Datasets:

Eftymak
/

OPbenchmark

+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "arrayShape": "cr:arrayShape",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "containedIn": "cr:containedIn",
+    "cr": "http://mlcommons.org/croissant/",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "prov": "http://www.w3.org/ns/prov#",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "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",
+    "isArray": "cr:isArray",
+    "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"
+  },
+  "@type": "sc:Dataset",
+  "conformsTo": "http://mlcommons.org/croissant/1.1",
+  "name": "Olfactory-Perception-Benchmark",
+  "alternateName": [
+    "OP Benchmark",
+    "Eftymak/OPbenchmark"
+  ],
+  "description": "The Olfactory Perception (OP) benchmark contains 1,010 questions across eight task categories evaluating large language models on olfactory reasoning: odor classification (175), odor primary descriptor identification (175), odor intensity (175), odor pleasantness (175), rate-all-that-apply multilabel descriptor prediction (100), odor mixture similarity (100), olfactory receptor activation (80), and smell identification from real-world odor sources (30). Each question is presented in two prompt formats — isomeric SMILES and compound names — to measure how molecular representation affects performance. Questions, distractor sets, and stratification schemes were manually constructed by the research team. Molecules and ground-truth labels are derived from established peer-reviewed olfactory science datasets, but the question structure is novel. The release additionally includes the Rate-All-That-Apply (RATA) task translated into 20 non-English languages spanning six language families using GPT-5.2; native speakers of nine languages (Arabic, French, Greek, Hindi, Italian, Mandarin Chinese, Persian, Russian, and Spanish) reviewed the translated descriptor lexicons and a sample of translated prompts.",
+  "url": "https://huggingface.co/datasets/Eftymak/OPbenchmark",
+  "creator": [
+    {
+      "@type": "sc:Person",
+      "name": "Eftychia Makri",
+      "affiliation": {
+        "@type": "sc:Organization",
+        "name": "Yale University"
+      }
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Nikolaos Nakis",
+      "affiliation": {
+        "@type": "sc:Organization",
+        "name": "Yale University"
+      }
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Laura Sisson",
+      "affiliation": {
+        "@type": "sc:Organization",
+        "name": "Patina"
+      }
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Geetanjali Minsky",
+      "affiliation": {
+        "@type": "sc:Organization",
+        "name": "MIT Media Lab"
+      }
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Leandros Tassiulas",
+      "affiliation": {
+        "@type": "sc:Organization",
+        "name": "Yale University"
+      }
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Vahid Satarifard",
+      "affiliation": {
+        "@type": "sc:Organization",
+        "name": "Yale University"
+      }
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Nicholas A. Christakis",
+      "affiliation": {
+        "@type": "sc:Organization",
+        "name": "Yale University"
+      }
+    }
+  ],
+  "license": "https://spdx.org/licenses/CC-BY-4.0.html",
+  "datePublished": "2026-11-06",
+  "version": "1.1.0",
+  "keywords": [
+    "olfaction",
+    "smell",
+    "benchmark",
+    "large language models",
+    "LLM evaluation",
+    "chemistry",
+    "molecular reasoning",
+    "SMILES",
+    "multilingual"
+  ],
+  "rai:dataLimitations": "OP is an evaluation-only benchmark not intended for model training or fine-tuning. The benchmark uses standardized descriptor vocabularies and discrete response formats (multiple choice, multilabel selection from fixed lists) which enable reproducible automatic evaluation but do not capture the full richness of odor perception, including contextual effects, individual variation, free-form descriptions, or temporal/concentration dynamics. Ground-truth labels reflect agreement with established olfactory science datasets and should not be interpreted as evidence of mechanistic olfactory understanding. The RATA descriptor lexicon (138 GS-LF descriptors) and the M2OR receptor set are themselves products of specific scientific traditions and may not generalize to all olfactory perception phenomena. Multilingual RATA variants rely on automated GPT-5.2 translation; native speakers reviewed the translations in nine languages (Arabic, French, Greek, Hindi, Italian, Mandarin Chinese, Persian, Russian, and Spanish); subtle connotation shifts may remain, particularly in the other 11 non-English languages. The benchmark is strictly not recommended for: clinical or safety-critical olfactory deployment (e.g., gas detection, food safety certification), training fragrance-design models without independent evaluation, evidence of mechanistic olfactory understanding, or commercial applications without further validation.",
