metadata
license: apache-2.0
task_categories:
- text-classification
- question-answering
- summarization
- translation
language:
- tl
- en
tags:
- tagalog
- distill
- gemini
- self instruct
- synthetic
size_categories:
- n<1K
SimpleQnA Dataset Description
SimpleQnA is a straightforward question-and-answer dataset in Tagalog. It covers many topics and task types, mixing simple and complex instructions. Some instructions are in English to avoid the model learning Tagalog-only responses when given Tagalog prompts.
The dataset includes constraints borrowed from Tulu 3, Muffin, and IFEval for instruction-following diversity. Random sampling is used to enhance variety.
Topics and Task Types Included
Basic and Common Tasks
- Answer general knowledge questions.
- Solve basic math word problems and arithmetic.
- Explain simple Python code or write pseudocode.
- Roleplay brief scenarios as specific characters.
- Generate short creative pieces like poems or story ideas.
- Tell or generate jokes on given topics.
- Create multiple-choice questions with answers and explanations.
- Make fill-in-the-blanks sentences with missing words.
- Generate true/false statements with validity.
- Extract specific info from short texts.
- Summarize short paragraphs concisely.
- Respond to simple conversational prompts.
- Answer trivia questions.
- Play simple word games (e.g., synonyms).
- Outline steps for common planning tasks.
- Explain STEM concepts simply.
- Answer open-ended factual questions.
- Classify short texts by category.
- Rewrite sentences in different tones.
- Brainstorm ideas on simple topics.
- Generate short content like product taglines.
Expanded and More Complex Tasks
- Write journal-style entries on topics, emotions, or experiences.
- Continue or finish open-ended story prompts with style traits.
- Rewrite texts in different styles or tones (formal, casual, specific authors).
- Create creative works under unusual constraints (e.g., fixed vowels, word counts).
- Perform tasks like drafting emails or writing commented pseudocode.
- Plan multi-step tasks, then estimate costs or resources.
- Follow instructions avoiding certain elements or styles.
- Summarize chat transcripts with main points.
- Solve math, logic, or coding problems with step-by-step explanations.
- Identify and correct errors in text, code, or logic with explanations.
- Perform unit conversions.
- Create multiple-choice questions with detailed rationales.
- Solve problems using chain-of-thought reasoning.
- Judge true/false statements with justification.
- Analyze mixed factual/fictional texts, flagging each.
- Simplify dense or technical texts for laypeople or kids.
- Rewrite paragraphs in multiple specified ways (simplify, clarify, shorten).
- Define complex terms with simple examples.
- Create or solve text puzzles and provide solutions.
- Reflect on hypothetical personal experiences in journal form.
- Generate creative prompts for specific genres or themes.
Complex and Advanced Topics
- Advanced general knowledge questions requiring synthesis.
- Multi-step algebra, geometry proofs, probability problems.
- Debugging and explaining complex code or algorithms.
- Nuanced roleplaying with conflicting emotions or goals.
- Wordplay or philosophical jokes.
- Fill-in-the-blanks with technical or academic vocabulary.
- True/false statements on specialized topics with detailed justification.
- Extracting multiple linked facts from dense texts.
- Summarizing research papers or lengthy opinion pieces.
- In-depth conversations on ethical, philosophical, or societal dilemmas.
- Difficult trivia requiring obscure or multi-step logic.
- Detailed project planning with budgets and timelines.
- Logical reasoning to find fallacies in arguments.
- Ethical guideline proposals for AI applications.
- Editing academic or legal texts for clarity and impact.
- Debunking complex misconceptions step-by-step.
- Analyzing social scenarios with hidden motives.
- Explaining advanced STEM concepts to non-experts.
- Open-ended critical thinking and multi-domain Q&A.
- Fine-grained classification with ambiguous inputs.
- Closed Q&A from lengthy technical documents.
- Rewriting biased reports into neutral summaries.
- Brainstorming innovative global solutions.
- Creating complex structured documents (character bios, grant proposals, FAQs).
AI Awareness & Safety Handling
- Explaining AI capabilities and limitations.
- Reflecting on AI knowledge cut-off and its effects.
- Handling sensitive, inappropriate, or potentially harmful requests responsibly.
SimpleQnA is diverse and rich. It tests simple question answering but also pushes the model to handle complex tasks, follow constraints, and respond naturally in Tagalog and English.