[Elamur Dictionary](https://md.dictionary.elamur.ai/free/en/) · [this page in HTML](https://dictionary.elamur.ai/free/en/ideas/page/3/)

[dictionaries](https://md.dictionary.elamur.ai/free/en/) / ideas

# 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** — A deterministic gloss-overlap rule for ambiguous readings: among candidate senses of a word, prefer the one whose gloss shares content words with the English gloss words already in the frame. It is a table lookup, not the paper's neural cosine similarity. ([the paper](https://research.elamur.ai/free/papers/ctx-2025-saeed-lemmatization-as-classification-task-results/))
- **to try** — Use the lemma-POS-gloss triple as the unit of the Hebrew dictionary (dictionary hebrew): store the three together and measure how many candidates per word remain after the POS tagger, the same way the paper's Table 2 does for Arabic. ([the paper](https://research.elamur.ai/free/papers/ctx-2025-saeed-lemmatization-as-classification-task-results/))
- **to try** — Label each aligned span in the reader with the correspondence type of the hierarchy (types 1 to 4), derived from the UD trees of both sides. Count where the exact-mode residual falls per type. The question is whether the residual sits in type-4 places, which would point to the reader or the encoder, not to the lexicon. ([the paper](https://research.elamur.ai/free/papers/czulo-2017-crossroads-between-contrastive-linguistics/))
- **to try** — : a valency flag on dictionary entries for transitive senses that have gone out of use (bloom, famish, and the other chapter IV verbs). The parser could then mark a reading as archaic instead of failing. Measure attachments on the verses the book quotes (Numbers xvii.8, Zephaniah ii.11), and without the flag. ([the paper](https://research.elamur.ai/free/papers/davies-1875-bible-english-chapters-old/))
- **to try** — : list the index words that are missing from the standard lexicon, with the sense the quotations give, as a small seed file. ([the paper](https://research.elamur.ai/free/papers/davies-1875-bible-english-chapters-old/))
- **to try** — Use the weak-\*some\* diagnostics as a guard against false inferences. «We could use some rain» should not yield «not all rain». Add negative-control tests for the existential \*there\* and mass-noun cases. These are cheap to check with UD features (expl, mass nouns from the lexicon). ([the paper](https://research.elamur.ai/free/papers/degen-2015-investigating-distribution-some-but-not/))
- **to try** — and : write the noise categories as explicit rules (dropped apostrophe: "didn't" to "didnt"; "-ing" to "-in"; vowel removal), each generated pair carrying the rule that made it. A held-out gold set of learner or informal pairs then says which rules help. ([the paper](https://research.elamur.ai/free/papers/dekker-2020-synthetic-data-english-lexical-normalization/))
- **to try** — Closed-vocabulary check for derivation: generate a derived form and accept it only if it is in the lexicon or in the open text's counts, so that a plausible but non-existent form ("-ity" on a noun that takes "-ness") is pruned. Report accuracy and without the constraint. ([the paper](https://research.elamur.ai/free/papers/deutsch-2018-distributional-orthographic-aggregation-model-english/))
- **to try** — A closed lexicon of label nouns as an enum in (fact, issue, proposal, task, reply, criticism, and so), used by a rule: a demonstrative plus a label noun ("this task") gets an abstract-antecedent flag, and the antecedent search looks for a clause or a sentence rather than a noun. ([the paper](https://research.elamur.ai/free/papers/dipper-2012-use-parallel-comparable-data-analysis/))
- **to try** — Polarity pairs from the en lexicon as a Contingency signal. The paper finds opposite-polarity pairs help Contingency, not Comparison. Test this on our tale text, one relation at a time, and report per-relation F rather than total accuracy. ([the paper](https://research.elamur.ai/free/papers/disc-2009-pitler-automatic-sense-prediction-implicit-discourse/))
- **to try** — tagging: a lexicon of polarity items for the readings of high-negation questions. Positive polarity items ("some", "already", "too") force the outer reading (checking p); negative ones ("any", "yet", "either") force the inner reading (checking not-p). Test on constructed minimal pairs. ([the paper](https://research.elamur.ai/free/papers/domaneschi-2017-bias-polar-questions-evidence-english/))
- **to try** — Keep the statistical dictionary as the control. A table of how often a word is changed in the training pairs, applied only when seen, is the standard encoder with its lookup. Report its numbers before any learned delta is credited. ([the paper](https://research.elamur.ai/free/papers/domingo-2018-spelling-normalization-historical-documents-machine/))
- **to try** — Dryden's three heads as measurable operating points of the translator. Metaphrase is the pair-trained word-by-word table, paraphrase is the meta-world with an author delta, imitation is generation from the world level. Place each Bible on the axis by a literalness score: the share of a verse's meta tokens anchored by the longest common subsequence against MACULA's glosses. The four Bibles then get one number each to compare their deltas. ([the paper](https://research.elamur.ai/free/papers/dryden-1900-essays-john-dryden/))
