similarity. It is worth stressing that the UKSC corpus was mainly used as a linguistic testbed (more than a contentrelated dataset) and this is the reason why the comparability of the two bodies has not represented a major methodological concern during the analysis. Table 2 summaries the final composition of the TC corpus. The building of the corpus has relied on LF Aligner v. 4.2 to automatically align source and target texts and therefore creating translation memories and on Sketch Engine (Kilgarriff et al. 2004) which has automatically recognised the languages, separated the subcorpora, aligned the TMX files and POS-tagged 8 them. In order to extract the data for the quantitative analysis, the queries on Sketch Engine have been complemented by using WordSmith Tools v. 5.0 (Scott 2008).
The following analysis is therefore based on this bilingual parallel corpus and the reference corpus. The main focus is the translated language in the framework of translation universals, conceived, according to the specific cases, as both S-Universals and/or T-Universals.

Simplification
Three elements related to textual complexity have been investigated to test the simplification universal, namely lexical variety, average sentence length and lexical density (see Zanettin 2012: 14-16).
As far as lexical variety is concerned, it indicates how rich the vocabulary of a text is and therefore it can be a measure of linguistic complexity. The STTR can therefore become an indicator of the complexity of a text: the higher the number of types (non-repeated words in a subcorpus), the more varied the language variety represented by it; therefore, a lower STTR in translated texts would point to a lower variety of language resulting from a process of simplification (see Zanettin 2012: 14-15).
The statistics obtained from WordSmith Tools are available in Table 2: the STTR of STC_ES (36.50) is higher than STC_EN (34.34), pointing to a higher lexical variety in original (ES) vs. translated (EN) texts, which seems to confirm the simplification hypothesis. UKSC's SSTR is almost equivalent to that of STC_EN (34.31 vs. 34.34).
Another potential indicator of lexical simplification is the comparison of the average sentence length: according to the simplification universal, the lexicon of translated texts would be less dense compared to original/non-translations. Table 3 shows the number of sentences with their mean length in each subcorpus. STC_ES has a higher number of sentences but a lower average sentence length compared to STC_EN. This seems to go against the simplification hypothesis since a lower number of sentences could point to the condensation of two or more sentences in one. Such quantitative data are counterintuitive considering that English syntax is simpler than Spanish syntax where subordinated sentences, especially in legal discourse, are a common feature (see Hernando Cuadrado 2003: 32-33). This is possibly due to some negative interference with the complex, baroque legal Spanish.
As far as the average sentence length is concerned, the sentences of STC_EN contain 181 more words than STC_ES which goes, again, against the simplification hypothesis. Compared with the reference corpus (UKSC) the number is even more striking (43.85 vs. 30.35): original texts use shorter sentences than translated text.
Finally, lexical density is measured by computing information load as a function of the ratio of lexical (content) words to grammatical (function) words (Stubbs 1996: 172 in Zanettin 2012. Translated texts are hypothesed to have a higher ratio of grammatical words and more repetition, and thus a simplified lexicon. In other words, higher lexical density is expected to be found in non-translated texts (Zanettin 2012: 15). Table 4 shows the number of content words (nouns, verbs, adjectives, adverbs) with their raw and relative frequency. Lexical density has been calculated by means of the following proportion: x : (lexical words) = 100 : (total words) Table 4 shows the results related to lexical density in the corpora under exam. The lexical density of original texts (UKSC: 53.74%) is slightly higher than that of translated text (STC_EN: 53.03%). This partially confirms the simplification hypothesis, even though there is no significant difference between the two datasets.
The tests conducted for the simplification universal do not seem to confirm robustly the translation universal: if lexical variety confirms it, average sentence length does not seem to go in this direction; moreover, lexical density partially (slightly) confirms the potential existence of the simplification hypothesis. As stated by Mauranen (2008), the simplification of some traits could result in the increase of complexity of other traits.

