Ideas behind the dictionary

673 ideas from the papers Mova reads that touch dictionaries — lexicons, glosses, word senses — each with the paper it comes from and what came of it when Mova tried it. Those taken up first, then those still to try.

to try — Keep Mova's hbo layers as separate per-token columns keyed by + token index (trees, prag, dictionary links, accents, later prosody), so a new layer never requires regenerating the others; check that the parse database already does this and say so in chapter 13. (the paper)
to try — Detect the reciprocal pattern in the trees: אִישׁ/אִשָּׁה as subject + a later prepositional phrase whose noun is אָח/רֵעַ/אָחוֹת with a 3rd-person suffix agreeing with it → mark the clause as reciprocal (e.g. MISC Reciprocal=Yes) and gloss "one another"; count the Torah's cases and the prepositions used. (the paper)
to try — In the Torah dictionary, give אִישׁ and אָח/רֵעַ a sense "(in pairs) each … the other" with the reciprocal examples. (the paper)
to try — A distributional profile per lemma, generated from the trees: for a noun, counts of its relations (nsubj/obj/obl/nmod as head or dependent of a construct), the verbs that govern it, the prepositions it takes, and its coordinated or parallel nouns. Add it to each Torah dictionary entry as evidence for the senses. (the paper)
to try — Split nezer in the dictionary by context: in Num 6 with the Nazirite vow vocabulary (נָזִיר, נדר) → "consecration/separation"; elsewhere in the Torah (Exod 29:6, 39:30; Lev 8:9, 21:12) with the priest's head and the holy diadem → "diadem". A test of whether context words separate two senses of one lemma. (the paper)
to try — Give the Torah dictionary an etymology line for אֲשֶׁר (from "place"; short form שֶׁ־; not from Akkadian ša) and for פַּעַם (concrete "foot/step" > "occurrence, time"), marked as a scholarly proposal with this source. (the paper)
to try — A frequency-band reading view for the Torah course: hide or show the gloss of each word by its lemma frequency rank (e.g. show only words below the top 300), generated from the dictionary's counts. (the paper)
to try — Align the Hebrew trees with JPS 1917 subtree by subtree (bottom-up, keeping the best few candidates per node, with dictionary glosses and Strong's numbers as anchors) instead of word by word; measure alignment against a hand-checked sample of verses. (the paper)
to try — Consistency flag for translations: if the same multiword subtree (lemma pattern) is glossed differently in two places, flag it; apply to the Torah dictionary's examples and the course's word-for-word layer. (the paper)
to try — Use the twelve-word inventory with glosses in the Torah dictionary; the Torah has few (e.g. אוֹי in Num 21:29, 24:23), so the entry should say how many and where, with the rest cited from the Prophets. (the paper)
to try — A rule for prag and the gloss: a predicative participle in direct speech, especially after הִנֵּה / הִנְנִי, is read as "real present" (English present progressive or immediate future), not as a past. (the paper)
to try — A tense rule for the English gloss and the world: thetic S-qatal → pluperfect/resultative ("had captured"), placed before the current mainline event in time; topic-selecting S-qatal → simple past in sequence. Check against JPS 1917, which often uses "had". (the paper)
to try — Write Blum's inversion pairs as a rule for reading verb forms: when an x is fronted, wayyiqtol becomes x-qatal and weqatal becomes x-yiqtol, with the same aspect. Use it in the tense/aspect gloss so that a fronted-subject qatal in narrative is read as part of the past sequence, not as a perfect. (the paper)
to try — Synonym-choice profiles as a feature for the comparison chapter: in the Torah dictionary, group lemmas that BDB glosses alike (e.g. the two words for "maidservant", the two first-person pronouns) and count, per book, which one is chosen. Report the counts, not a clustering. (the paper)
to try — Give each Qal *qatil/qatul* lemma in the Torah dictionary a default aktionsart (state / activity / achievement / accomplishment), and list the contextual shifters (wayyiqtol, imperative, punctual time adverb) that move it to an event reading; e.g. a stative like "be heavy" or "be old" read as "became …" in wayyiqtol. (the paper)
to try — In glossing and in the world, read a stative verb in wayyiqtol as an inchoative event ("grew old", "became strong"), and in bare qatal as a present or resulting state; check the rule against JPS 1917 on the Torah's stative lemmas. (the paper)