+  "rai:dataBiases": "OP inherits the biases of its source datasets. (i) Compound coverage is drawn from industry-curated databases (GS-LF, IFRA FIG: fragrance/flavor industry odorants) and peer-reviewed academic resources (M2OR, DREAM, Leibniz-LSB@TUM); some odorants outside commercial fragrance/flavor may be underrepresented. (ii) Descriptor lexicon: the 138-descriptor RATA vocabulary follows GS-LF conventions developed primarily in English-speaking and European perfumery contexts (iii) Receptor coverage: the M2OR-derived ORA task covers human olfactory receptors that have been experimentally characterized; rare or recently discovered receptors may be underepresented. (iv) Multilingual coverage: family-level coverage is unbalanced (more Indo-European than non-Indo-European languages); GPT-5.2 translation quality reflects the model's training distribution and is expected to be stronger for higher-resourced languages.",
+  "rai:personalSensitiveInformation": "The dataset contains no directly Personally Identifiable Information (PII) — no names, contact details, faces, biometric data, health records, or national identifiers. The dataset does not contain personal disclosures from human subjects. Individual-level data was not collected by the authors of this work. The multilingual translation process was performed by an automated language model (GPT-5.2). No crowdsourcing was involved in benchmark construction.",
+  "rai:dataUseCases": "OP is intended to measure factual olfactory knowledge in large language models — specifically, whether LLMs can reason from molecular information (compound names or isomeric SMILES) to perceptual properties of smell. Validated use cases include: (1) any-overlap accuracy on single-answer tasks (OC, OPD, OIn, OPl, OS, SIT) for closed-set olfactory question answering; (2) per-question multilabel F1 on RATA and ORA for descriptor and receptor knowledge evaluation; (3) Pearson correlation against human psychophysical ratings on intensity, pleasantness, and mixture similarity; (4) molecular-representation comparison via the dual-prompting (isomeric SMILES vs. compound name) design; (5) cross-lingual evaluation of olfactory descriptor knowledge via the 21-language RATA variant and ensemble approaches. Validity is established through benchmarking of 26 model configurations from 10 providers, paired statistical tests with Holm-Bonferroni correction, hierarchical bootstrap confidence intervals, and comparison to specialist models (POM ensemble, fair-retrained). Use cases NOT validated and explicitly out of scope include: training or fine-tuning olfactory prediction models (test-set integrity is required for benchmark validity); clinical or industrial olfactory deployment; fragrance or flavor product development without further validation; safety-critical gas or contaminant detection; evidence of mechanistic olfactory understanding; demographic representativeness of olfactory perception (the benchmark covers Western-centric science-derived descriptors); free-form or open-ended olfactory description evaluation (the benchmark uses constrained multiple-choice and multilabel formats); and individual-level psychophysical prediction (ground-truth ratings are aggregate means).",
+  "rai:dataSocialImpact": "OP enables systematic, reproducible evaluation of LLM capability on olfaction — a sensory modality previously absent from sensory-alignment benchmarks despite its scientific and practical importance. Positive impacts include identifying capability gaps in current LLMs (e.g., the consistent SMILES-vs-name gap, the mixture-similarity failure mode, receptor knowledge idiosyncrasies) which informs research priorities, providing a baseline against which to measure progress in molecular-perceptual reasoning, surfacing cross-cultural variation in olfactory linguistic resources via the 21-language RATA evaluation, and enabling fair comparison between general-purpose LLMs and specialist olfactory models. Risks and mitigations: (i) Premature claims of olfactory capability: high benchmark scores could be misinterpreted as evidence of LLM olfactory understanding suitable for fragrance, food, or safety applications; we mitigate via explicit out-of-scope documentation in this Croissant card and in the paper's limitations section. (ii) Translation residuals: GPT-5.2-generated translations may carry subtle connotation shifts in languages without native-speaker validation; downstream users should interpret per-language results with appropriate caution. The dataset is released under CC BY 4.0 to support open research and reuse with attribution.",
+  "rai:hasSyntheticData": true,
+  "prov:wasDerivedFrom": [
+    {
+      "@id": "https://github.com/ARY2260/openpom",
+      "prov:label": "GS-LF (GoodScents and Leffingwell) odor descriptor data, as used by the Principal Odor Map (Lee et al. 2023)",
+      "sc:license": "https://spdx.org/licenses/CC-BY-NC-4.0.html",
+      "sc:description": "Source of RATA molecules and labels."
