Discourse in Translation
The goal of this article is to introduce the relevance of discourse in translation and discuss its current status in machine translation…
Discourse in Translation
The goal of this article is to introduce the relevance of discourse in translation and discuss its current status in machine translation and human translation studies as well as its open challenges.
Why am I explicitly considering machine translation and human translation perspectives?
In human translation studies, quality is often viewed broadly, taking into account specific contexts, users, and purposes. It looks not only at the final translation, but also at the process and social aspects of translation. In contrast, machine translation (MT) treats translation as a narrowly defined task with strict rules and specialized evaluation methods. This difference creates a bias: MT focuses on certain aspects of translation while overlooking others, simply because of how the task is defined [1].
Discourse Definition:
In the definition of discourse, Jurafsky and Martin [2, chapter24] mention:
“Like movies, language does not normally consist of isolated, unrelated sentences, but instead of collocated, structured, coherent groups of sentences. We refer to such a coherent structured group of sentences as a discourse”
They also define discourse models as [2, chapter 23]:
“A discourse model is a mental model that the understander builds incrementally when interpreting a text, containing representations of the entities referred to in the text, as well as properties of the entities and relations among them.”
In their comprehensive Discourse Reader, Jaworski and Coupland [3] discuss ten definitions of discourse, summarized by Schiffrin et al. [4] as:
(1) anything beyond the sentence (from a linguistics tradition)
(2) language use (from sociolinguistics)
(3) a broad range of social practices that construct power, ideology, etc. (from critical theory).
Paltridge [5,6] defines discourse analysis as:
“Discourse analysis examines patterns of language across texts and considers the relationship between language and the social and cultural context in which it is used. Discourse analysis also considers the ways that the use of language presents different views of the world and different understandings. It examines how the use of language is influenced by relationships between participants as well as the effects the use of language has upon social identities and relations. It also considers how views of the world, and identities, are constructed through the use of discourse.”
(A) Machine Translation:
Discourse Phenomena in Machine Translation Studies:
Discourse translation has been a relevant topic since the earliest works in machine translation using rule-based systems, then statistical machine translation systems [7], neural machine translation systems [12], and general purpose large language models [25].
The relevance of discourse translation stems from the difficulty of handling discourse phenomena across long pieces of texts (typically referred to as documents).
Below, I present very simple examples of relevant discourse phenomena in machine translation from our daily lives where errors might be critical (this is not an exhaustive list) [23]:
1- pronoun resolution
English: I have two sisters. They are my best friends in the world.
Arabic: لدي أُختان. إنهما صديقتاي المفضلتان في العالم.
In this example, the pronoun “they” refers to “two sisters”. In Arabic, dual subjects have a different formula for the pronoun compared to plural subjects. Looking at the context beyond the sentence is necessary to disambiguate the pronoun.
2- lexical cohesion
English: I’ve been struggling and grappling with questions about the nature of God. I’m very aware that when you say the word “God,” many people will turn off immediately.
French: J’ai lutté et j’ai été aux prises avec des questions sur la nature de Dieu. Je suis parfaitement conscient que lorsqu’on prononce le mot de “Dieu”, nombreux sont ceux qui se détournent immédiatement.
The entity “God” is mentioned more than once and it’s important that it is translated consistently across the document.
3- formality
English: How are you my dear friend? Would you like to go to the cinema with me?
German: Wie geht es dir, mein lieber Freund? Möchtest du mit mir ins Kino gehen?
In German, the pronoun “you” has two translations based on your relationship to the listener: du (informal) and Sie (formal). The context “my dear friend” enables disambiguating the pronoun.
4- verb forms
English: Maria said she was too sick. However, she was seen walking in the park.
Portuguese: A Maria disse que estava muito doente. No entanto, ela foi vista a passear no parque.
The verb “seen” has two forms depending on the gender of the object. The context Maria allows disambiguating the verb and selecting the correct feminine form.
Discourse Research in Machine Translation:
Research efforts in discourse in machine translation have spanned a great deal of areas, including:
- data: creating test-suites and benchmarks [8,9].
- evaluation: building metrics and evaluation tools [10,11].
- modelling: developing architectures specific for discourse translation.[12,13]
- interpretability: interpreting the role of context in document translation.[14]
- multimodality: assessing the role of multimodal information in handling discourse phenomena.[15]
- reinforcement learning: reward modelling to improve discourse performance.[16]
Open Challenges:
In my opinion, these are a few open challenges that are worth exploring in order to push forward the research in discourse in machine translation:
- a global evaluation protocol: there is still no universally accepted framework or metric to evaluate model’s performance on discourse phenomena in machine translation. Many efforts are either phenomenon-specific or language-specific.