to try — In the Torah dictionary, record for key nouns their governing verbs (the heads of which they are obj) and their coordinated synonyms, counted from the trees; show them as the entry's "frame", which separates senses better than glosses alone. (the paper)
to try — Gloss rule: בֵּין A וּבֵין B and בֵּין A לְB → "between A and B"; a check against JPS 1917 that no "and between" survives in the English. (the paper)
to try — Gloss and world: one event, with the minor verb as a modifier — קום + V "set about V-ing / promptly V", שׁוב + V "V again", לֹא יסף + V "no longer V", מהר + V "quickly V". Check against JPS 1917 ("rose up and went" vs. "went"). (the paper)
to try — A cheap, explainable check of the paper's prediction on our data: gloss overlap. For each root, compare the English glosses of the noun with those of the denominal verb and of the root's other verbs; denominals should share more gloss words with their noun. (the paper)
to try — In the Torah dictionary, mark verbs that BDB labels "denom." and, where visible, the extra pattern consonant (e.g. a verb built on a מ- or ת- noun); show the source noun in the entry rather than only the root. (the paper)
to try — Three readings of אֵין + (lə-)infinitive in the trees and the gloss: (a) existential "there is no one to V" (infinitive as a dependent of the noun or of אֵין); (b) impersonal "one may not / cannot V"; (c) personal, with a lə- subject, "X need not / cannot V". Write the patterns into the particles chapter with Torah examples (Gen 2:5 is in our data). (the paper)
to try — For Mova's own Hebrew analysis, look words up whole first (a full-form lexicon from the Torah trees and the dictionary) and only then split off prefixes and suffixes; report accuracy by number of segments per word, the paper's diagnostic. (the paper)
to try — Context-activated sense: when a "boil" verb has a sister or nearby argument naming water or a pot (בַּמַּיִם, סִיר), mark the gloss "boil"; otherwise give the default with "cook" as an alternative. A small rule over the trees, checked against JPS 1917 ("sodden", "boil", "seethe"). (the paper)
to try — Count, over the Torah, each lemma's share of occurrences with an explicit instrument/medium argument, as a cheap measure of how much a verb's default meaning carries — a column for the dictionary. (the paper)
to try — Dictionary senses split by stem: for בשׁל the Torah dictionary should list "ripen" under qal/hiphil and "boil, cook" under piel/pual, and pick the sense by the stem in the tree, not by one gloss per root. (the paper)
to try — Note the Ex 12:9 / Deut 16:7 tension as a worked example of lexical semantics in (polysemy, default construal). (the paper)
to try — Spelling normalisation: map plene and defective spellings to one consonantal key (drop medial ו/י as matres) when looking a word up in the Torah dictionary; measure how many Mishnah word types then find a Torah lemma. (the paper)
to try — Dictionary: give לִפְנֵי the senses "before (place, time)", "in the service of", "in the face of (a threat)", chosen by the governing verb. (the paper)
to try — The principle carries over to text: build synthetic "damaged" copies of Torah verses by rule (a letter lost, a lacuna, the classic confusions ד/ר, ה/ח, ו/י, ב/כ) and measure how often Mova's dictionary lookup and the converter's tokenisation still recover the right lemma. A robustness test with a known answer, generated deterministically from a seed. (the paper)
to try — Lacuna filling as a measured task: delete 15% of words of a Torah chapter in 1–3-word gaps (by seed) and score a rule/frequency-based filler from the dictionary and trees; a baseline for later models. (the paper)
to try — Period matters for any Hebrew lexicon-based correction: a word list from the Torah must not be used to correct or normalise later Hebrew, nor the reverse. A rule for Mova's own lookups: each lexicon carries its period, and a lookup outside the period is flagged. (the paper)
to try — Period, not genre, is what separates Hebrew stages for a model: a quantitative argument for keeping Biblical, Mishnaic and Modern Hebrew lexicons and rules apart, to quote in the comparison chapter. (the paper)
to try — The measured confusion table of Hebrew letters (ח/ה, ד/ר, ג/נ, ב/כ, ם/ס, ח/ת, ו/י) is a ready, sourced list of visually similar letters: use it to generate deterministic "damaged" Torah verses and test how robust Mova's dictionary lookup is, and cite it in the script chapter. (the paper)