+    },
+    {
+      "@id": "https://m2or.chemsensim.fr/",
+      "prov:label": "M2OR (Molecule-to-Olfactory-Receptor)",
+      "sc:license": "https://spdx.org/licenses/CC-BY-4.0.html"
+    },
+    {
+      "@id": "https://www.synapse.org/Synapse:syn2811262",
+      "prov:label": "DREAM Olfaction Prediction Challenge",
+      "sc:license": "Synapse data use terms; cite Keller et al. 2017 (DOI:10.1126/science.aal2014)"
+    },
+    {
+      "@id": "https://github.com/pyrfume/pyrfume-data/tree/main/bushdid_2014",
+      "prov:label": "Bushdid et al. 2014 mixture similarity dataset (via pyrfume)",
+      "sc:license": "https://spdx.org/licenses/MIT.html"
+    },
+    {
+      "@id": "https://github.com/pyrfume/pyrfume-data/tree/main/snitz_2013",
+      "prov:label": "Snitz et al. 2013 mixture similarity dataset (via pyrfume)",
+      "sc:license": "https://spdx.org/licenses/CC-BY-4.0.html"
+    },
+    {
+      "@id": "https://doi.org/10.1073/pnas.2116576119",
+      "prov:label": "Mayhew et al. 2022, odorous/odorless molecules (source of OC)",
+      "sc:license": "Terms of the original publication and supplementary data; cite Mayhew et al. 2022"
+    },
+    {
+      "@id": "https://ifrafragrance.org/docs/default-source/glossary/ifra-fragrance-ingredient-glossary---april-2020.pdf",
+      "prov:label": "IFRA Fragrance Ingredient Glossary, 2020 (source of OPD)",
+      "sc:license": "IFRA terms of use; cite the IFRA Fragrance Ingredient Glossary"
+    },
+    {
+      "@id": "https://doi.org/10.1038/s41586-020-2891-7",
+      "prov:label": "Ravia et al. 2020 mixture similarity data (source of OS)",
+      "sc:license": "Terms of the original publication; cite Ravia et al. 2020"
+    },
+    {
+      "@id": "https://portal.fis.leibniz-lsb.tum.de/en/publications/valuable-tool-for-flavor-scientists-the-leibniz-lsbtum-odorant-da/",
+      "prov:label": "Leibniz-LSB@TUM Odorant Database (source of SIT)",
+      "sc:license": "Terms of use of the database website; cite Kreissl et al. 2022"
+    }
+  ],
+  "prov:wasGeneratedBy": [
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "Question Construction and Curation",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:Agent",
+          "@id": "op_benchmark_research_team",
+          "prov:label": "Authors: E. Makri, N. Nakis, L. Sisson, G. Minsky, L. Tassiulas, V. Satarifard, N. A. Christakis",
+          "sc:description": "Manual curation, question construction, distractor selection, and quality review performed by the paper's research team."
+        }
+      ]
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "Dual-Prompt Generation",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:Agent",
+          "@id": "op_benchmark_research_team",
+          "prov:label": "Authors: E. Makri, N. Nakis, L. Sisson, G. Minsky, L. Tassiulas, V. Satarifard, N. A. Christakis"
+        }
+      ]
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "Multilingual RATA Translation (synthetic)",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:SoftwareAgent",
+          "@id": "gpt_5.2_(high_reasoning)",
+          "prov:label": "GPT-5.2 (high reasoning)",
+          "sc:description": "GPT-5.2 with high-reasoning settings used for descriptor and UI translation across 20 non-English languages."
+        },
+        {
+          "@type": "prov:Agent",
+          "@id": "native_speaker_annotators",
+          "prov:label": "Native-speaker annotators (Arabic, French, Greek, Hindi, Italian, Mandarin Chinese, Persian, Russian, and Spanish)",
+          "sc:description": "Independent review of the translated descriptor lexicons (whether each term is a real word with an olfactory meaning) and of 20 translated RATA prompts per language, in nine of the 21 languages."
+        },
+        {
+          "@type": "prov:Agent",
+          "@id": "op_benchmark_research_team",
+          "prov:label": "Authors: E. Makri, N. Nakis, L. Sisson, G. Minsky, L. Tassiulas, V. Satarifard, N. A. Christakis",
+          "sc:description": "Designed the translation prompt template, two-step QC procedure, and oversaw the validation process."
+        }
+      ]
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "Model Evaluation",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:Agent",
+          "@id": "op_benchmark_research_team",
+          "prov:label": "Authors: E. Makri, N. Nakis, L. Sisson, G. Minsky, L. Tassiulas, V. Satarifard, N. A. Christakis"
+        },
+        {
+          "@type": "prov:SoftwareAgent",
+          "@id": "model_providers",
+          "prov:label": "Model APIs and local inference (OpenAI, Anthropic, Google, xAI, DeepSeek, Meta, Z.ai, Moonshot AI, Alibaba, Ai2), via provider APIs, Groq, Fireworks, and local vLLM",
+          "sc:description": "26 model configurations; no web search, tool use, or retrieval enabled."
