| { |
| "@context": { |
| "@language": "en", |
| "@vocab": "https://schema.org/", |
| "citeAs": "cr:citeAs", |
| "column": "cr:column", |
| "conformsTo": "dct:conformsTo", |
| "cr": "http://mlcommons.org/croissant/", |
| "rai": "http://mlcommons.org/croissant/RAI/", |
| "dct": "http://purl.org/dc/terms/", |
| "sc": "https://schema.org/", |
| "data": { |
| "@id": "cr:data", |
| "@type": "@json" |
| }, |
| "dataBiases": "cr:dataBiases", |
| "dataCollection": "cr:dataCollection", |
| "dataCollectionMissingData": "cr:dataCollectionMissingData", |
| "dataCollectionRawData": "cr:dataCollectionRawData", |
| "dataCollectionTimeframe": "cr:dataCollectionTimeframe", |
| "dataCollectionType": "cr:dataCollectionType", |
| "dataImputationProtocol": "cr:dataImputationProtocol", |
| "dataLimitations": "cr:dataLimitations", |
| "dataManipulationProtocol": "cr:dataManipulationProtocol", |
| "dataPreprocessingProtocol": "cr:dataPreprocessingProtocol", |
| "dataSocialImpact": "cr:dataSocialImpact", |
| "dataUseCases": "cr:dataUseCases", |
| "dataAnnotationProtocol": "cr:dataAnnotationProtocol", |
| "dataAnnotationPlatform": "cr:dataAnnotationPlatform", |
| "dataAnnotationAnalysis": "cr:dataAnnotationAnalysis", |
| "annotationsPerItem": "cr:annotationsPerItem", |
| "annotatorDemographics": "cr:annotatorDemographics", |
| "machineAnnotationTools": "cr:machineAnnotationTools", |
| "personalSensitiveInformation": "cr:personalSensitiveInformation", |
| "dataReleaseMaintenance": "cr:dataReleaseMaintenance", |
| "dataField": "cr:dataField", |
| "dataType": { |
| "@id": "cr:dataType", |
| "@type": "@vocab" |
| }, |
| "dct": "http://purl.org/dc/terms/", |
| "extract": "cr:extract", |
| "field": "cr:field", |
| "fileProperty": "cr:fileProperty", |
| "fileObject": "cr:fileObject", |
| "fileSet": "cr:fileSet", |
| "format": "cr:format", |
| "includes": "cr:includes", |
| "isLiveDataset": "cr:isLiveDataset", |
| "jsonPath": "cr:jsonPath", |
| "key": "cr:key", |
| "md5": "cr:md5", |
| "parentField": "cr:parentField", |
| "path": "cr:path", |
| "recordSet": "cr:recordSet", |
| "references": "cr:references", |
| "regex": "cr:regex", |
| "repeated": "cr:repeated", |
| "replace": "cr:replace", |
| "sc": "https://schema.org/", |
| "separator": "cr:separator", |
| "source": "cr:source", |
| "subField": "cr:subField", |
| "transform": "cr:transform", |
| "prov": "http://www.w3.org/ns/prov#" |
| }, |
| "@type": "sc:Dataset", |
| "name": "MMTT-Bench", |
| "description": "MMTT-Bench (Multimodal Text-and-Time-series Benchmark) is a synthetic benchmark for evaluating information-theoretic metrics on text-annotated time series. It provides controlled experimental conditions for measuring how much textual annotations inform predictions about co-occurring time-series signals, enabling rigorous evaluation of metrics including KSG mutual information, partial information decomposition (PID), V-information, and pointwise mutual information (PMI). The synthetic signals is a sine wave with constant-period discontinuities, annoted with correct, incorrect and irrelevant text annotations", |
| "conformsTo": "http://mlcommons.org/croissant/1.0", |
| "citeAs": "@dataset{mmtt_bench_2026, title={When Does Text Inform? Benchmarking Information-Theoretic Metrics for Multimodal Time-Series Forecasting}, year={2026}}", |
| "license": "https://creativecommons.org/licenses/by/4.0/", |
| "url": "https://huggingface.co/datasets/WhenDoesTextInform/MMTT-Bench", |
| "version": "1.0.0", |
| "keywords": [ |
| "multimodal", |
| "time series", |
| "fusion", |
| "forecasting", |
| "information theory", |
| "mutual information", |
| "benchmark", |
| "synthetic", |
| "fusion" |
| ], |
| "inLanguage": "en", |
| "isLiveDataset": false, |
|
|
| "rai:dataCollection": "All data is synthetically generated. Time-series signals are procedurally constructed (sine wave with controlled constant-period discontinuities; Rössler chaotic attractor). Text annotations are generated algorithmically to encode varying degrees of mutual information with the time-series signals, providing ground-truth control over the information-theoretic relationships being benchmarked.", |
| "rai:dataCollectionType": "Synthetic / programmatically generated", |
| "rai:dataCollectionMissingData": "No missing data. The benchmark is fully synthetic and all fields are populated by construction.", |
| "rai:dataCollectionRawData": "There is no upstream raw data source. Signals and annotations are generated in Python using controlled random seeds. Source code for data generation is available in the accompanying paper repository.", |
| "rai:dataCollectionTimeframe": "Generated in 2026 for submission to NeurIPS 2026 Evaluations and Datasets Track.", |
| "rai:dataImputationProtocol": "Not applicable. No real-world data was collected and no imputation was required.", |
| "rai:dataPreprocessingProtocol": "Time-series values are normalised per signal type prior to serialisation. Text annotations are generated at the segment level and serialised as string fields alongside the corresponding numeric time-series arrays. No further preprocessing is applied; downstream metric implementations are expected to operate on the values as stored and implement preprocessing appropriate to their constraints, like text embeddings and dimensionality reduction.", |
