open-standards/aac/aac-rs/tools/speech/review.json

{
  "date": "2026-09-29",
  "about": "The review behind crates/aac-core/src/speech_lexicon.json. For each site language: the board glosses (translations_data.json) and default-board words an engine could misread when said alone — single letters, single kana, lone particles, prepositions and conjunctions, abbreviations, bound forms — and what was decided. 'left_alone' words were checked and need no spoken form; 'unresolved' words are misread by at least one engine but no spelling was found that fixes them, or the engines cannot be heard well enough to tell; 'gloss_problems' are glosses a spoken form cannot fix, listed for native review (the glosses themselves are not changed here). check_pronunciation.py speaks every lexicon entry and every 'left_alone' and 'unresolved' word. Model-reviewed only: no native speaker has checked any of this.",
  "all_languages": {
    "gloss_problems": [
      { "gloss": "the", "problem": "Languages with no definite article gloss 'the' as a demonstrative or third-person pronoun (ur وہ, fa آن, mr तो, gu તે, pa ਉਹ, ta அந்த, te ఆ, ml ആ, kn ಆ, th นั้น, id/ms itu, sw ile, yo náà, fi se, pl/cs ten, uk цей, vi cái). Engines say these correctly, but the cell then means 'that' or 'he'. A native reviewer should decide whether the 'the' cell belongs on these boards." },
      { "gloss": "default-board words with no gloss", "problem": "These default-board words have no entry in translations_data.json, so every language speaks them in English with its own voice: is, for, ball, hug, blanket, question, rest, shoe, silly, student, helper, house, better, worse, worried, hurting, emergency, OK, how many, why not, what's that, my turn, your turn, and the board names Actions, Core, Describing, Nouns, People, Places, Questions, Verbs, Speak, Clear." }
    ]
  },
  "languages": {
    "en": {
      "left_alone": {
        "a": "Lower-case, so not announced as a capital; engines give the article's citation form /eɪ/.",
        "TV": "Said as the letters T V, which is how the word is said.",
        "OK": "Apple voices say 'okay'."
      }
    },
    "es": {
      "left_alone": { "a": "The letter's name /a/ is the preposition." }
    },
    "fr": {
      "left_alone": {
        "où": "Said /u/ without naming the accent (Apple, eSpeak).",
        "là": "Said /la/ without naming the accent."
      }
    },
    "de": {
      "gloss_problems": [
        { "gloss": "favorite", "written": "Lieblings-", "problem": "A bound prefix (Lieblings-essen); it cannot stand alone. Engines drop the hyphen and say 'Lieblings'." }
      ]
    },
    "pt": {
      "left_alone": { "TV": "Said as the letters 'tevê', which is the word." },
      "unresolved": {
        "é": "The verb 'is' (not a board gloss, but a common one-word reply). Apple voices say /ɛ/; eSpeak says 'é agudo'. No spelling fixes eSpeak without changing the vowel ('eh' gives /e/)."
      }
    },
    "it": {
      "left_alone": {
        "a": "The letter's name /a/ is the preposition.",
        "TV": "Said as the letters 'tivù', which is the word."
      }
    },
    "nl": {
      "left_alone": { "TV": "Said as the letters 'tee-vee', which is the word." }
    },
    "zh": {
      "unresolved": {
        "教": "'teach' is jiāo; Whisper heard 叫 (jiào) from Tingting, but Whisper does not hear tone reliably. Needs a native ear.",
        "干": "'dry' is gān; gàn ('to do') is the other reading. Whisper output was unusable.",
        "数": "'count' is shǔ; shù ('number') is the other reading. Whisper heard 书 (shū).",
        "量": "'measure' is liáng; liàng ('amount') is the other reading.",
        "空": "'empty' is kōng; kòng ('free time') is the other reading.",
        "累": "'tired' is lèi; Whisper heard 雷 (léi).",
        "切": "'cut' is qiē; qiè is the other reading."
      },
      "left_alone": {
        "长": "'long' is cháng; Whisper heard 'Chang' (cháng), not zhǎng.",
        "还": "'still' is hái; Whisper heard hāi, not huán."
      }
    },
    "ja": {
      "left_alone": {
        "と": "The particle is read 'to', as written.",
        "ひ": "Read 'hi'.",
        "て": "Read 'te'.",
        "め": "Read 'me'.",
        "は": "Read 'ha' (not the topic particle 'wa') when alone; correct for 'teeth'.",
        "き": "Read 'ki'.",
        "え": "Read 'e'.",
        "も": "Read 'mo'.",
        "に": "Read 'ni'; Whisper heard the number 2.",
        "ご": "Read 'go'; Whisper heard the number 5."
