Lexical Bundles in European Parliament English: How Interpreters, Translators and MEPs Use Four-Word Formulas Differently (Wu, 2021, Open Access)

Conference English #2: Four-word Lexical Bundles in European Parliament Conference English

Summary. This open-access article by Yinyin Wu (Graduate Program in Translation and Interpretation, National Taiwan University) compares four-word lexical bundles in three kinds of English produced at the European Parliament (EP). The three registers are simultaneous interpretations into English, written translations of the same speeches, and original English speeches. Using the European Parliament Interpretation Corpus (EPIC) and WordSmith Tools, the study identified 892 bundle types. Only about 10% of them occur in two or all three registers. Noun-phrase and prepositional-phrase bundles dominate in all three. Interpreted English stands out functionally: stance, referential and subject-specific bundles each make up about a third of its bundle types. Obligation bundles such as we have to account for almost half of its stance bundles.

The article at a glance

• Title: Lexical Bundles in English EU Parliamentary Discourse—Variation Across Interpreted, Translated, and Spoken Registers
• Chinese title: 歐盟議會英語語篇之詞串研究—口譯、筆譯及演說之語域差異
• Author: Yinyin Wu (吳茵茵), National Taiwan University
• Journal: Compilation and Translation Review (編譯論叢), Vol. 14, No. 2 (September 2021), pp. 37–86
• DOI: 10.29912/CTR.202109_14(2).0002
• Access: Open access [https://naeraj.naer.edu.tw/wSite/PDFReader?xmlId=2057911&fileName=1631174409649&format=pdf]
• Data: EPIC, plenary sessions of February 2004
• Method: Corpus-driven identification of four-word bundles, followed by structural and functional analysis
• Funding: Ministry of Science and Technology, Taiwan (MOST 107-2410-H-002-255)

What are lexical bundles?

Lexical bundles are word sequences that recur frequently in a body of text. Examples are on the basis of, in the light of and when it comes to. They are not idioms. Their meaning is transparent and they are usually not complete grammatical units. Even so, they work as the building blocks of discourse. Because software identifies them by frequency, they offer an objective way to study formulaic language. Four-word bundles are the most widely studied length: three-word sequences are too numerous to analyse, and five-word sequences are too rare.

Why compare interpreters, translators and MEPs?

The study builds on Line Henriksen's article "The song in the booth: Formulaic interpreting and oral textualisation" (Interpreting, 9(1), 2007). Henriksen borrowed Milman Parry's concept of the Homeric formula to describe simultaneous interpreting in the EU. She asked ten Danish staff interpreters at the European Commission to interpret the same speeches. Some of them rendered recurrent source-text ideas with the same stock Danish formulas, glossed in English as "take the lead" and "lead the way", whatever the English wording. Henriksen argued that formulas spread among colleagues until a booth's output becomes uniform. She compared this to the "oral textualisation" of the Homeric poems. In her view, diplomats, officials, translators and interpreters all take part in weaving the text of EU discourse.

Wu (2021) extends Henriksen's argument in three ways:

1. It examines the English produced by interpreters, translators and MEPs, not by interpreters alone.
2. It operationalizes formulaic language as lexical bundles identified computationally, not chosen by the researcher.
3. It uses authentic plenary data, not renditions elicited in an experiment.

There is also a teaching motivation. In Taiwan, most interpreters and translators are non-native speakers of English but often have to work into English as a B language. Most English-booth interpreters and English translators in the EU are native speakers. Their output is therefore a useful model for training.

What data were used?

The study compares three corpora, 157 texts and 79,899 words in total:

• Interpretation corpus: 38 simultaneous interpretations into English from Spanish and Italian (18,611 words).
• Translation corpus: the official English translations of the same 38 source speeches (20,577 words).
• Speech corpus: 81 original English speeches delivered in the plenary (40,711 words).