Explicitation
Explicitation has been investigated as both a S-and T-Universal (Zanettin 2012: 16-17). As a T-universal, it requires the use of monolingual comparable corpora (original vs. translated texts) used as testbeds to see whether translators tend to "spell out things in translation" (Baker 1996: 176). As a S-Universal, it requires the use of bilingual parallel corpora (source vs. target texts) to see if, for example, translators resort to more words in their texts to explain culturally or linguistically distant features of the source texts.
In order to test the explicitation hypothesis in the TC corpus, a linguistic category has been chosen: discourse markers and, more specifically, the so-called marcadores reformuladores (Portolés 2001: 141-143) / linking adverbials (Biber et al. 1999: 875-879). To guarantee a consistent analysis, only adverbials having the same function have been selected thus establishing a functional comparison between Spanish and English (other irrelevant discourse markers have not been included in the analysis): 1) reformuladores explicativos / apposition linking adverbials; 2) reformuladores de rectificación / concession linking adverbials; 3) reformuladores de distanciamiento / contrast linking adverbials; 4) reformuladores recapitulativos / summation linking adverbs.
If explicitation takes place in the corpus, then translated texts (STC_EN) will present a higher percentage of linking adverbials than their source texts (STC_ES) and original texts (UKSC). Table 5 presents an overview of the data extracted from the TC corpus.  Table 5 confirms that STC_EN contains a higher percentage of linking adverbials compared to STC_ES, which would point to translated texts being more explicit than their source texts. This frequency is always higher except from the last category (recapitulativos/summation). Table 6 and 7 present all the occurrences of the discourse markers in each subcorpus with a view to establish the frequency of the types of adverbials.  The explicitation hypothesis, considered as a T-Universal, would imply a comparison of linking adverbials with original texts (UKSC). Table 8 shows the relative frequency of each category in the comparable corpora. Data reveal that translated texts contain a significant higher percentage of linking adverbials (2.78 vs. 1.58%). Table 9 compares the occurrences and relative frequency of each discourse marker in STC_EN and UKSC. This allows to establish if STC_EN contains more different adverbials (types) compared to the original corpus UKSC. Table 9. Linking adverbials STC_EN vs. UKSC Data confirm a higher frequency of linking adverbials in STC_EN and a more limited range of different adverbials. Translated texts seem to use only some of the linking adverbials whereas original texts resort to a wider range of discourse markers (see the list in Biber et al. 1999: 875-879). Table 10 summarises the comparison among STC_ES, STC_EN and UKSC. The results seem to support the claim that translators make syntactic relations more explicit by using linking adverbials more frequently than is typical of non-translated language. Moreover, original texts (UKSC) contain a more varied use of discourse markers which could point to the fact that translators tend to repeat the same linking adverbials, thus simplifying the overall structure of the text.

Normalisation
In order to test if translations show a preference for more habitual options offered by a target repertoire (Toury 1995: 268) and therefore if translators tend to adhere to conventional expressions at the expense of more creative ones (Zanettin 2012: 19), a feature of legal language has been selected, namely legal phraseology and, more specifically, the so-called locuciones preposicionales / complex prepositions (see Pontrandolfo 2020). Following Biber et al. (1999: 75), "complex prepositions are multi-word sequences that function semantically and syntactically as single preposition". These phraseological units play a key role in legal and judicial discourse (see Biel 2015: 141-142). As Biel points out (2015: 141), the distinctiveness of complex prepositions in legal language results from their increased frequency compared to everyday language.
The normalisation universal has been tested as a T-Universal and therefore by means of a comparison between STC_EN and UKSC to see if translators tend to get closer to the stylistic conventions of original texts (UKSC).
Complex prepositions have been semi-automatically extracted by using the Concordance-CQL function in Sketch Engine and setting the cut-off frequency at 30 occurrences.
The most frequent (and standard) complex prepositions used in UKSC are summarised in Table 11.  Table 11 shows that the most frequent types (different forms) of complex prepositions are 63; these can be considered as the most common ones in the reference corpus and can be compared with those extracted from STC_EN, as presented in Table 12. Data show that there are 15 most frequent complex prepositions in STC_EN. Due to the differences in size of the two corpora, only relative (normalised) frequencies have been considered for the analysis and only the complex prepositions common to both datasets have been investigated (see Table 13). The comparison between the two subcorpora shows that all the complex prepositions in STC_EN are also present in the most frequent patterns in UKSC. This could point to the existence of the normalisation hypothesis. In general, the most frequent complex prepositions common to both subcorpora are also the most frequent in STC_EN, which could suggest a tendency to adhere to the textual conventions of the target language. However, the most frequent complex prepositions in translation (STC_EN) are not equally frequent in the original texts (UKSC); in fact, the most frequent ones in STC_EN tend to be the less frequent ones in UKSC.