to try — A "disputed sense" field in the Torah dictionary: for words like תְּשׁוּקָה, keep BDB's gloss but list the alternatives with their sources (ancient versions, later uses, modern proposals), so the dictionary never presents a contested gloss as settled. (the paper)
to try — Evidence from later Hebrew (Qumran, Mishnah) and the ancient versions as a sense test: for a rare word, a gloss that fits the Torah but fails every later use is weak. A short method note for the lexicon chapter. (the paper)
to try — Rare lemmas (three uses or fewer in the Bible) carry glosses that rest mostly on tradition; count them in the Torah and mark them, so a low-evidence sense is visible in explain output. (the paper)
to try — טֶרֶם/בְּטֶרֶם + yiqtol = a past action not yet completed when the main event happened; a tense rule for the gloss and the world's timeline (Gen 37:18, 27:33). (the paper)
to try — Every sense in the Torah dictionary should carry its source and date (BDB 1906, OSHB gloss, our own reading), so a sense inherited from an older dictionary can be traced and questioned rather than taken as fact; this is lexicographic criticism done in data. (the paper)
to try — When two sources disagree on a gloss, keep both with their sources instead of merging them, and count how often BDB's gloss and OSHB's differ over the Torah lemmas. (the paper)
to try — An interference check for any Hebrew a teacher model produces: flag Modern Hebrew markers — plene spelling of piel and other forms the Torah writes defective (compare with the Torah dictionary's spelling column), lemmas absent from the Torah lexicon, missing כי/אֲשֶׁר where Biblical syntax requires them. Rule-based, explainable, a gate before any generated Hebrew enters a course. (the paper)
to try — Add a borrowing field to the Torah dictionary for the few candidates the paper discusses in Torah poems (e.g. Gen 49:4, 5, 11, 12, 20, 21, 22), with the type (direct, indirect, culture word, semantic loan, calque) and the confidence the paper gives, never more. (the paper)
to try — Mark the archaic poems (Gen 49, Ex 15, Num 23–24, Deut 32–33) as a register of their own in the parse database, so that counts in the grammar can be given and without them; their lexicon and syntax differ from Torah prose. (the paper)
to try — In the English gloss and the world reader, take the reading of a form from its profile and its context: qatal in a quote → test for perfect ("I have given") before simple past; qatal of a stative verb → present; wayyiqtol → past perfective unless inside a habitual chain. (the paper)
to try — Use the paper's six path statements as the default sense order for each form when Mova glosses a verb: try the senses in path order and let context choose (e.g. yiqtol: ongoing in a participle-like context, habitual in a past-iterative chain, future in speech, modal after a volitive or in a command). The order gives a deterministic tie-break that can be explained. (the paper)
to try — Poetic parallelism (the second colon restating the first with different words) is the case lexical overlap and the finetuned model both miss. A transparent attempt: align the two cola of a verse by tree role (same relation to the verb, same frame) and test whether the aligned words are related in the Torah dictionary (shared English gloss, same semantic field). Measure on the Torah's poems (Gen 49, Ex 15, Deut 32–33) by hand-checked pairs. (the paper)
to try — Name pairs as a lexicon test: רבה/רבא and אבה/אבא are distinct forms that copyists confused; a lexicon for and should keep them as separate entries and a normaliser must never merge them. (the paper)
to try — Dictionary hygiene: a lexicon entry should record whether it is attested in speech or text, or only in dictionaries and Bible translations; Mutzafi's classicisms are the negative examples. Mova's Aramaic lexicon tables can carry an attested field and refuse unattested forms as gold. (the paper)
to try — The Swadesh lists with etymologies are a small cross-variety test: for each basic meaning, the exponent in Syriac, Targumic, Talmudic and Neo-Aramaic varieties, checked against Mova's Aramaic lexicon for coverage. (the paper)
to try — A multilingual glossary of the grammatical terms (Hebrew, Aramaic, Syriac, Arabic) for the script and phonology chapters, so that a term met in a Masoretic or Syriac source is mapped to the chapter's own term; it doubles as a lexicon test for on mixed technical passages. (the paper)