The interpretations and speeches come from EPIC (Russo et al., 2005). They were recorded at the plenary sessions of 10, 11, 12, 25 and 26 February 2004. The translations come from the English verbatim reports published by the EP. According to Bernardini et al. (2016), those translations were produced independently, without reference to the interpreters' output. Hesitations, repetitions and fillers were removed from the transcripts so that the three corpora could be compared fairly. Politics, justice, and economics and finance are the main topics in all three corpora. The speech corpus also includes many speeches on health.

How were the bundles identified and classified?

The corpora are small and of different sizes, so raw frequency thresholds were used:

• Interpretation corpus: bundles occurring at least twice (108 per million words), yielding 302 bundle types and 704 occurrences.
• Translation corpus: bundles occurring at least twice (97 per million words), yielding 364 bundle types and 918 occurrences.
• Speech corpus: bundles occurring at least three times (74 per million words), yielding 226 bundle types and 922 occurrences.

In total, 892 bundle types with 2,544 occurrences were analysed. All three thresholds are above 40 per million words, the highest cut-off used in previous lexical bundle studies.

The structural classification follows Biber et al. (2004), as adapted by Jablonkai (2010). It has four categories: verb-phrase-based bundles, dependent-clause bundles, noun-phrase/prepositional-phrase-based bundles, and bundles with adjectives, adverbs or numbers.

The functional classification follows Biber et al. (2004): stance bundles, discourse organizers and referential bundles. It adds the category of subject-specific (EU-related) bundles from Jablonkai (2010) and several subtypes from Simpson-Vlach and Ellis (2010), such as evaluation, hedges, discourse markers and metadiscourse. The concordance lines of every bundle were checked to decide its main function in context.

How many bundles do the three registers share?

Very few. Differences far outweigh similarities in actual bundle use.

• Interpretations and translations of the same speeches share only 58 bundle types: 19% of the interpretation bundles and 16% of the translation bundles.
• Interpretations and speeches share 35 types. Translations and speeches share 37 types.
• 92 bundle types occur in two or all three corpora. They make up only 10% of all bundle types but 30% of all bundle occurrences (768 of 2,544).
• 19 bundle types occur in all three corpora. They make up 2% of the types but 11% of the occurrences (289).

The translation corpus has more bundle types than the interpretation corpus (364 vs. 302). Translators work under less time pressure and can vary their wording more. This is in line with Bernardini et al. (2016), who found interpreted EP texts simpler than their translated counterparts.

The 19 bundles found in all three registers are: I would like to, of the European Union, the European Union and, in the European Union, on the basis of, the rule of law, that the European Union, the implementation of the, of the Lisbon strategy, the end of the, the fight against terrorism, when it comes to, it is important to, the Commission and the, the Council and the, in the Member States, of the European Parliament, in the fight against, with the European Union.

What structural patterns did the study find?

All three registers show the same order of structural categories:

• Noun/prepositional phrase-based bundles are the largest category: 49% of bundle types in interpretations, 56% in translations and 64% in speeches.
• Verb-phrase-based bundles come second: 33% in interpretations, 25% in translations and 20% in speeches.
• Dependent-clause bundles are similar across registers: 16%, 17% and 15%.
• Other bundles (with adjectives, adverbs or numbers) make up only 1–2%.

This is the same order as in Jablonkai's (2010) study of EU documents. The dominance of noun- and prepositional-phrase bundles marks EU parliamentary English as formal and informational, like academic English.

Within that shared pattern, interpreted English is the most oral of the three: it has the highest share of verb-phrase bundles. The speeches are the most literate: they have the lowest share of verb-phrase bundles and the highest share of noun/prepositional-phrase bundles. This is less surprising once we know how the speeches were delivered. 53% were read out, 17% were partly read out, and only 30% were impromptu.

What functional patterns did the study find?

Interpreted English is functionally distinctive. Referential bundles make up 32% of its bundle types, stance bundles 30%, subject-specific (EU-related) bundles 30% and discourse organizers 8%. With three categories at about a third each, it combines communicative priorities (stance) with informational ones (referential and subject-specific). This places it in the middle of a continuum, close to academic spoken registers such as classroom teaching.