Levelling out
Levelling out has not been frequently empirically investigated as a translation universal. In order to test if translated texts tend to show more uniform patterns compared to original texts (levelling out as a S-Universal), two indicators have been considered: lexical variety in terms of STTR and average sentence length, which were also considered as indicators of simplification, and have been tested in other corpus-based studies (see Corpas Pastor et al. 2008, Redelinghuys 2016. Uniformity is to be understood as the degree of variation of a specific linguistic trait in the corpus. To this aim, each judgment or the corpus has been considered by assigning it a medium value related to the STTR and medium sentence length (data obtained from WordSmith Tools). The degree of variation has been calculated by subtracting the maximum and minimum value; the difference is the interval in which all the other values are comprised and it is therefore the maximum degree of variation available in the corpus. By comparing these intervals, it is possible to isolate the minimum intervals that could point to a more uniformity in the corpus (due to the similarities of the values of the judgments composing that specific corpus).
If the levelling-out hypothesis is confirmed, then the STTR and the average sentence length of STC_EN will present more restricted intervals (less variation = more homogeneity) compared with STC_ES.
As far as lexical variety is concerned, Figure 1 shows the trend in source texts (STC_ES): each point in the graph indicates the STTR of each judgment.  As far as the English translated texts are concerned, the maximum value is 36.82 and the minimum 30.47 (interval equal to 6.35). The comparison between the two datasets is shown in Figure 3. It shows that the trend is more homogeneous in STC_EN, which seems to confirm the levelling out hypothesis applied to STTR:  The global picture is shown in Figure 6 which compares the two datasets. The comparison between the two subcorpora reveals a slightly more homogenous trend in STC_EN (translated texts), which is in line with the levelling out hypothesis. It is also interesting to observe that the values of STC_EN are almost always higher than their source texts, which matches the simplification universal. In this sense, the hypothesis tested in the parallel corpus seems to be confirmed.

Untypical collocations
As a T-Universal, the untypical collocation hypothesis has been tested on the most frequent type of collocation found in the corpus namely Verb + Noun (both as subject and object) (based on Corpas Pastor 1996's classification of phraseological units).
No cut-off frequency has been established since one of the aims of the analysis is to spot untypical/infrequent patterns (hapax legomena have also been considered for the purpose of the study). The selection of the nodes -based on Sketch Engine's Wordlist toolof the collocations is based on three criteria: a) judicial terms are common to both STC_EN and UKSC; b) they are close to a verb (span: +/-3 words on the left and right); c) they meet the two previous criteria and have a raw frequency of at least 100 in order to be tested later on the basis of their syntactic function: V + Subject / V + Object). The resulting nodes have been used for the quali-quantitative analysis based on the Concordance function of Sketch Engine: appeal, appellant, argument, case, court, decision, judgment, law, provision e right. The departing point has been obviously the STC_EN subcorpus used as a testbed to spot untypical collocations. Table 14 shows the results of the quantitative analysis. A distinction should be made between infrequent and atypical collocations: the former are collocations which tend to be less used in STC_EN compared to UKSC, while the latter are patterns which are indeed rare or untypical in Mauranen's (2008: 44) acception (rare or absent from actual target language texts, even though they are possible in the TL system).
The final results show that 201 out of 1,732 collocations are infrequent legal English (UKSC), corresponding to 19.69% of the whole collocational patterns. Untypical collocationswhich the qualitative analysis has revealed to be an effect of some interference with the source language (Spanish)amount to 15 types with a total of 27 occurrences, representing 1.56% of the collocations (see Biel 2014: 113-114).
Although there are instances of this phenomenon in the translated corpus (STC_EN), the results are not robust and therefore do not provide a solid picture able to (quantitatively) confirm the universal.

