Dataset search log

Use this log while you work through Before You Start. Fill in one copy per project. Keep it with your project files, and share it with your team. It shows that you checked for existing work, and it gives you the related-work section of your paper.

Download

Download this template as a Word document (.docx), fill it in on your computer, or print it and fill it in by hand.

Project

Field	Your answer
Project name
Task	e.g. sentiment analysis
Language	e.g. Swahili
Other names for the language	e.g. Kiswahili
All language codes	e.g. sw, swa, swh
Variety, script, and domain you need	e.g. Kenyan Swahili, Latin script, social media
Searched by
Date started

Step 1: Searches

Record every search, including those that found nothing. A search with no results is still useful evidence.

Build your search terms from three parts, and try several combinations:

• Language: every name and code, for example Hausa, Harshen Hausa, hau, ha.
• Task: the task name and its synonyms, for example sentiment, opinion, polarity, emotion.
• Resource word: dataset, corpus, benchmark, annotated, shared task.

The table shows example searches for a Hausa sentiment project. Replace them with your own.

Date	Where	Exact search terms	Relevant results
	ACL Anthology	e.g. Hausa sentiment, Hausa corpus, Hausa dataset
	Lanfrica	e.g. Hausa, then filter by task
	AtlasNLP	e.g. language: Hausa; task: sentiment analysis
	Hugging Face (language filter)	e.g. language hau, task text-classification
	AfricaNLP workshop proceedings	e.g. skim the last two years' paper titles for "Hausa" and "sentiment"
	Google Scholar / arXiv	e.g. "Hausa" sentiment dataset 2020..2026
	Asked the community (where?)	e.g. Masakhane Slack: "Does anyone know of Hausa sentiment data?"

Step 2: Datasets found

One row per dataset. For each check, write Pass, Partly, or Fail, and add a short note where needed.

Dataset	Link	Languages	Size	Licence	Access	Fit	Labels	Splits	Guidelines	Quality	Documentation

Step 3: Decision

Question	Your answer
Which row of the Step 3 table applies?	Reuse / Extend / Build a new one, learning from the old one / Borrow, then build small / Build from scratch
Which dataset(s) is the decision based on?
Why? (two or three sentences)
Decided by, and date

Step 4: People contacted

Date	Who (name, dataset)	How (email, Slack, meeting)	What you asked	What you learned