| "rai:dataManipulationProtocol": "Dataset splits (train / val / test) are constructed temporally by cycle, with training consisting of the first 62.5% of the signal, validation the next 18.75% and testing the final 18.75%.", |
| "rai:dataAnnotationProtocol": "Annotations are machine-generated. Each time-series segment is paired with a text annotation whose information content with respect to the segment is set by a controlled parameter. There are no human annotators.", |
| "rai:dataAnnotationPlatform": "Custom Python scripts (no external annotation platform).", |
| "rai:dataAnnotationAnalysis": "Mutual information between text annotations and time-series segments is true by design of the dataset, and analytically verified through MI estimators and model performance.", |
| "rai:annotationsPerItem": "1 text annotation per text category (correct, incorrect, irrelevant) per time-series point. Annotations are stored as a list to enable extension to multiple text samples per category per point.", |
| "rai:annotatorDemographics": "Not applicable. All annotations are machine-generated.", |
| "rai:machineAnnotationTools": "Custom Python data-generation scripts. No third-party annotation models (e.g., LLMs) were used to produce the annotations.", |
| "rai:dataBiases": "As a fully synthetic benchmark, MMTT-Bench does not reflect biases present in naturally occurring text or time-series corpora. However, the benchmark is intentionally narrow in scope: signal diversity is limited to a single signal, and text annotation vocabulary is controlled. Results on MMTT-Bench may not generalise to real-world multimodal datasets with organic language variation or distribution shift.", |
| "rai:dataUseCases": "Intended use: controlled evaluation and comparison of information-theoretic metrics (KSG mutual information, PID, V-information, PMI) on multimodal data pairing text with time series. Suitable for benchmarking metric sensitivity, bias, and computational properties under known ground-truth conditions. Not intended for deployment in production systems or for training general-purpose models.", |
| "rai:dataLimitations": "The benchmark covers only one signal types. Text annotations follow a constrained vocabulary and do not reflect the complexity of natural language. Metric evaluation results should be interpreted within the synthetic setting and not extrapolated to real-world multimodal corpora without further validation.", |
| "rai:dataSocialImpact": "MMTT-Bench contains no personal data, sensitive content, or real-world subjects. It is a methodological contribution intended to improve the rigour of information-theoretic metric evaluation in the ML research community. No direct negative societal impact is anticipated.", |
| "rai:personalSensitiveInformation": "None. The dataset is entirely synthetic and contains no personal, demographic, medical, or otherwise sensitive information.", |
| "rai:dataReleaseMaintenance": "The dataset is released as a static benchmark under CC-BY 4.0. Versioned releases will be maintained on Hugging Face. Bug reports and questions can be raised via the Hugging Face community discussion tab.", |
| "rai:hasSyntheticData": true, |
| "prov:wasDerivedFrom": [ |
| { |
| "@id": "https://anonymous.4open.science/r/MMTT-Bench-0C83/src/mmtt_bench/data_generation/sine_templates.py", |
| "prov:label": "Templated text annoations, templates in MMTT-Bench code", |
| "sc:license": "MIT" |
| } |
| ], |
| "prov:wasGeneratedBy": [ |
| { |
| "@type": "prov:Activity", |
| "prov:type": { |
| "@id": "https://www.wikidata.org/wiki/Q655427" |
| }, |
| "prov:label": "Signal generation", |
| "sc:description": "Time-series signals are procedurally generated using deterministic algorithms: a sine wave with controlled constant-period discontinuities. Ground-truth labels (phase, true_direction, true_magnitude) are derived analytically from signal values at each timestep." |
| }, |
| { |
| "@type": "prov:Activity", |
| "prov:type": { |
| "@id": "https://www.wikidata.org/wiki/Q109719325" |
| }, |
| "prov:label": "Template-based annotation", |
| "sc:description": "Natural language annotations are produced by a template-based Python text generator. Slot values (direction_word, magnitude_word, tense) are drawn from a controlled vocabulary and filled into sentence templates to produce one or more candidate texts per point. Distractor annotations (category='incorrect') are generated by the same mechanism with deliberately mismatched slot values. No language models, crowdworkers, or third-party annotation platforms were involved." |
| } |
| ], |
|
|
| "distribution": [ |
| { |
| "@type": "cr:FileObject", |
| "@id": "train-json", |
| "name": "train.json", |
| "description": "Training split. A single JSON object with two top-level keys: 'metadata' (split='train', n_points=3840, n_samples=1, and a 't' array of timestamps) and 'points' (array of per-point records each containing signal values, ground-truth labels, and text annotations).", |
| "contentUrl": "https://huggingface.co/datasets/WhenDoesTextInform/MMTT-Bench/resolve/main/train.json", |
| "encodingFormat": "application/json", |