to try — Shewa as "moved" vs "still": the medieval binary (mutaḥarrik / sākin) is exactly the decision a syllabifier has to make; name the two outputs that way in the phonology rules, with the medieval term as the gloss. (the paper)
to try — A precision-first merge rule: the paper's lesson is that low-precision lexical links hurt. Apply it to the concept merges in the lexicon (open problem 2): keep a merge only if its precision on held-out verses is above a set threshold, and fall back to the copy otherwise. (the paper)
to try — Use the book's per-affix ranking as an explicit, explainable feature in world when the reader derives a headword from a prefixed form, so each decision can say which affix rule fired and how segmentable it is. (the paper)
to try — Make the explanation of an irony candidate in follow the paper's annotation units: the evaluation span, its modifier (intensifier), its target, and the target's implicit polarity with its source (polarity lexicon, a negative situation from the seed list, or a narrated frame). The incongruent pair is already printed; adding target and source makes the decision checkable by a reader. (the paper)
to try — Grow the 45 seed negative situations of into a lexicon of prototypical sentiment for situations, built from Mova's own texts rather than Twitter: in the tales and lessons, count for each situation frame (verb + object, or a state) whether the frames after it show a negative or positive reaction of the experiencer (cry, be afraid, groan vs laugh, rejoice) or an explicit evaluation. Keep multiword situations whole, as the paper finds summing words unreliable. Review the top of the list by hand before it is used. (the paper)
to try — A table of the book's periods and sections (sounds, morphology, syntax, lexicon) as a checklist of which period each en-x-emode chapter covers, so that no period is left without a source. (the paper)
to try — Sense-free keys: since sense IDs cost little for generation, count the Hebrew reader's words by Strong's number or gloss class, and compare with a table keyed by the finer sense. Pick the key by the residual, not by the argument. (the paper)
to try — Report the link rate between words and concepts: the share of source words that reach a concept in the meta-world, per author. This is the ontology-linking statistic that the report gives, and it shows where the lexicon is thin. (the paper)
to try — Weight the alternatives by a deterministic similarity (for example shared lexicon entries) rather than GloVe cosine, and measure whether the weighted score still predicts the paper's ratings. (the paper)
to try — Concreteness as a merge-risk feature. Humboldt predicts that words for physical objects match across languages and abstract words do not. Split the lexicon's merges and the shuffled-lexicon control by concreteness: concrete nouns versus abstract nouns, verbs of feeling and judgement. The prediction is that merges of concrete concepts lower the bits and merges of abstract ones raise them, which would focus open problem 2. (the paper)
to try — The lexicon as a "synonymy of languages". A concept should keep its members with their side meanings, not discard them. Each member of a concept carries features (era, register, number and person, connotation class) that the encoder can choose. This extends open problem 5 from FEATS to register and era. (the paper)
to try — Decomposable versus non-decomposable idioms as a test: an idiom whose parts carry meaning allows passive and internal modification, one that does not keeps its literal reading in those frames. Write minimal pairs and test that prag reads the idiom only where the class allows it. Negative control: the literal sentence must not get the idiomatic reading. (to check) (the paper)
to try — The class as an enum: every entry of the pragmatics expressions base and the en dictionary's multiword entries gets one of the paper's classes, and the class decides what the parser allows (words-with-spaces only for fixed expressions; inflection for semi-fixed; full syntax with a lexical link for flexible ones). (to check: the exact class names and boundaries) (the paper)
to try — Key the frames by sense and not by lemma: should carry the frame id per sense, as the *run* example shows. Compare lemma-level and sense-level frame accuracy on the same sentences. (the paper)
to try — A brittleness item: a pattern that passes in the short form and fails in a longer verse, to test the frame and encoder path rather than the lexicon. (the paper)