- language and culture coverage: there is still a lot of work to do to identify the relevant language-specific discourse phenomena specially for low-resource languages.
- downstream applications: the implications of discourse phenomena performance on downstream tasks that use the translation output is still an open question.
(B) Human translation:
Definition of Translation Quality:
Translation quality has been a source of debate in translation studies for decades [14], since it is considered highly subjective and dependent on how translation and quality are defined. One common denominator is the central role played by accuracy and fluency, a view shared by the field of machine translation [17].
Koby et al. [14] provide two definitions of translation quality, a narrow one and a broad one.
Narrow definition:
“A high-quality translation is one in which the message embodied in the source text is transferred completely into the target text, including denotation, connotation, nuance, and style, and the target text is written in the target language using correct grammar and word order, to produce a culturally appropriate text that, in most cases, reads as if originally written by a native speaker of the target language for readers in the target culture.”
Broad definition:
“A quality translation demonstrates accuracy and fluency required for the audience and purpose and complies with all other specifications negotiated between the requester and provider, taking into account end-user needs.”
Discourse Phenomena in Human Translation Studies:
In addition to the phenomena discussed in MT studies, the Skopos theory [19], proposed to focus on preserving the purpose of the source text in the translation. House [20] deems it difficult to exactly determine the purpose and proposes to divide a text into register and genre, each further subdivided, for a detailed analysis of category based equivalence. With more attention on the recipient of the translation, criteria such as readability and comprehensibility were introduced. For instance, Gopferich [21] proposes several dimensions of comprehensibility, that is, concision, correctness, motivation, structure, simplicity, and perceptibility [17].
Discourse Research in Human Translation Studies:
Munday and Zhang [5] provide a thorough categorization of research works on discourse analysis in translation studies:
Open Challenges:
While translation studies look into much broader aspects of quality compared to MT, some challenges still persist:
- integrating MT technologies in human translators’ workflows. [24]
- designing a systematic catalog of translation quality definitions, criteria, and evaluations of their measurability. [17]
- Discourse analysis in translation studies often draws from linguistics, critical discourse analysis, and sociological approaches, but integrating these perspectives systematically remains a challenge. This limits the ability to fully capture how social, cultural, and pragmatic factors shape translation choices. [5,22]
I hope you enjoyed reading the article and I’m looking forward to reading your thoughts about the relevance of discourse translation in the common era of language technologies in the comments section. Let’s grow and learn together through fruitful discussions :)
References:
Disclaimer: the goal of this article is not to serve as an exhaustive survey of works in discourse translation, but rather to highlight the main research directions to the best of our knowledge. Please let us know if there are relevant works in new areas that we missed :)
Acknowledgement: huge thanks to Cristina Meza Castro for sharing resources and giving feedback on the article.
[2]https://web.stanford.edu/~jurafsky/slp3/
[3]https://www.academia.edu/9827608/The_Discourse_Reader_Jaworski_Adam_Coupland_Nikolas_1
[4]https://repository.dinus.ac.id/docs/ajar/discourse-analysis-full.pdf
[5]https://repository.dinus.ac.id/docs/ajar/Discourse-Analysis-in-Translation-Studies.pdf
[6]https://www.bloomsbury.com/uk/discourse-analysis-9781350093638/
[8]https://aclanthology.org/2023.acl-long.435/
[9]https://arxiv.org/abs/2004.14607
[10]https://arxiv.org/abs/2208.09118
[11]https://aclanthology.org/2023.eacl-main.278/
[12]https://aclanthology.org/N18-1118/
[13]https://aclanthology.org/W17-4811/
[14]https://aclanthology.org/2024.findings-eacl.113/
[16]https://arxiv.org/abs/1811.05683
[17]https://aclanthology.org/2023.eamt-1.37.pdf
[18]https://ddd.uab.cat/pub/tradumatica/tradumatica_a2014n12/tradumatica_a2014n12p413.pdf
[21]https://www.fachportal-paedagogik.de/literatur/vollanzeige.html?FId=2736295
[22]https://scispace.com/pdf/pragmatic-and-sociocultural-adjustments-in-translation-w698c587tw.pdf
메타데이터
- post_id
- c27093f5378d
- slug
- discourse-in-translation-c27093f5378d
- url
- https://medium.com/@wafamoh97/discourse-in-translation-c27093f5378d
- canonical_url
- https://medium.com/@wafamoh97/discourse-in-translation-c27093f5378d
- author_url
- https://medium.com/@wafamoh97
- status
- ok
- fetched_at
- 2026-06-22 05:41:33
