WildChat-GIO: Filtered English/German Prompt Pool for the GIO Pilot Annotation Study

This dataset contains a filtered subset of the WildChat-1M dataset (Zhao et al., 2024), prepared for the pilot annotation study of the Generative Intent Operationalization (GIO) framework.

Filtering pipeline (AP1):Starting from 1,048,576 conversations in WildChat-1M, the following filters were applied to first-turn user prompts:

Language: English and German only

Minimum word count: ≥ 5 words

Code removal: prompts dominated by code blocks excluded

Deduplication: near-duplicate prompts removed

Result: 230,289 filtered prompts (226,042 English / 4,247 German)

File format: CSV with columns: conversation_id, prompt, language, word_count, block

Usage: This pool serves as input for AP2 (stratified sampling), where 55 prompts are selected for expert annotation of GIO modes and grounding necessity variables. The full experiment pipeline is available at: https://github.com/kaispriestersbach/gio-pilot-study

Source dataset: Zhao, W. X. et al. (2024). WildChat-1M. Licensed under ODC-BY. https://huggingface.co/datasets/allenai/WildChat-1M