to try — Aquila is the baseline Mova already has: MACULA's glosses read word by word, which beat the Hebrew reader today. Add a stricter "Aquila" baseline that also writes a word for every prefix and for אֵת, and report both. If the reader cannot beat glosses, it should at least beat Aquila on the function-word class. That gives the function-word residual a floor and a ceiling. (the paper)
to try — Jerome's §7 example is an author adding emphasis words that have no source (*I say unto thee*). Count inserts with no source concept as their own sub-class of the reader residual, apart from lexical substitutions. Additions are the signature of a sense translator and should land in the delta, not in the lexicon. (the paper)
to try — : measure how many of the 1,524 original-to-modern pairs the archaic function already maps to the modern lemma, and which pairs it misses. The dictionary is a crawled list with no licence in our records, so use it for this test only. (the paper)
to try — Check for gaps in the dictionary before counting merges. The paper adds the second-person forms by hand because they are frequent but missing from the dictionary. The same audit for *thou*, *thee*, *ye*, *hath* and *doth* would show which author features the lexicon cannot yet reach. (the paper)
to try — Copy rule:, copy unchanged any word that has no dictionary entry. Measure the exact-match rate and without the rule, on the same test set. (the paper)
to try — The neural retrofitting is not for Mova; the count form is a co-occurrence table where dictionary links act as hard constraints on merges. (the paper)
to try — Use the dictionary as a prior on lexicon merges. A pair of forms linked by a dictionary gets a bonus for merging into one concept, and a merge is kept only if reader and encoder bits go down (open problem 2). Run it twice on dev and compare the merge set. (the paper)
to try — : build an expected-inference table from verb and frame signatures in the lexicon. For each embedded event, output the table's label next to the rule's label, and report the cases where they diverge. The table is the baseline a learned model must beat. (the paper)
to try — The sources matter: Urban Dictionary scraping is not usable under our rules. A Wiktionary dump is CC BY-SA, so quote only and do not carry it into a release. (the paper)
to try — Type constraints at tagging time: for known words, allow only the tags in the lexicon. Report accuracy on known and unseen words apart, because the paper's gains are almost all on unseen words. (the paper)
to try — and : a table of multi-word entries with gaps, such as either... or and particle verbs like show up, each stored with its positions. Measure how many gold multi-word units in UD EWT dev are recovered by the table. (the paper)
to try — Turn the "abstract for concrete" shift (§ 20) into a small lexicon of constructions with a date range, so that the early-modern mode can say which reading it used and why. (the paper)
to try — Store the meaning group of each verb-particle pair (directional, removal, completive, perfective, figurative) as a lexicon table. Then the school student can say whether "pick up" in a sentence is literal or figurative, and give the reason from the table. (the paper)
to try — Lexical sophistication for : build the word-level list from open corpora only (not from restricted lists) and count, per window, the share of words above a frequency cut. (the paper)
to try — : a deterministic linker for a historical lemma with only a gloss. Rank candidate modern senses by the overlap of collocates in a corpus against the neighbourhood of each candidate in the lexicon, with ties broken in BTreeMap order. Measure precision at 1 on a held-out set of hand-matched pairs. (the paper)
to try — : add a sense-level shift relation as a Rust enum (for example,), not a free string, and store it on senses that are linked to a neighbouring sense. Count how many school-dictionary senses could carry a shift. (the paper)
to try — : record the provenance of each sense as an enum (main dictionary, fallback dictionary, added by hand), and keep a fallback order in code. Each school sense then says where its gloss came. (the paper)
to try — The sense-shift link as a typed edge between two senses in the pupil dictionary, with the kind as an enum. Count how many multi-sense school words have a shift that could be labelled. (the paper)
to try — Use a semantic field as a unit of evaluation: take one field (emotion or shame words) and test the SLM student on questions about it in the school lessons, so a single field can be measured before the whole lexicon is covered. (the paper)
