{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.23943", "kind": "paper", "url": "https://arxiv.org/abs/2606.23943"}], "authors": ["Maria Contreras"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.23943", "source_url": "https://arxiv.org/abs/2606.23943"}, "categories": ["Language & Knowledge", "Software & AI Compute", "cs.CL"], "description": {"en": "A comparison of tokenization strategies (BPE, Unigram LM, WordPiece, PRPE) for Southern Quechua using a 200k-sentence corpus and a finite-state morphological analyzer as reference, with metrics including fertility rate, OOV rate, and morphological boundary accuracy."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.23943", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.23943", "title": "QuechuaTok"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Peruvian University of Applied Sciences"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.23943", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.23943", "source_url": "https://openalex.org/W7165770401"}, "key": "claire-radar:2606.23943", "modality": null, "name": "QuechuaTok", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_quechuatok_31b7234f", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "eb0fd916466cc9ff29efdc93979730199a2ed7497773ffea028f5a3211c56ee8", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.23943"}, "publisher": null, "released": "2026-06-22", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.23943", "source_url": "https://arxiv.org/abs/2606.23943"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-23943", "source": "claire_radar", "source_benchmark_id": "2606.23943", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Software & AI Compute"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2606.23943", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-04", "state": "stale"}}, "status": "stale"}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "7790df887552703ec0ab8f7d6a75a3d5ebd8533a63c2dccb304189ca55d56638", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A comparison of tokenization strategies (BPE, Unigram LM, WordPiece, PRPE) for Southern Quechua using a 200k-sentence corpus and a finite-state morphological analyzer as reference, with metrics including fertility rate, OOV rate, and morphological boundary accuracy.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence"], "snippet": "We present QuechuaTok, a systematic benchmark comparing four tokenization strategies - BPE, Unigram LM, WordPiece, and a morphology-aware PRPE tokenizer - for Southern Quechua (quz), a low-resource agglutinative language spoken by 8-10 million people in South America."}, "familyId": "bmf_7dbc10d7377f", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_quechuatok_31b7234f", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Software & Cloud"], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.23943", "paper": "https://arxiv.org/abs/2606.23943", "pdf": "https://arxiv.org/pdf/2606.23943", "project": null, "report": "https://arxiv.org/abs/2606.23943"}, "metrics": [], "motivation": "Tokenization is a foundational step in NLP pipelines, yet standard evaluation metrics such as fertility rate fail to capture morphological correctness for agglutinative languages.", "name": "QuechuaTok", "oneLine": "A comparison of tokenization strategies (BPE, Unigram LM, WordPiece, PRPE) for Southern Quechua using a 200k-sentence corpus and a finite-state morphological analyzer as reference, with metrics including fertility rate, OOV rate, and morphological boundary accuracy.", "paperTitle": "QuechuaTok: Morphological Boundary Accuracy as a Necessary Metric for Tokenizer Evaluation in Agglutinative Low-Resource Languages", "primaryDomain": "Software & AI Compute", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.23943", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-22", "source": {"authors": ["Maria Contreras"], "categories": ["cs.CL"], "id": "2606.23943", "title": "QuechuaTok: Morphological Boundary Accuracy as a Necessary Metric for Tokenizer Evaluation in Agglutinative Low-Resource Languages", "type": "arxiv", "url": "https://arxiv.org/abs/2606.23943"}, "sourceUpdatedAt": "2026-06-22T21:04:48Z", "topics": ["cs.CL"], "venueAttempts": [], "whyItMatters": "Evaluates tokenizer quality for agglutinative low-resource languages, showing that fertility rate alone is insufficient and that morphological boundary accuracy provides a more meaningful signal."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