| "sha256": "831b4ce98d414d1efe137c1a0716246f85b2f434ab491b956901519dc5ce4243" |
| }, |
| { |
| "@type": "cr:FileObject", |
| "@id": "val-json", |
| "name": "val.json", |
| "description": "Validation split. Same format as train.json: a single JSON object with 'metadata' (split='val', n_points=1152) and 'points' array.", |
| "contentUrl": "https://huggingface.co/datasets/WhenDoesTextInform/MMTT-Bench/resolve/main/val.json", |
| "encodingFormat": "application/json", |
| "sha256": "884a78bcd346fdc2edeb1d676c39648a88cea5847997915b90e18edd62cb5462" |
| }, |
| { |
| "@type": "cr:FileObject", |
| "@id": "test-json", |
| "name": "test.json", |
| "description": "Test split. Same format as train.json: a single JSON object with 'metadata' (split='test', n_points=1152) and 'points' array.", |
| "contentUrl": "https://huggingface.co/datasets/WhenDoesTextInform/MMTT-Bench/resolve/main/test.json", |
| "encodingFormat": "application/json", |
| "sha256": "c6fce1fc780bdc94df516c06871c6b61627ec96e80a29a945cf4752deac1674c" |
| } |
| ], |
|
|
| "recordSet": [ |
| { |
| "@type": "cr:RecordSet", |
| "@id": "train-records", |
| "name": "train", |
| "description": "Training split. One record per point in the time series.", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "train-records/t", |
| "name": "t", |
| "description": "Timestamp of the point.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileObject": { "@id": "train-json" }, |
| "extract": { "jsonPath": "$.points[*].t" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "train-records/y", |
| "name": "y", |
| "description": "Signal value at this point.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileObject": { "@id": "train-json" }, |
| "extract": { "jsonPath": "$.points[*].y" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "train-records/dy", |
| "name": "dy", |
| "description": "Derivative of the signal at this point.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileObject": { "@id": "train-json" }, |
| "extract": { "jsonPath": "$.points[*].dy" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "train-records/phase", |
| "name": "phase", |
| "description": "Phase label for the point (e.g. ascending_zero).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "train-json" }, |
| "extract": { "jsonPath": "$.points[*].phase" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "train-records/true_direction", |
| "name": "true_direction", |
| "description": "Ground-truth direction label (e.g. increasing/decreasing).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "train-json" }, |
| "extract": { "jsonPath": "$.points[*].true_direction" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "train-records/true_magnitude", |
| "name": "true_magnitude", |
| "description": "Ground-truth magnitude label (e.g. steep/gradual).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "train-json" }, |
| "extract": { "jsonPath": "$.points[*].true_magnitude" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "train-records/category", |
| "name": "category", |
| "description": "Whether the text annotation is correct, incorrect or irrelevant", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "train-json" }, |
| "extract": { "jsonPath": "$.points[*].category" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "train-records/tense", |
| "name": "tense", |
| "description": "Grammatical tense of the annotation (past/present/future).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "train-json" }, |
| "extract": { "jsonPath": "$.points[*].tense" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "train-records/text", |
| "name": "text", |
| "description": "Natural language annotation for this point (first sample).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "train-json" }, |
| "extract": { "jsonPath": "$.points[*].samples[0].text" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "train-records/irrelevant_group", |
| "name": "irrelevant_group", |
| "description": "Distractor group identifier; null for correct annotations.", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "train-json" }, |
| "extract": { "jsonPath": "$.points[*].samples[0].irrelevant_group" } |
| } |
| } |
| ] |
| }, |
| { |
| "@type": "cr:RecordSet", |
| "@id": "val-records", |
| "name": "val", |
| "description": "Validation split. One record per point in the time series.", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "val-records/t", |
| "name": "t", |
| "description": "Timestamp of the point.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileObject": { "@id": "val-json" }, |
| "extract": { "jsonPath": "$.points[*].t" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "val-records/y", |
| "name": "y", |
| "description": "Signal value at this point.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileObject": { "@id": "val-json" }, |
| "extract": { "jsonPath": "$.points[*].y" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "val-records/dy", |
| "name": "dy", |
| "description": "Derivative of the signal at this point.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileObject": { "@id": "val-json" }, |
| "extract": { "jsonPath": "$.points[*].dy" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "val-records/phase", |
