Ingilingo — recovery and everyday senses

2026-10-12 · Installed foundations release used as baseline. Live index matched the baseline SHA-256 before building this release. VPS deployment remains user-run.

Release

{
  "new_distinct_pairs": 189,
  "new_editorial_pos_sense_records": 191,
  "existing_context_notes": 87,
  "editorial_proposals": 278,
  "attested_mapping_candidates": 7702,
  "new_form_entry_links": 7673,
  "controlled_editorial_form_links": 44,
  "new_vocabulary_from_forms": 0,
  "active_bilingual_records": 56735,
  "human_verified_new_entries": 0,
  "semantic_validation": "All candidates inspected for contextual distinctions during authoring; no comprehensive independent-source or human accuracy verification."
}

Measured retrieval coverage

Fixed 5,000 alphabetic word forms per language from wordfreq 3.1.1; includes names and inflections, not a list of unique lemmas. The improvement is chiefly recovery of existing headwords. This does not measure bilingual translation accuracy or dictionary completeness.

[
  {
    "language": "en",
    "denominator": 5000,
    "unit": "alphabetic word forms, including names and inflections; not lemmas",
    "before": {
      "exact": 3303,
      "grammar_explanation": 8,
      "explicit_form": 237,
      "folded": 0,
      "no_direct_match": 1452
    },
    "after": {
      "exact": 3311,
      "grammar_explanation": 8,
      "explicit_form": 1010,
      "folded": 0,
      "no_direct_match": 671
    }
  },
  {
    "language": "tr",
    "denominator": 5000,
    "unit": "alphabetic word forms, including names and inflections; not lemmas",
    "before": {
      "exact": 2326,
      "grammar_explanation": 0,
      "explicit_form": 120,
      "folded": 67,
      "no_direct_match": 2487
    },
    "after": {
      "exact": 2326,
      "grammar_explanation": 0,
      "explicit_form": 1537,
      "folded": 50,
      "no_direct_match": 1087
    }
  }
]

Inventory

{
  "before": {
    "active_translation_records": 56544,
    "unique_normalized_translation_pairs": 52076,
    "unique_english_headwords": 29401,
    "unique_turkish_headwords": 35726,
    "multiword_translation_records": 18818,
    "subject_records": {
      "Genel": 2885,
      "Hukuk": 710,
      "Muhasebe ve Finans": 218,
      "Tıp": 971,
      "Teknoloji": 180,
      "Oyun": 38,
      "İfadeler": 1474,
      "Deyimler": 905,
      "Atasözleri": 124,
      "Sözlük": 49039
    },
    "review_status": {
      "editorial_draft": 4886,
      "expert_review_pending": 80,
      "source_translation": 51334,
      "unspecified": 244
    },
    "sense_boundary_nodes": {
      "editorial_draft": 9941,
      "provisional_import": 103160
    },
    "sense_limit": "Imported sense nodes are provisional record-level nodes, NOT independently verified semantic senses. Editorial nodes are drafts; each pair has two language nodes.",
    "lookup_links_by_kind": {
      "editorial_alternative_name": 1,
      "explicit_inflected_form": 524,
      "inherited_alias": 4048
    },
    "withheld_translation_records": 2
  },
  "after": {
    "active_translation_records": 56735,
    "unique_normalized_translation_pairs": 52265,
    "unique_english_headwords": 29458,
    "unique_turkish_headwords": 35804,
    "multiword_translation_records": 18907,
    "subject_records": {
      "Genel": 3010,
      "Hukuk": 718,
      "Muhasebe ve Finans": 218,
      "Tıp": 971,
      "Teknoloji": 190,
      "Oyun": 38,
      "İfadeler": 1474,
      "Deyimler": 905,
      "Atasözleri": 124,
      "Sözlük": 49039,
      "İş dünyası": 40,
      "Muhasebe": 1,
      "Eğitim": 1,
      "Çevre": 1,
      "Finans": 5
    },
    "review_status": {
      "editorial_draft": 5077,
      "expert_review_pending": 80,
      "source_translation": 51334,
      "unspecified": 244
    },
    "sense_boundary_nodes": {
      "editorial_draft": 10323,
      "provisional_import": 103160
    },
    "sense_limit": "Imported sense nodes are provisional record-level nodes, NOT independently verified semantic senses. Editorial nodes are drafts; each pair has two language nodes.",
    "lookup_links_by_kind": {
      "attested_inflection": 7673,
      "controlled_inflection": 44,
      "editorial_alternative_name": 1,
      "explicit_inflected_form": 524,
      "inherited_alias": 4048
    },
    "withheld_translation_records": 2
  }
}

Lawful source use

BOUN Turkish and English EWT form/lemma annotations: CC BY-SA 4.0. BOUN cites manually corrected morphology; EWT morphology includes automatic annotation and subsequent corrections. Annotation is fallible. EWT underlying sentence-text rights differ from annotation rights: no source sentences are included. Only form/lemma/POS evidence and sentence/token identifiers are shipped. Dataset snapshot hashes, licences, contributor READMEs and exclusions are included. No translation import from these corpora; no protected dictionary copied.

Semantic limits and controls

{
  "scope": "278 authored proposals examined during model-assisted authoring for POS, naturalness, common confusions and restrictive contexts. This is not independent expert review.",
  "independent_bilingual_source_checks": 0,
  "human_verified": 0,
  "source_annotations": "All accepted mappings have at least two non-typo observed annotations, existing compatible POS targets and a constrained surface relationship. Turkish nonactive voice and adverbs excluded. No invented suffix analysis is applied at query time.",
  "exceptions": "104 original controlled Turkish form proposals, 44 genuinely new links; others already present. All model-assisted drafts, not corpus-attested claims.",
  "drafts": 278,
  "semantic_accuracy_percentage": null,
  "controls": [
    "causatives/other nonactive voice do not link to an active root in the new corpus mappings",
    "securities and minutes retain exact lexical senses ahead of form alternatives",
    "no legal-expert approval or imported translations from UD",
    "all original source fields retained; additive context notes only"
  ]
}

Installation and rollback

14 checks passed: compatibility, changed-master refusal, safe private storage, installation, integrity, source/alias preservation, reproducibility, repeated-install refusal and rollback protections. No production mutation during tests.

Retrieval validation

{
  "status": "passed",
  "checks": 129449,
  "active_records": 56744,
  "semantic_verification": "Retrieval checks do not establish bilingual semantic accuracy."
}

Actual app event/loader flow tested in a minimal DOM harness, including form submit, translation links, autocomplete, loading failure and retry, and inflection notices. No browser-engine rendering or physical iOS Safari check was possible here.

Remaining work

671 English and 1,087 Turkish benchmark forms still unmatched. These are unresolved forms, not a verified list of missing headwords. See remaining-gaps.json. Common-base coverage, contextual sense review, colloquial contractions, idioms, derivation, jurisdiction distinctions and qualified human review remain priorities.

278 editorial proposals

189 new normalized pairs, 191 new POS-specific records (two reuse an existing pair under another POS), 87 additive context notes. No reverse lookup, inflection or spelling alias counted as new vocabulary.