      },
      "gloss_problems": [
        { "gloss": "four", "written": "し", "problem": "Said alone, four is よん (yon); し is the counting-sequence reading. Whisper heard 4, so engines say shi." },
        { "gloss": "nouns, verbs and adjectives", "written": "hiragana", "problem": "Almost every gloss is in hiragana where Japanese writes kanji (手 目 歯 木 日 絵 本 家 水 山 川 雨 紙 …). Kanji tell the engine the reading and pitch accent; kana leave it to guess between homophones (あめ 雨/飴, かみ 紙/髪/神, はし 箸/橋)." },
        { "gloss": "nose, flower", "written": "はな", "problem": "Both glosses are はな (鼻, 花); written in kanji they would differ." },
        { "gloss": "need, be, stay", "written": "いる", "problem": "'need' is 要る, 'be/stay' is 居る; they differ in pitch accent and are indistinguishable in kana." }
      ]
    },
    "ko": {
      "left_alone": {
        "에": "Hangul syllables are read as sounds; letter names apply only to lone jamo (ㅇ is read '이응'), which no gloss uses.",
        "TV": "Koreans say the letters (티비); Whisper heard 'TV'."
      }
    },
    "ar": {
      "gloss_problems": [
        { "gloss": "the", "written": "ال", "problem": "The article is a prefix and cannot stand alone; engines say 'al'." },
        { "gloss": "promise, count", "written": "أعد", "problem": "Same letters, different words (أَعِدُ 'I promise', أَعُدُّ 'I count'); bare, the engine picks one for both cells. Vowel marks on the glosses would separate them." }
      ]
    },
    "he": {
      "gloss_problems": [
        { "gloss": "the", "written": "ה", "problem": "The article is a prefix and cannot stand alone. Apple (Whisper heard 'hey') and eSpeak (/he/) both say the letter's name. A spoken form 'הַ' gives 'ha', but a lone ה is more often a spelled letter, so it is not in the lexicon." },
        { "gloss": "to, don't", "written": "אל", "problem": "'to' is אֶל (el), 'don't' is אַל (al); both cells show אל and engines say 'al'. The lexicon maps a written word to one spoken form, so this needs vowel points on the 'to' gloss, or another gloss." },
        { "gloss": "there, name", "written": "שם", "problem": "'there' is שָׁם (sham), 'name' is שֵׁם (shem); engines say 'sham' for both." },
        { "gloss": "here, mouth", "written": "פה", "problem": "'here' is פֹּה (po), 'mouth' is פֶּה (pe); both cells show פה." }
      ]
    },
    "ur": {
      "gloss_problems": [
        { "gloss": "will", "written": "گا", "problem": "The future suffix (-gā) cannot stand alone." },
        { "gloss": "I, in", "written": "میں", "problem": "'I' is مَیں (main), 'in' is مِیں (mẽ); both cells show میں." }
      ]
    },
    "fa": {
      "left_alone": {
        "کی": "'when' (key); eSpeak says /kej/.",
        "پر": "'full' (por); eSpeak says /por/, not 'par'.",
        "کت": "'jacket' (kot); eSpeak says /kot/."
      }
    },
    "hi": {
      "left_alone": {
        "से": "A Devanagari syllable is read as its sound, never as a letter name.",
        "को": "Read as the postposition 'ko'.",
        "में": "Read 'mẽ'.",
        "पर": "Read 'par'.",
        "भी": "Read 'bhī'."
      }
    },
    "bn": {
      "left_alone": { "ও": "The letter's sound /o/ is the word 'too'." },
      "gloss_problems": [
        { "gloss": "the", "written": "টা", "problem": "The classifier -টা is a suffix and cannot stand alone." },
        { "gloss": "to", "written": "কে", "problem": "The case ending -কে is written joined to its noun; alone it is read as কে 'who'." }
      ]
    },
    "ta": {
      "gloss_problems": [
        { "gloss": "to", "written": "க்கு", "problem": "The dative case ending is a suffix and cannot stand alone; eSpeak says 'kku'." }
      ]
    },
    "te": {
      "gloss_problems": [
        { "gloss": "to", "written": "కు", "problem": "The dative case ending is a suffix and cannot stand alone." },
        { "gloss": "with", "written": "తో", "problem": "The instrumental/comitative case ending is a suffix and cannot stand alone." }
      ]
    },
    "ml": {
      "gloss_problems": [
        { "gloss": "to", "written": "ക്കു", "problem": "The dative case ending is a suffix and cannot stand alone." }
      ]
    },
    "mr": {
      "gloss_problems": [
        { "gloss": "to", "written": "ला", "problem": "The dative postposition is written joined to its noun and does not stand alone." },
        { "gloss": "toward", "written": "कडे", "problem": "Written joined to its noun (घराकडे)." }
      ]
    },
    "kn": {
      "gloss_problems": [
        { "gloss": "to", "written": "ಗೆ", "problem": "The dative case ending is a suffix and cannot stand alone." },
        { "gloss": "since", "written": "ನಿಂದ", "problem": "The ablative case ending is a suffix and cannot stand alone." }
      ]
    },
    "gu": {
      "gloss_problems": [
        { "gloss": "to", "written": "ને", "problem": "The postposition is written joined to its noun and does not stand alone." },
        { "gloss": "since", "written": "થી", "problem": "Written joined to its noun." }
      ]
    },
    "vi": {
      "left_alone": {
        "to": "A Vietnamese syllable ('big'); Latin letters in Vietnamese are read as syllables, not named.",
        "no": "A Vietnamese syllable ('full')."