to try — : keep stopwords in the idiom-to-definition index. The ablation in the paper (128 against 156 median rank) is the first measure to repeat with Mova's own idioms. (the paper)
to try — Count-based collocates: compute PMI and frequency from counts in a BTreeMap, not a library model, and check the ranking against dictionary example sentences, with a shuffled-count control. (the paper)
to try — Sort the remaining failures by estimated fix cost (a lexicon entry, an encoder rule, a reader change). That gives the order of work across the eight open problems. (the paper)
to try — Use the A and B split for the lexicon work (open problem 2). Fit the merges on the A verses only, and report the B-verse residual after each change as the side-effect check. A merge that raises the B residual is noise. (the paper)
to try — Frame files for the school verbs: one file per verb with numbered roles (agent, theme, source, benefactive), kept or and mapped to the world's roles. Start with the verbs in the first school lessons. (the paper)
to try — Before writing any new author delta, rank per-verse features (source book, translator, OOV rate of the Hebrew concept, verse length, reader confidence) by mutual information with the residual bits. Features at the top point to where a delta or a lexicon merge will pay. (the paper)
to try — Test Mova's noun-event decision against QANom's verbal/non-verbal labels: decides by WordNet's first sense and event lexicographer files, without context. Measure its precision and recall on QANom's candidates (tests only, outside the repo, like NOMLEX in world frames nominal-check), and look at the context-dependent misses (*value*, *time*, *light*) for a rule (a determiner + *of*-phrase reads verbally; *time for* does not). (the paper)
to try — Keep Kruisinga's distinction between native-looking forms and forms he calls literary or spread by schooling. Store it as a register flag on the lexicon entry, so that the flag can say why a form is marked, and use it as an example for the register lessons. (the paper)
to try — Compute the four chain features on Mova's own English texts from the coreference output, plus lexical chains from the lexicon's hyponym and meronym links. Report chain number, length, distance and relation type per text, as a deterministic table, with the register label from the text's metadata. (the paper)
to try — Split compounds before keying the lexicon. The paper shows that unsplit compounds fragment the chains. Keep a rule-based splitter for the concept key, with BTreeMap rules and a count of how many keys change. This touches open problem 2 (phrase keys). (the paper)
to try — Verb-class feature from the lexicon: experiencer-subject verbs (*enjoy, fear, like*) lower the word-order weight and raise the animacy weight. Add this as a lexicon field, not a rule over surface strings. (the paper)
to try — The alternation pairs as a lexicon relation: XuYaZ is always the passive of a XiYeZ verb and huXYaZ always the passive of a heXYiZ verb; other pairs (XaYaZ↔niXYaZ anticausative/passive, XiYeZ↔hitXaYeZ reflexive/anticausative, XaYaZ↔heXYiZ causative) are possible but lexically listed. A root-indexed table of attested pairs lets the frame reader map a nifʿal clause to the same event as its paʿal partner, with the roles swapped. (the paper)
to try — Cognate search for Hebrew and Aramaic: join triples (Hebrew word, Aramaic word, gloss) from with a weighted correspondence machine for the letters, and check the top pairs against known cognate sets. (the paper)
to try — Keep a dictionary only for the ambiguous cases, as the paper does for broken plurals: list the Hebrew plural forms whose singular is not predictable, and report accuracy on a held-out set of unseen documents, not on the set that was tuned. (the paper)
to try — Make a root-versus-lemma test for Hebrew: for every root shared by two or more lemmas in the dictionary hebrew Torah dictionary, check the glosses and count pairs with unrelated meanings. The count measures the over-grouping that Froud describe for Arabic, and it tells us how far a root key can be trusted in world before we build on it. (the paper)
to try — Their error example (a noun mis-rooted to a verb of thirst) is a useful negative case for any root table built for Hebrew: a root must match the stem's consonant order and its gloss, not only its letters. (the paper)

This page in Markdown → https://md.dictionary.elamur.ai/free/en/ideas/page/4/