Cross-lingual Explicitation of Implicit Discourse Relations

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 217 papers37 conceptsfrozen 2026-10-04

When we communicate, we rarely state every logical connection. We might say, "It started raining; I opened my umbrella." We don't need to say "because" for a listener to understand the causal link. These unstated links are implicit discourse relations.

The difficulty for machine translation (MT) is that different languages have different "densities" of these markers. As Frances Yung notes in Towards a discourse relation-aware approach for Chinese-English machine translation, languages like Chinese are often "implicit-heavy," relying on context and juxtaposition. In contrast, English is more "explicit-heavy," requiring transitional words to feel cohesive. When an MT system translates literally, the result often feels "broken" or disjointed because the "connective tissue" of the discourse is missing.

The Explicitation Hypothesis

In translation studies, the Explicitation Hypothesis suggests that translators tend to make the implicit explicit. When a human translates from a paratactic language (one that favors short, unlinked clauses) to a hypotactic one (one that favors complex, linked structures), they naturally insert connectives like "but," "because," or "although" to satisfy the target language's grammatical expectations.

Research by Muhammad Alif Redzuan Abdullah and colleagues in Discourse Reconstruction in the English Translation of Sunzi’s The Art of War illustrates this in classical texts. Translators of Sunzi’s work must frequently "fill in" omitted subjects and logical conjunctions to make the ancient Chinese text readable for modern English audiences. This isn't just about adding words; it's about reconstructing the rhetorical structure of the original thought in a new linguistic environment.

Triangulating the Unspoken

One of the most significant hurdles in discourse processing is the lack of labeled data. It is hard to train a model to recognize a "hidden" causal link if it isn't written down. Researchers have turned to cross-lingual explicitation as a clever solution to this data scarcity.

The intuition is simple: if a logical relation is implicit in Language A but made explicit by a human translator in Language B, we can use Language B as a "gold label" for Language A. Wei Shi, Frances Yung, and Vera Demberg pioneered this in Acquiring Annotated Data with Cross-lingual Explicitation for Implicit Discourse Relation Classification. By looking at how English sentence pairs were translated into French, German, or Czech, they could "triangulate" the intended meaning. If multiple translators independently added a contrastive marker (like "however"), it provides a high-confidence signal that the original English pair contained an implicit contrast.

This method has even been used to bootstrap resources for low-resource languages. In Semi-automatic discourse annotation in a low-resource language, Marian Marchal and colleagues used English translations to build the first discourse connective lexicon for Nigerian Pidgin, demonstrating that the underlying semantic links are universal even when the surface markers differ.

Sharpening and Shifting

Explicitation isn't just about adding missing words; it's also about sharpening vague ones. Frances Yung and Ekaterina Lapshinova‐Koltunski found in Investigating Explicitation of Discourse Connectives in Translation using Automatic Annotations that translators often replace generic connectives like "and" with more specific ones like "furthermore" or "consequently." This is particularly common when moving from spoken source texts to written translations, where the reader lacks prosodic cues (like tone of voice) to help them infer the logic.

However, this process is not perfect. The DiscoGeM 2.0 project, led by Sárka Zikánová, reveals that discourse relations are not always preserved during translation. A translator might interpret a subtle relationship differently than the original author intended, leading to a "shift" in the rhetorical structure. This suggests that while translation is a powerful lens for seeing implicit logic, it also introduces its own layer of interpretation.

Improving Machine Translation Flow

The ultimate goal of this research is to move MT beyond sentence-by-sentence accuracy toward document-level cohesion. Mei Tu and Yu Zhou demonstrated in Enhancing Grammatical Cohesion: Generating Transitional Expressions for SMT that by explicitly modeling the structural relationships between clauses, systems can generate the necessary "transitional expressions" that make English output readable.

While early experiments by Kevin Duh in Crosslingual Annotation and Analysis of Implicit Discourse Connectives for Machine Translation showed that simply forcing connectives into the source text doesn't always boost automated scores like BLEU, the consensus remains that discourse-aware models are essential for human-like fluency. By teaching machines to see the "unspoken" logic between sentences, we move closer to translations that don't just convey the facts, but also the flow of the argument.

Papers in this version

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 24 papers
  2. Version 1 · 2026-10-04 · 17 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_implicit_discourse_relations_8de1d6b3,
  title = {implicit discourse relations},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/implicit-discourse-relations/rdr_8de1d6b31e894f4f8be2e1cb610fc1f5},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_cross_lingual_explicitation_of_implicit__8de1d6b3_v1,
  title = {Cross-lingual Explicitation of Implicit Discourse Relations},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/cross-lingual-explicitation-of-implicit-discourse-relations/rdr_8de1d6b31e894f4f8be2e1cb610fc1f5/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}