+        }
+      ]
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "Answer Extraction Pipeline",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:Agent",
+          "@id": "op_benchmark_research_team",
+          "prov:label": "Authors: E. Makri, N. Nakis, L. Sisson, G. Minsky, L. Tassiulas, V. Satarifard, N. A. Christakis"
+        }
+      ]
+    }
+  ],
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "repo",
+      "name": "repo",
+      "description": "The Hugging Face git repository hosting the OP benchmark.",
+      "contentUrl": "https://huggingface.co/datasets/Eftymak/OPbenchmark",
+      "encodingFormat": "git+https",
+      "sha256": "https://github.com/mlcommons/croissant/issues/80"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "op-benchmark-csv",
+      "name": "op-benchmark-csv",
+      "description": "The master CSV file containing all 1,010 OP benchmark questions across the eight task categories. Each row is one question, with both isomeric SMILES and compound-name prompt formats.",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "contentUrl": "Benchmark/OP_Benchmark.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "https://github.com/mlcommons/croissant/issues/80"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "rata-100-english-csv",
+      "name": "rata-100-english-csv",
+      "description": "The English RATA-100 CSV (subset of OP_Benchmark.csv corresponding to the rate-all-that-apply task), used as the source for the multilingual translations.",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "contentUrl": "Benchmark/4-Intermediate_1/rata_100.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "https://github.com/mlcommons/croissant/issues/80"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "rata-translations",
+      "name": "rata-translations",
+      "description": "GPT-5.2-generated translations of the 100 RATA questions into 20 non-English languages (Arabic, Bengali, Egyptian Arabic, French, German, Greek, Hindi, Italian, Japanese, Korean, Mandarin Chinese, Persian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Turkish, Ukrainian). Each file shares the same 11-column schema as the master CSV; descriptor vocabulary, OPTIONS, and prompt text are translated, while SMILES, molecule identifiers, and ground-truth answers are retained from the English source. Synthetic content per rai:hasSyntheticData=true; see Multilingual RATA Translation activity in prov:wasGeneratedBy.",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "encodingFormat": "text/csv",
+      "includes": "Benchmark/4-Intermediate_1/Translations_GPT5.2/rata_100_*.csv"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "questions",
+      "name": "questions",
+      "description": "The 1,010 OP benchmark questions across eight task categories. Each row is one question.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "questions/question_ID",
+          "name": "question_ID",
+          "description": "Unique question identifier (UUID).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "op-benchmark-csv"
+            },
+            "extract": {
+              "column": "question_ID"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "questions/question_category",
+          "name": "question_category",
+          "description": "Task category. One of: odor_classification (OC), primary_odor_descriptor (OPD), odor_intensity (OIn), odor_pleasantness (OPl), rata (RATA), mixture_similarity (OS), or_activation (ORA), or smell_identification (SIT).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "op-benchmark-csv"
+            },
+            "extract": {
+              "column": "question_category"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "questions/compound_name_1",
+          "name": "compound.name_1",
+          "description": "Compound name(s) for the first molecule (or list of molecules for mixture/multi-molecule tasks).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "op-benchmark-csv"
+            },
+            "extract": {
+              "column": "compound.name_1"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "questions/SMILES_1",
+          "name": "SMILES_1",
+          "description": "Isomeric SMILES for the first molecule (or list of molecules for mixture/multi-molecule tasks).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "op-benchmark-csv"
+            },
+            "extract": {
+              "column": "SMILES_1"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "questions/compound_name_2",
+          "name": "compound.name_2",
+          "description": "Compound name for the second molecule. Used in paired-comparison tasks (OIn, OPl) and mixture-pair tasks (OS); empty for other tasks.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "op-benchmark-csv"
+            },
+            "extract": {
+              "column": "compound.name_2"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "questions/SMILES_2",
+          "name": "SMILES_2",
+          "description": "Isomeric SMILES for the second molecule. Used in paired-comparison tasks (OIn, OPl) and mixture-pair tasks (OS); empty for other tasks.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "op-benchmark-csv"
+            },
+            "extract": {
+              "column": "SMILES_2"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "questions/OPTIONS",
+          "name": "OPTIONS",
+          "description": "Answer options (semicolon-separated) for multiple-choice and multilabel tasks.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "op-benchmark-csv"
+            },
+            "extract": {
+              "column": "OPTIONS"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "questions/prompt_1",
+          "name": "prompt.1",
+          "description": "The full prompt text using isomeric SMILES notation.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "op-benchmark-csv"
+            },
+            "extract": {
+              "column": "prompt.1"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "questions/prompt_2",
+          "name": "prompt.2",