      }
    },
    "ru": {
      "left_alone": {
        "о": "The letter's name /o/ is the preposition.",
        "я": "The letter's name /ja/ is the pronoun."
      }
    },
    "pl": {
      "left_alone": { "o": "The letter's name /ɔ/ is the preposition." }
    },
    "uk": {
      "left_alone": { "я": "The letter's name /ja/ is the pronoun." }
    },
    "cs": {
      "left_alone": { "o": "The letter's name /o/ is the preposition." }
    },
    "sv": {
      "left_alone": {
        "i": "The letter's name /iː/ is the preposition.",
        "yr": "eSpeak says /yːr/; Whisper could not hear Apple's reading (inconclusive).",
        "TV": "Said as the letters 'teve', which is the word."
      }
    },
    "da": {
      "left_alone": {
        "i": "The letter's name /iː/ is the preposition.",
        "tv": "Said as the letters 'te-ve', which is the word."
      }
    },
    "nb": {
      "left_alone": {
        "i": "The letter's name /iː/ is the preposition.",
        "tv": "Said as the letters 'te-ve', which is the word."
      }
    },
    "am": {
      "gloss_problems": [
        { "gloss": "the", "written": "ው", "problem": "The definite ending -u/-w is a suffix and cannot stand alone; eSpeak says 'wɨ'." },
        { "gloss": "before, after, during, since", "written": "ከ...በፊት, ከ...በኋላ, በ...ጊዜ, ከ...ጀምሮ", "problem": "Circumpositions written with an ellipsis for the missing noun; a cell speaks the two halves with nothing between." }
      ]
    },
    "ha": {
      "left_alone": { "i": "Hausa 'yes' is /iː/, the letter's own sound; lower-case, so no 'capital'." },
      "gloss_problems": [
        { "gloss": "the", "written": "-n", "problem": "The definite ending is a suffix and cannot stand alone." }
      ]
    },
    "yo": {
      "left_alone": { "ó": "The third-person pronoun is the vowel with a high tone." }
    },
    "zu": {
      "gloss_problems": [
        { "gloss": "the", "written": "i", "problem": "Zulu has no article; i- is a noun-class prefix and cannot stand alone." },
        { "gloss": "with", "written": "no", "problem": "na- fused with a noun's initial vowel (no- before u-); it does not stand alone." },
        { "gloss": "to", "written": "ku", "problem": "A locative/class prefix; it does not stand alone." },
        { "gloss": "have", "written": "na", "problem": "The associative 'have' needs a subject concord (ngina-, una-); na alone is not a verb." },
        { "gloss": "be", "written": "ba", "problem": "The verb stem -ba needs a subject concord (ngiba …)." },
        { "gloss": "tall, long", "written": "de", "problem": "The adjective stem -de needs a class prefix (-de → mude, lide …)." }
      ]
    },
    "tr": {
      "left_alone": {
        "o": "Lower-case 'o' is read /o/, the pronoun; only the capital is announced ('büyük o'), and the lower-case rule handles that.",
        "TV": "Said as the letters 'te-ve'."
      },
      "gloss_problems": [
        { "gloss": "to", "written": "-e/-a", "problem": "The dative case suffix in two harmony forms; it cannot stand alone, and the cell reads 'e a'." }
      ]
    },
    "el": {
      "unresolved": {
        "ο": "The masculine article /o/. Every engine names the lone letter (Apple 'μικρό όμικρον', eSpeak 'όμικρον', also with a full stop); ω is named too, and 'ω!' changes the word. Not a board gloss on its own."
      },
      "gloss_problems": [
        { "gloss": "the", "written": "ο/η/το", "problem": "Three articles joined by slashes; eSpeak reads the slash aloud in English ('o slash i slash to'). A cell should hold one word." }
      ]
    },
    "hu": {
      "gloss_problems": [
        { "gloss": "the", "written": "a/az", "problem": "Two forms joined by a slash; eSpeak reads it 'a per az'." },
        { "gloss": "in, on, with, to", "written": "-ban/-ben, -on/-en/-ön, -val/-vel, -hoz/-hez/-höz", "problem": "Case suffixes in their harmony forms; they cannot stand alone, and eSpeak reads the slash as 'per' ('ban per ben')." }
      ]
    }
  }
}