Interference
In order to test the final universal (interference), a typical discursive trait of Spanish judicial language has been analysed, namely antinormative/incorrect gerunds (see Pontrandolfo 2019b). Biber et al. (1999: 198) and Swan (2005: 378) provide specific indications on the correct use of -ing froms in English both as gerunds (when the -ing form has the function of noun) and present participles (having the function of verbs, adjectives and adverbs); these guidelines have been followed to identify interference cases in the translated corpus (STC_EN).
If the interference universal takes place in the corpus, then the STC_EN will contain negative influences from the source language transferred to the target texts in the process of translation (see Zanettin 2012: 21).
The test has been carried out on the TC corpus. The first step has been isolating all the gerunds in STC_ES by means of the Concordance tool in Sketch Engine. 454 gerunds out of 1,014 have been considered as antinormative (based on a manual/qualitative analysis). The subsequent step was using the Parallel Concordance tool in Sketch Engine to exclude the cases in which the gerund had been reformulated in the target text (these are cases of re-elaborative effort on the part of the translator, Toury 1975) and end up with the gerunds which are actually incorrect (because of a negative interference) in English. They have been classified following Pontrandolfo (2019b: 729-730 based on CMLJ 2011: 119 122)'s typologies: 1) BOE's gerunds (gerund with the function of a relative clause); 2) gerund of posteriority (which expresses an action which has developed after the action expressed in the main clause); 3) ilative gerund (functioning as a simple coordinated sentence); 4) consequence gerund (expressing an action which is the result of the main sentence). Table 15 presents an overview of the anti-normative gerunds in STC_ES and STC_EN: 166 result from reformulation whereas 287 have been considered as potentially incorrect. Among the 287 -ing forms detected in the corpus, some of them are clearly incorrect and are typical examples of negative interference (e.g. El título II regula la sucesión de ordenamientos y administraciones públicas, formando parte de su contenido la integración de […] > Title II regulates the succession of Regulations and Administrations, being part of its content the incorporation of […] where the illative gerund in the ST is replicated in the TT thus resulting in an incorrect sentence in English). Other are ambiguous constructions or stylistically improvable sentences.
The analysis has confirmed that 50% of the Spanish gerunds (453/1,014) are not antinormative and that 36.6% of them (166/453) had been reformulated in translation. The remaining cases (63.4%) can be classified as literal translations, i.e. examples in which translators opt for repeating the -ing form in their translations. However, the full-blown incorrect cases are generally limited in number and therefore the universal is not robustly confirmed.

Discussion and final remarks
The pilot study presented in this paper is not exhaustive in its design and results, and it is only based on a small-scale study conducted in a specific language combination (Spanish-English) and within a specific genre (judgments). However, it has proven the feasibility of the tests which can be replicated in the future with larger corpora, different language combinations as well as different genres.
The results presented in this paper seem to point to the existence of the translation universals identified in the literature, although the quantitative results are not solid enough to confirm the hypotheses.
The TC corpus, and more precisely the English subcorpus (STC_EN), has been used as testbed to verify S-Universals whereas the UKSC corpus has been used as a reference corpus to compare translated (STC_EN) and non-translated/original English (UKSC) in T-Universals.
As far as simplification is concerned, lexical variety and density slightly confirm the tendency of translators to simplify source texts, whereas the average sentence length does not seem to follow this direction.
As for explicitation, results show that STC_EN contain a higher percentage of linking adverbials compared to STC_ES which would point to translated texts being more explicit than their source texts. Moreover, original texts (UKSC) contain a more varied use of discourse markers which could indicate that translators tend to repeat the same linking adverbials, thus simplifying the overall structure of the text.
As far as normalisation is concerned, the comparison between the two subcorpora on a specific discursive element (complex prepositions) shows that the phraseological patterns in STC_EN are also present in the most frequent patterns in UKSC. This could confirm the normalisation hypothesis. In general, the most frequent complex prepositions common to both subcorpora are also the most frequent in STC_EN, which would suggest a tendency to adhere to the textual conventions of the target language.
The levelling out hypothesis has been tested on STTR and average sentence length and the comparison between the two subcorpora reveals a slightly more homogenous trend in STC_EN (translated texts), which is in line with the levelling out hypothesis.
With regards to untypical collocations, the final results show that 19.69% of the whole collocational patterns are infrequent in original legal English.
The analysis related to interference has confirmed that 63.4% of antinormative gerunds in ST have been translated literally in the target texts but full-blown incorrect cases are limited in the overall picture and therefore the universal has not been confirmed robustly.
The study demonstrates that these regularities in translations tend to be interrelated (see the case of simplification and explicitation) and that more systematic tests should be carried out to confirm or disconfirm these universals in legal translation.
As far as this pilot study is concerned, it has been characterised by a quantitative approach. Future studies can explore the qualitative dimension as well, as an effective way to counterbalance the preliminary results.
The results obtained are useful both for the professional and training perspective, as mentioned in the introduction of this paper.
The untypical collocation hypothesis has confirmed the importance of distinguishing between a variant and an alternative (see Göpferich 1995) 11 : not all the translation solutions adopted by legal translators/trainees, although correct from the viewpoint of grammar (alternative), are acceptable and in line with the discursive conventions (variant) (see Garzone 2007: 218-219). The same applies to the universal of interference: the risk of being negatively influenced by the structuring of the sentences in the source texts can play a major role in the acceptability of the final result.
Future studies on larger corpora and different language combinations can definitely help taking further steps in this promising area of legal translation studies. 11 An alternative consists of a linguistic form that is not conventional but that can accomplish the communicative goal whereas a variant is the reduced array of variation that is accepted to replace any given convention. According to Göpferich (1995), translators have to be acquainted not only with the most conventional features in any given genre, but also with the possible variants.