Translations and speeches are more alike. Subject-specific bundles are the largest category in both: 37% in translations and 50% in speeches. They are followed by referential bundles (33% and 24%), stance bundles (21% and 20%) and discourse organizers (9% and 6%). Translations and speeches are more informational than interpreted English, but less so than EU documents, where stance bundles make up only 3% of bundle types (Jablonkai, 2010).

What makes interpreted English distinctive?

• Obligation bundles dominate. Obligation/directive bundles make up almost half of the stance bundle types in interpreted English (41 of 91). Interpreters use we have to, we need to, we must and we've got to more than translators and MEPs do. These bundles emphasize the action to be taken. They are also light words that give the interpreter a little time.
• It is always "we", never "you". In all three EP corpora, obligation bundles never contain you. In university teaching, by contrast, they often do: you have to be, you might want to (Biber et al., 2004). Lecturers speak with authority over students. MEPs, and the interpreters and translators who serve them, speak as equals.
• Desire is expressed in one way. Every desire bundle in interpreted English contains want: I don't want to, we want to do. MEPs express desire in many ways: I hope that we, we look forward to, I welcome the opportunity.
• Simpler sentence frames. Eight of the 34 focus bundles in interpreted English are simple frames with that is, it is or this is, such as that's not the way and it's not just a. The translation and speech corpora each contain only two such focus bundles.
• "On the one hand" without "on the other hand". Interpreters used on the one hand but never followed it with on the other hand. Under time pressure, they may forget the first half of the frame or switch to a shorter word such as but.

What is distinctive about translated and spoken English?

Translated English contains no desire bundles, no discourse markers and no metadiscourse bundles. It is the only register with text-deixis bundles such as as stated in the, and the only one with non-causal elaboration bundles such as that is to say. Its topic-introduction bundles are the most varied, for example should like to highlight, the third priority is and to point out that.

MEP speeches look the most literate in structure, but their functions still show oral features. Desire bundles make up a quarter of their stance bundle types. They contain discourse markers such as thank you very much and metadiscourse bundles, and they have no text-deixis bundles.

What does this mean for interpreter and translator training?

• Prioritize shared bundles. Interpreting trainees can focus on bundles found in both interpretations and speeches, to support their comprehension and production of conference English. Translation trainees can focus on bundles found in both translations and speeches, because translators sometimes work with scripted political speeches.
• Teach by function. A functional typology shows trainees which chunks belong to which register and what those chunks do there: expressing obligation, introducing a topic, framing an abstract concept, referring to an EU institution, and so on.
• Teach complete chunks. Overlapping bundles can be merged into longer patterns with variable slots, such as (in) the fight against terrorism / drugs / international organised crime. These are more complete in meaning and easier to memorize.

What are the limitations?

• The three corpora are small compared with those of most corpus-driven studies. Patterns in larger corpora need further study.
• Overlapping bundles may inflate counts. For example, in the fight against and the fight against terrorism often belong to the same five-word sequence.

Future research could examine lexical variety, lexical density, collocations and sentence length in EP English. It could also compare interpreted, translated and spoken registers in other languages, such as Mandarin Chinese.

Related posts in the Conference English series

• Phrasal verbs in European Parliament conference English (Wu, 2023, The Interpreter and Translator Trainer, 17(2))
• Multi-word constructions in European Parliament conference English (Wu, 2024, The Interpreter and Translator Trainer, 18(3)), which builds on this lexical bundle study

How to cite

Wu, Y. (2021). Lexical bundles in English EU parliamentary discourse—Variation across interpreted, translated, and spoken registers. Compilation and Translation Review, 14(2), 37–86. https://doi.org/10.29912/CTR.202109_14(2).0002

About the author: Yinyin Wu teaches at the Graduate Program in Translation and Interpretation, National Taiwan University, Taiwan. This is the second post in the Conference English series, which moves from two-word phrasal verbs to longer word sequences and the rhetorical structure of whole speeches.