Explain task categories and refine dataset documentation

README.md

@@ -58,6 +58,24 @@ MMDM provides a single `test` split containing **4,961 examples**: **4,351 hard
 
 ![MMDM task distribution: 4,961 examples across 17 categories](assets/task-distribution.png)
 
+- **Deductive reasoning:** Derive conclusions from stated premises, rules, or natural-language entailment.
+- **Policy exceptions:** Apply policies while resolving exceptions, rule precedence, and conflicting conditions.
+- **Knowledge QA:** Answer factual and subject-specific questions across multiple domains.
+- **Probabilistic reasoning:** Infer outcome probabilities from uncertainty, conditional information, sampling, or noisy evidence.
+- **Commonsense context:** Use everyday knowledge to answer a question or select a plausible continuation.
+- **Constraint planning:** Choose an action or plan that satisfies interacting constraints and trade-offs.
+- **Quantitative reasoning:** Solve numerical questions involving arithmetic, word problems, financial information, or tables.
+- **Option logic:** Jointly interpret logical statements and dependencies distributed across the candidate options.
+- **Intent routing:** Identify the user intent and select an appropriate tool, route, or action.
+- **Multi-hop reasoning:** Connect facts across passages, documents, or entities to reach an answer.
+- **Evidence validation:** Check whether evidence supports a claim or whether a response meets stated semantic and format requirements.
+- **Answer preference:** Choose the response that best satisfies the request among candidate answers.
+- **Score rubrics:** Assign a score by applying explicit criteria for completeness, coverage, compliance, or task completion.
+- **Temporal / numeric:** Apply rules involving dates, durations, numeric thresholds, or time-dependent conditions.
+- **Code / database:** Reason about code behavior or select the database query that answers a request.
+- **Reading comprehension:** Answer questions using information stated or implied in a supplied passage.
+- **Emotion recognition:** Identify the emotion conveyed by a piece of text.
+
 ## Dataset Creation
 
 MMDM draws on [ARC](https://huggingface.co/datasets/allenai/ai2_arc), [BANKING77](https://huggingface.co/datasets/PolyAI/banking77), [BoolQ](https://github.com/google-research-datasets/boolean-questions), [CodeContests](https://github.com/google-deepmind/code_contests), [CommonsenseQA](https://www.tau-nlp.sites.tau.ac.il/commonsenseqa), [ContractNLI](https://stanfordnlp.github.io/contract-nli/), [FinQA](https://github.com/czyssrs/FinQA), [GoEmotions](https://github.com/google-research/google-research/tree/master/goemotions), [HellaSwag](https://rowanzellers.com/hellaswag/), [HelpSteer2](https://huggingface.co/datasets/nvidia/HelpSteer2), [Instruction-following rubric adaptations](https://huggingface.co/datasets/allenai/IF_multi_constraints_upto5), [LogiQA 2.0](https://github.com/csitfun/LogiQA2.0), [MASSIVE](https://huggingface.co/datasets/AmazonScience/massive), [MathQA](https://math-qa.github.io/), [MMLU auxiliary data](https://huggingface.co/datasets/cais/mmlu), [MultiDoc2Dial](https://doc2dial.github.io/multidoc2dial/), [MuSiQue](https://github.com/StonyBrookNLP/musique), [OpenBookQA](https://huggingface.co/datasets/allenai/openbookqa), [OR-ShARC](https://github.com/Yifan-Gao/open_retrieval_conversational_machine_reading), [RuleTaker](https://github.com/allenai/ruletaker), [SNLI](https://nlp.stanford.edu/projects/snli/), [Spider](https://yale-lily.github.io/spider), [TAT-QA](https://github.com/NExTplusplus/TAT-QA), and [TempReason](https://github.com/DAMO-NLP-SG/TempReason). These labeled datasets are converted into a common state/question/options format while retaining their source answers. They are complemented by independently program-composed decision problems and LLM-assisted synthetic scenarios with independently reviewed labels. Program-defined tasks use executable rules or probability calculations; soft targets specify the target decision distribution and are not universally annotator-vote frequencies.