fffoivos/greek-forum-reasoning-traces
Greek Forum Reasoning Traces Greek has almost none of the post-training data English takes for granted. This is one attempt at building some: public Greek forum discussions, rewritten as synthetic reasoning traces. Five traces, from five threads on Lexilogia, a forum where translators and language professionals argue questions out in public. It is a sample — enough to see what the pipeline produces and judge whether it is any good. How a discussion becomes a trace — the… See the full description on the dataset page: https://huggingface.co/datasets/fffoivos/greek-forum-reasoning-traces.
Greek Forum Reasoning Traces
Greek has almost none of the post-training data English takes for granted. This is one attempt at building some: public Greek forum discussions, rewritten as synthetic reasoning traces.
Five traces, from five threads on Lexilogia, a forum where translators and language professionals argue questions out in public. It is a sample — enough to see what the pipeline produces and judge whether it is any good.
[How a discussion becomes a trace](https://claude.ai/code/artifact/9e71418a-f97b-43e4-a053-d98ab15a12ac) — the walkthrough, with real material at every step.
What a trace is
One discussion becomes one exchange. A question in the voice of someone who actually has the problem, then a reply in four parts: an opening that restates what is really being asked, the reasoning, a pre-conclusion naming what the reasoning did and did not settle, and an answer. Where the thread reached no consensus, a quality note says so rather than inventing one.
How close these are to the source
The reasoning, the pre-conclusion and the answer are synthesised. They are not assembled from anyone's posts.
The question is rewritten from the thread's opening post, and stays closer to it. Measured across these five, the similarity between a question and its opening post runs from 0.50 to 0.85, and the longest verbatim run is 184 characters. Treat the questions as close paraphrase, not clean-room rewriting. Every trace carries its thread URL, so all of this can be checked.
The discussions behind them
The threads these were written from are published in full at `Greekpt/lexilogia-discussions` — 17,814 threads, 235,889 posts. Read its card before using it: unlike this sample, it names the members who wrote the posts.
The code
The pipeline that produced these is open, at fffoivos/greek-forum-reasoning-traces-pipeline: the gates that decide which discussions are worth using, the annotation step, and the assembly that turns a thread into one trace. The prompts are included. The collection code is not.
Rights and removal
Lexilogia's terms leave copyright with the individual authors. This sample is published for non-commercial research, as transformed and attributed material, and it is not a redistribution of the forum. No trace here names a participant.
If you posted in one of these threads, or you are Lexilogia staff, and you want a trace removed, open a discussion on this repository or write to foivos.karounos@eellak.gr. It will be removed, no justification needed, and the removal will carry into anything released from this work afterwards.
What is not released
The collection code, the filtering and acceptance gates, and the corpora from the other forums are not part of this repository.
Citation
Karounos, F. (2026). Greek Forum Reasoning Traces. Greekpt. https://huggingface.co/datasets/Greekpt/greek-forum-reasoning-traces