WordSearch is an environment for evaluating agents on finding hidden words in letter grids. This environment wraps the WordSearch implementation from TextArena, a framework for text-based game environments.
- Pattern recognition in 2D grids
- Spatial reasoning across horizontal, vertical, and diagonal directions
- Systematic search strategies
- Testing visual-spatial processing in text format
WordSearch does not require a sandbox. It has minimal compute requirements.
MIT.
There are two splits: train (600 tasks) and test (300 tasks). Each split contains 100 tasks (train) or 50 tasks (test) across each of 6 variants:
- WordSearch-v0
- WordSearch-v0-hardcore
- WordSearch-v0-hardcore-raw
- WordSearch-v0-hardcore-train
- WordSearch-v0-raw
- WordSearch-v0-train
Each task is seeded for reproducibility.
This is a sparse reward environment. Rewards are mapped from TextArena's native range of {-1, 0, 1} to {0.0, 0.5, 1.0} via (raw + 1) / 2.
We do not use LLM graders for this environment; reward is determined programmatically.
Game state is generated procedurally by the TextArena engine using seeded randomness. No external data files are required.
Agents are given a single tool:
find_word(start_row, start_col, end_row, end_col): Mark a word's position by specifying start and end coordinates
WordSearch is a multi-turn environment.
Easy to Medium. Word Search requires systematic grid scanning and pattern matching. Difficulty increases with grid size and word placement complexity in hardcore variants.
There are no further environment requirements; WordSearch works out of the box without any secrets or API keys.
Agents in WordSearch interact only with a letter grid puzzle and have no access to external systems, the internet, or sensitive data. The environment does not present safety risks.
@software{textarena2024,
author = {Guertler, Leon and Banting, Wilfried and Pignatelli, Eduardo},
title = {TextArena},
year = {2024},
publisher = {GitHub},
url = {https://github.com/LeonGuertler/TextArena}
}