| "name": "phase", |
| "description": "Phase label for the point (e.g. ascending_zero).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "val-json" }, |
| "extract": { "jsonPath": "$.points[*].phase" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "val-records/true_direction", |
| "name": "true_direction", |
| "description": "Ground-truth direction label (e.g. increasing/decreasing).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "val-json" }, |
| "extract": { "jsonPath": "$.points[*].true_direction" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "val-records/true_magnitude", |
| "name": "true_magnitude", |
| "description": "Ground-truth magnitude label (e.g. steep/gradual).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "val-json" }, |
| "extract": { "jsonPath": "$.points[*].true_magnitude" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "val-records/category", |
| "name": "category", |
| "description": "Whether the text annotation is correct, incorrect or irrelevant.", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "val-json" }, |
| "extract": { "jsonPath": "$.points[*].category" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "val-records/tense", |
| "name": "tense", |
| "description": "Grammatical tense of the annotation (past/present/future).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "val-json" }, |
| "extract": { "jsonPath": "$.points[*].tense" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "val-records/text", |
| "name": "text", |
| "description": "Natural language annotation for this point (first sample).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "val-json" }, |
| "extract": { "jsonPath": "$.points[*].samples[0].text" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "val-records/irrelevant_group", |
| "name": "irrelevant_group", |
| "description": "Distractor group identifier; null for correct annotations.", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "val-json" }, |
| "extract": { "jsonPath": "$.points[*].samples[0].irrelevant_group" } |
| } |
| } |
| ] |
| }, |
| { |
| "@type": "cr:RecordSet", |
| "@id": "test-records", |
| "name": "test", |
| "description": "Test split. One record per point in the time series.", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "test-records/t", |
| "name": "t", |
| "description": "Timestamp of the point.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileObject": { "@id": "test-json" }, |
| "extract": { "jsonPath": "$.points[*].t" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "test-records/y", |
| "name": "y", |
| "description": "Signal value at this point.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileObject": { "@id": "test-json" }, |
| "extract": { "jsonPath": "$.points[*].y" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "test-records/dy", |
| "name": "dy", |
| "description": "Derivative of the signal at this point.", |
| "dataType": "sc:Float", |
| "source": { |
| "fileObject": { "@id": "test-json" }, |
| "extract": { "jsonPath": "$.points[*].dy" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "test-records/phase", |
| "name": "phase", |
| "description": "Phase label for the point (e.g. ascending_zero).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "test-json" }, |
| "extract": { "jsonPath": "$.points[*].phase" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "test-records/true_direction", |
| "name": "true_direction", |
| "description": "Ground-truth direction label (e.g. increasing/decreasing).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "test-json" }, |
| "extract": { "jsonPath": "$.points[*].true_direction" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "test-records/true_magnitude", |
| "name": "true_magnitude", |
| "description": "Ground-truth magnitude label (e.g. steep/gradual).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "test-json" }, |
| "extract": { "jsonPath": "$.points[*].true_magnitude" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "test-records/category", |
| "name": "category", |
| "description": "Whether the text annotation is correct, incorrect or irrelevant.", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "test-json" }, |
| "extract": { "jsonPath": "$.points[*].category" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "test-records/tense", |
| "name": "tense", |
| "description": "Grammatical tense of the annotation (past/present/future).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "test-json" }, |
| "extract": { "jsonPath": "$.points[*].tense" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "test-records/text", |
| "name": "text", |
| "description": "Natural language annotation for this point (first sample).", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "test-json" }, |
| "extract": { "jsonPath": "$.points[*].samples[0].text" } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "test-records/irrelevant_group", |
| "name": "irrelevant_group", |
| "description": "Distractor group identifier; null for correct annotations.", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { "@id": "test-json" }, |
| "extract": { "jsonPath": "$.points[*].samples[0].irrelevant_group" } |
| } |
| } |
| ] |
| } |
| ] |
| } |
