Finding Interesting Papers in ACL Workshop Proceedings

NLP workshops at ACL venues have a rich tradition. Topics range from representation learning to text technologies for social sciences and speech communities, including meta-research such as teaching NLP. Some workshops span many years and sometimes metamorphose into conferences (e.g., the conference of machine translation -WMT-, where the “W” indicates its provenance).

Digging through workshop volumes can be quite entertaining. Since workshops are also usually a bit smaller in size, you don’t have to crawl through a mountain of papers - simply find a workshop with a topic that interests you, and then explore the papers! (Since the workshop landscape is quite dynamic, individual venues are best inspected on a per-year basis in the ACL anthology, e.g., see this link for all ACL-related workshops of 2026.)

My impression is that workshop papers tend to be perceived as of slightly lower value when compared with conference papers. Expectations towards meeting certain conventions may indeed be slightly lower for workshop papers, but that doesn’t need to be a downside. Instead, it makes those papers often a bit less dense, as it cuts the authors a bit of slack, which can, e.g., help allow for presentation of more free and creative ideas.

Three Recent Workshop Papers that Stuck

Here’s a small sample of workshop papers that I found very interesting:

On Tables with Numbers, with Numbers (Kogkalidis & Chatzikyriakidis, LM4UC 2025). This paper questions the obsession with numbers in empirical sciences and argues that (in NLP) it sometimes goes way too far. A problem is that in the fog of numbers, their “objective appearance” makes us overlook what they actually represent, and we may forget to think about whether they even represent a sensible and useful measurement. I agree with this! The paper also argues that the infatuation with numbers privileges high-resource labs that can run many experiments (including many redundant ones). And finally it pushes our critical thinking even a bit further, with an extremely provocative claim: “Tables are irrelevant”. Agree, or not - It’s good food for thought!
Degree Zero of Translation: Using Interlinear Baselines to Quantify Translator Intervention (Rapacz & Smywiński-Pohl, LaTeCH-CLfL 2026). My take from this paper is that the time has come where we should focus less on whether a translation is correct, but also take other features into account. This makes much sense to me, when we accept that many LLMs are good translators, but they’ve all different styles and quirks. The concrete paper proposes a radical metric of translator intervention. The metric is qualitatively neutral, by which I mean non-judgemental: a translation can have a large amount of intervention and be considered highly useful for some audience, and it can have a minimal amount of intervention and be considered preferable by another audience. Gave me a push to think about translation quality measurements in a new light.
Superframes: A Schema for Lexicon-free Frame-semantic Annotation (Evang, DMR 2026). What is the difference between “owning something” and “buying something”? A key difference is that owning implies a state of possession, whereas buying initializes a possession. Since annotating text with such linguistic structures is cumbersome and usually requires extensive and complicated resources, “Superframes” proposes an innovative, fairly lightweight and multilingual semantic annotation scheme, building on the beloved syntactic Universal Dependency Schema. This could be very useful for structured linguistic analyses and searches of corpora, including cross-lingual ones.

Conclusion

There are many gems to be found in workshop proceedings. The three papers above is just a small, non-representative sample - I’m sure that I’ve overlooked many similarly interesting papers.

So the take-away from this post is: Don’t view Workshops as a venues of lesser value. View them as a rich and diverse mine with many crafty ideas and interesting studies.

💎💎💎