- **to try** — Add an emotional-shift check to the irony candidates in : a positive reply (joy, praise, thanks) right after a narrated or quoted negative event or an angry line from another character is a context incongruity, the dialogue form of the implicit contrast already in the rule. Polarity of the previous frame comes from Mova's own polarity lexicon and the polarity field of the previous sentence. Measure precision on silver lines whose means names irony. ([the paper](https://research.elamur.ai/free/papers/ducret-2020-linguistic-features-automatic-sarcasm-detection/))
- **to try** — Keep a feature only when it raises the gate, as the paper drops changes that lower it. This mirrors the lexicon rule in open problem 2: a merge or a feature stays only if it lowers the residual. ([the paper](https://research.elamur.ai/free/papers/dusek-2012-formemes-english-czech-deep-syntactic/))
- **to try** — A closed list of direction words (four compass terms and six relative terms, such as "left" and "right") with every grammatical form counted per lemma. The paper's hand-made form-to-lemma list is the same mechanism as a lexicon table, and it gives a small Modern Hebrew lexical test. ([the paper](https://research.elamur.ai/free/papers/eacl-2021-weise-talk-me-left-right-angles/))
- **to try** — Build a small gold list of literal vs figurative lexemes from the piyyut corpus, and check whether the lexicon entries in Mova's Hebrew chapters mark late senses (for example the "kidney, guts" line, which the paper gives as an example of a figurative use). ([the paper](https://research.elamur.ai/free/papers/eacl-2024-toker-dataset-metaphor-detection-early-medieval/))
- **to try** — Run an extrinsic check, as the paper does: score an author's corpus with the register lexicon and compare with the human label. The delta is credited only if the score moves in the right direction. ([the paper](https://research.elamur.ai/free/papers/eder-2021-acquiring-formality-informed-lexical-resource/))
- **to try** — The regression is neural. The count form averages per-word scores over a chunk; build the lexicon from the Bible corpora and measure separately whether it tracks formality and whether it tracks archaism, since the two are different. ([the paper](https://research.elamur.ai/free/papers/eder-2021-acquiring-formality-informed-lexical-resource/))
- **to try** — Refuse merges across different nuance profiles. A lexicon merge of two concepts that differ in strength or type is probably noise, as the be/have merges already are. Make the merge check compare nuance profiles and count how many merges it removes. ([the paper](https://research.elamur.ai/free/papers/edmonds-1998-translating-near-synonyms-possibilities-preferences/))
- **to try** — Coverage report by part of speech, as in the paper's Table 6: for plus, report the share of unknown tokens by tag on UD EWT dev, to show where the lexicon is thin (the paper finds nouns and adjectives, not verbs). ([the paper](https://research.elamur.ai/free/papers/egedi-1994-freely-available-syntactic-lexicon-english/))
- **to try** — and : add a frame field to the pupil dictionary, so that each verb records its complement pattern (intransitive, transitive, with a sentential complement). Use it to check the verb-argument structure of a parse and to flag clauses that break the pattern, with the reason written out. ([the paper](https://research.elamur.ai/free/papers/egedi-1994-freely-available-syntactic-lexicon-english/))
- **to try** — Use the stage model of Anglo-French contact as an origin tag, so that a French-origin word can be explained by its stage. ([the paper](https://research.elamur.ai/free/papers/einenkel-1887-streifzuge-durch-die-mittelenglische/))
- **to try** — Chunk-to-chunk pairs learned from unaligned verse pairs. Enumerate the decompositions of the two trees into bounded chunks, weight them by the current counts, and keep a chunk pair (\*give a kiss to\* to \*kiss\*) only when it lowers the residual. This is the count-based version of the residual-driven lexicon (open problem 2). ([the paper](https://research.elamur.ai/free/papers/eisner-2003-learning-non-isomorphic-tree-mappings/))
- **to try** — : take the word-stock classes of Chapter VII (native, borrowed, compounded, derived) as the origin tags of dictionary entries, so that the school dictionary can say where a word comes. ([the paper](https://research.elamur.ai/free/papers/emerson-1894-history-english-language/))
- **to try** — Predicate-context incompatibility lists: predicates that only take individuals and predicates that only take abstract objects decide the anaphor type, with the list entry named in the explanation. Grow the lists from our own lexicon and measure their coverage on dialogue lessons. (to check: the paper's list of contexts; the 1999 card's I4 covers the same idea for the short version) ([the paper](https://research.elamur.ai/free/papers/enprag-2000-eckert-dialogue-acts-synchronizing-units-anaphora/))
