Repository navigation
JSP3 (Semester Project 3, Linguistics), 2025/26 edition, as run from October 2025 to June 2026.
This release preserves the repository as it stood at the end of the year. It is the first tagged edition.
Task: Word Sense Disambiguation
The class took a text (Čapek, Válka s mloky / The War with the Newts) and sense-annotated its open-class words with a wordnet, automatically, mostly by prompting LLMs. The class worked in pairs, coordinated through GitHub and met fortnightly.
| Task | Goal | Directory |
|---|---|---|
| EVAL | Score tagged text against a gold standard, with partial credit for near-miss senses | tasks/eval/ |
| WSD-C | Compare different amounts of context in the prompt | tasks/wsd-c/ |
| WSD-M | Compare different LLMs, including Czech-capable models | tasks/wsd-m/ |
| WSD-W | Compare different amounts of wordnet information in the prompt | tasks/wsd-w/ |
| ALIGN | Align paragraphs and sentences across the Czech and English versions | tasks/align/ |
| SENTI | Sentiment and named entities over the course of a story | tasks/senti/ |
README.md also proposes TEXT (compare versions of the text), EXPAND and NEW (grow the Czech wordnet). No work on them is committed.
Contents
prepare/: scripts that build the NTU-MC Czech/English wordnet databases and the sample data (db2lmf.py,ntumc2json.py,prepare.bash)tasks/*.md: the task specifications (minimum / desired / stretch)tasks/<task>/: each pair's code, results and reportreport.tex: the report template for the tasksattendance.md: the attendance record
Timeline: first commit 2025-10-05, last commit 2026-06-26. The busiest months were February to April 2026.