+          "description": "The full prompt text using compound names.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "op-benchmark-csv"
+            },
+            "extract": {
+              "column": "prompt.2"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "questions/answer",
+          "name": "answer",
+          "description": "Ground-truth answer. Single token for single-answer tasks; semicolon-separated set for multilabel tasks (RATA, ORA); category label for ordinal tasks (OS).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "op-benchmark-csv"
+            },
+            "extract": {
+              "column": "answer"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "questions/other_info",
+          "name": "other_info",
+          "description": "Optional supplementary information (e.g., source dataset, ground-truth numerical ratings for OIn/OPl, perceptual distance for OS). Often empty.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "op-benchmark-csv"
+            },
+            "extract": {
+              "column": "other_info"
+            }
+          }
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "rata_translations",
+      "name": "rata_translations",
+      "description": "Translated RATA-100 questions across 20 non-English languages, sourced from the rata-translations FileSet. Each row is one question in one language; the source file's language is encoded in the filename (e.g., rata_100_French_gpt5.csv). Schema is identical to the master questions RecordSet, but descriptor vocabulary, OPTIONS, and prompt text are GPT-5.2-translated.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "rata_translations/question_ID",
+          "name": "question_ID",
+          "description": "Unique question identifier; matches the corresponding RATA row in the master questions RecordSet.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "rata-translations"
+            },
+            "extract": {
+              "column": "question_ID"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "rata_translations/compound_name_1",
+          "name": "compound.name_1",
+          "description": "Compound name for the molecule (retained from English source; not translated).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "rata-translations"
+            },
+            "extract": {
+              "column": "compound.name_1"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "rata_translations/SMILES_1",
+          "name": "SMILES_1",
+          "description": "Isomeric SMILES for the molecule (retained from English source; not translated).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "rata-translations"
+            },
+            "extract": {
+              "column": "SMILES_1"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "rata_translations/compound_name_2",
+          "name": "compound.name_2",
+          "description": "Empty for RATA (single-molecule task).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "rata-translations"
+            },
+            "extract": {
+              "column": "compound.name_2"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "rata_translations/SMILES_2",
+          "name": "SMILES_2",
+          "description": "Empty for RATA (single-molecule task).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "rata-translations"
+            },
+            "extract": {
+              "column": "SMILES_2"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "rata_translations/OPTIONS",
+          "name": "OPTIONS",
+          "description": "Translated 138-descriptor option set (semicolon-separated) for the target language. GPT-5.2 generated; native-speaker reviewed for Arabic, French, Greek, Hindi, Italian, Mandarin Chinese, Persian, Russian, and Spanish.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "rata-translations"
+            },
+            "extract": {
+              "column": "OPTIONS"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "rata_translations/question_category",
+          "name": "question_category",
+          "description": "Task category, always 'rata' in this RecordSet.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "rata-translations"
+            },
+            "extract": {
+              "column": "question_category"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "rata_translations/prompt_1",
+          "name": "prompt.1",
+          "description": "Translated prompt text using isomeric SMILES notation.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "rata-translations"
+            },
+            "extract": {
+              "column": "prompt.1"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "rata_translations/prompt_2",
+          "name": "prompt.2",
+          "description": "Translated prompt text using compound names.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "rata-translations"
+            },
+            "extract": {
+              "column": "prompt.2"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "rata_translations/answer",
+          "name": "answer",
+          "description": "Ground-truth answer (translated descriptor set, semicolon-separated).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "rata-translations"
+            },
+            "extract": {
+              "column": "answer"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "rata_translations/other_info",
+          "name": "other_info",
+          "description": "Optional supplementary information; usually empty.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "rata-translations"
+            },
+            "extract": {
+              "column": "other_info"
+            }
+          }
+        }
+      ]
+    }
+  ],
+  "citeAs": "@inproceedings{makri2026op, title={Benchmark for Assessing Olfactory Perception of Large Language Models}, author={Makri, Eftychia and Nakis, Nikolaos and Sisson, Laura and Minsky, Geetanjali and Tassiulas, Leandros and Satarifard, Vahid and Christakis, Nicholas A.}, booktitle={Advances in Neural Information Processing Systems (NeurIPS), Evaluations and Datasets Track}, year={2026}}",
+  "sameAs": "https://github.com/Satarifard/Olfactory-Perception-Benchmark"
+}