What it is
A string that comes back is not thereby a unit. Recurrence has several causes and only one of them is that a community of writers has settled on a wording. A drafting policy produces it, a template produces it, one prolific author produces it, and so does a requirement that a body's texts agree with each other.
The case that matters for translation is the archive of translated text. It is large, it is searchable, it is the first place anyone looks, and everything in it was produced under the constraints of translation rather than those of drafting. A pattern read off it describes what translators into that language have done. It is evidence, but not of the thing it is usually taken for.
The terminology section keeps the acceptability a standardizing body confers on a designation, and keeps the competition between designations for one concept. What is at issue here is the evidential status of a recurrent string, which nobody confers and somebody measures. The Polish connector case that section already carries is not repeated here.
Where the fixity comes from
Nothing, and that is the finding. What can be shown is what recurrence in translated text has been measured to mean, and both studies that show it were run on Polish.
- Koźbiał, comparing judgments of the Court of Justice with judgments of the Polish Supreme Court, confirms that the high distribution of prepositions in judgments measured against a representative sample of general Polish is a generic feature of judgments, and finds that the phraseological profile of EU judgments differs from that of national judgments in the distribution of prepositions and in their functions, which may result from the impact of the translation process on target-language texts.
- Biel, Koźbiał and Wasilewska examine four EU genres in Polish: legislation, judgments, reports and websites. Translations show increased aggregate formulaicity in bundle tokens for every genre except judgments; the overlap of bundles between translations and non-translations is consistently low; and translations develop formulaic profiles of their own, levelled out against the EU English corpora and overlapping minimally with domestic genres. Their bundles are admitted at 40 occurrences per million words, and a second threshold left the question open.
Both findings are about Polish and both state their direction. Neither licenses a claim about another language, and no figure from either study's tables appears on this page.
Consequences for translation
The reviser's test is short and needs no corpus: a pattern is evidence of the target community's usage only if it is attested in text that community drafted.
- An archive of translations is evidence of translation practice. Where the question is what the target community writes, it is the wrong corpus, however closely it matches the job in hand.
- A genre-level finding licenses nothing about a single combination. That a genre is more formulaic in translation says nothing about whether this sequence is fixed.
- Recurrence inside one document may be a drafting requirement rather than a unit, and the two are indistinguishable from outside the document.
- Where a body requires continuity with its own archive, following the archive is right and is still not evidence about the language. Both hold at once, and a line in the file is cheaper than the argument later.
- Both link tiers are empty because nothing is done to a stretch of text here. This record is a test applied to evidence, not an operation performed on wording.
Examples
EN>PL
- Source: Member States shall encourage, by any means which they consider appropriate
- Target: Państwa członkowskie wspierają wszelkimi środkami, które uznają za właściwe
- Comment: Directive 2008/52/EC, Articles 4(1) and 9, where the English opening recurs word for word and the Polish repeats it word for word. A count over the act returns a recurrent sequence in both versions. What produced it is the source text and the requirement that an act agree with itself, and the studies behind this record found EU Polish overlapping minimally with domestically drafted Polish.
EN>NL
- Source: Member States shall encourage, by any means which they consider appropriate
- Target: met alle door hen passend geachte middelen at Article 4(1), op een wijze die zij passend achten at Article 9
- Comment: The same two provisions. Dutch does not repeat itself, so one English recurrence yields a recurrent sequence in one target version and none in the other. Whatever is being counted belongs to the version rather than to the wording, and a reviser calling the Dutch inconsistent would be right about the act and wrong about the language.
References
- Baker, M. (1993). "Corpus linguistics and translation studies: implications and applications." In Text and Technology: In Honour of John Sinclair. Amsterdam: John Benjamins.
- Laviosa, S. (1998). "Core patterns of lexical use in a comparable corpus of English narrative prose." Meta 43(4).
- Koźbiał, D. (2019). "Prepositions in EU and domestic judgments: a corpus analysis of the distribution and functions of simple, complex and secondary prepositions." Comparative Legilinguistics 35, 89 to 119, published 3 January 2019. In Polish.
- Biel, Ł., Koźbiał, D. and Wasilewska, K. (2019). "The formulaicity of translations across EU institutional genres: a corpus-driven analysis of lexical bundles in translated and non-translated language." Translation Spaces 8(1), 67 to 92.
- Directive 2008/52/EC of the European Parliament and of the Council of 21 May 2008 on certain aspects of mediation in civil and commercial matters, Articles 4(1) and 9, English, Dutch and Polish language versions.