- **to try** — and : measure the share of tagger errors on words missing from the AGID and UD dictionary in code (lexicon gap and unknown word, about 9% in the paper), on EWT dev. ([the paper](https://research.elamur.ai/free/papers/enud-2011-manning-part-speech-tagging-97-100/))
- **to try** — suspect labels: sort each tagger error on EWT dev into Manning's classes (inconsistent standard, gold wrong, difficult linguistics, lexicon gap, unknown word), and check that the suspect-label rule catches the inconsistent-standard class first. ([the paper](https://research.elamur.ai/free/papers/enud-2011-manning-part-speech-tagging-97-100/))
- **to try** — Default and shadow arguments as a source of commonsense defaults for the student, but rebuilt from WordNet glosses (WordNet 3.0 licence, already used by Mova) rather than taken from VerbAtlas: a verb sense whose gloss names its object or place (\*dine in\*: at home; \*drink\*: a liquid) gives the frame a default filler marked as implied. It fills «nothing remembered matched» cases where a question asks what was eaten or drunk and the lesson only says «the bear ate». Measure with world knowledge diagnose on patient questions; implied fillers must score below stated ones. ([the paper](https://research.elamur.ai/free/papers/fabio-2019-verbatlas-novel-large-scale-verbal/))
- **to try** — Selectional preferences as hypernyms (food, liquid, person, location) computed from Mova's own lessons per VerbNet class — the frequency of each WordNet top category among a class's agents and patients — as a soft score in (see stowe-2021-semlink-2-0-chasing-lexical I1 for the same table). VerbNet's own restrictions (+animate) stay as the hard part. ([the paper](https://research.elamur.ai/free/papers/fabio-2019-verbatlas-novel-large-scale-verbal/))
- **to try** — and the lexicon in : mark nouns that have count and mass uses, and let the determiner or the bare form choose the sense (\*a rabbit\* against \*rabbit\* as food). Test the choice on the lesson sentences that use these nouns. ([the paper](https://research.elamur.ai/free/papers/falkum-2017-lexical-pragmatics-count-mass-polysemy/))
- **to try** — Build a government table for prepositions and conjunctions from the entries (verb or adjective, then the preposition it takes), as a lexicon. Each row keeps its source entry id, so a parse can say why a preposition was attached to a verb. ([the paper](https://research.elamur.ai/free/papers/fernald-1904-connectives-english-speech-correct/))
- **to try** — Use a hypernym chain from as back-off features for words the tagger has not seen. Report OOV accuracy separately, because the paper shows that this is where the ontology helps. ([the paper](https://research.elamur.ai/free/papers/finch-2006-lexical-dependency-ontological-knowledge-improve/))
- **to try** — Build a substitution set from synonyms of the same UPOS (the paper's POS filter), keep the gold trees, and report the LAS drop of en parse-eval on EWT dev. ([the paper](https://research.elamur.ai/free/papers/findings-2021-wang-closer-look-into-robustness-neural/))
- **to try** — Measure how many substituted words are unknown to the dictionaries and, and report LAS separately for sentences with many unknown words. ([the paper](https://research.elamur.ai/free/papers/findings-2021-wang-closer-look-into-robustness-neural/))
- **to try** — A closed-class spatial lexicon for Hebrew: cardinal directions, "between" (בין), "next to" (ליד), "opposite" (מול), "north of" (צפונית ל) and the like, each mapped to a relation type in world (north-of, adjacent, between, along). The paper's counts show these carry most of the geometry, so start there and measure how many descriptions get a relation at all. ([the paper](https://research.elamur.ai/free/papers/findings-2023-argaman-hegel-novel-dataset-geo-location/))
- **to try** — Use mT5 only as a teacher: its segmentations of unpointed Hebrew could seed silver labels for the lexicon tables of the segmenter, checked against the gold treebank. The model never runs at test time. ([the paper](https://research.elamur.ai/free/papers/findings-2023-eyal-multilingual-sequence-sequence-models-hebrew/))
- **to try** — Lexicon-free check: measure the segmenter on words outside its lex table and on foreign or code-switched words, and without the table, and report the drop. ([the paper](https://research.elamur.ai/free/papers/findings-2024-shmidman-mrl-parsing-without-tears-case/))
- **to try** — Average-rank scoring for the translator: evaluate each variant (author delta or off, lexicon versions) over many tasks (author pair, book, granularity) and rank variants by their average rank, with the weights split hierarchically. Report the spread, not one number. ([the paper](https://research.elamur.ai/free/papers/freitag-2022-results-wmt22-metrics-shared-task/))
- **to try** — Verb position as features for the Hebrew reader. Verb-initial, verb-second and verb-final classes (open problem 7 for word order: verb-subject to subject-verb) can be counted from MACULA trees and used by the normaliser the paper's idea suggests. Measure the order agreement with the English gloss before and after. ([the paper](https://research.elamur.ai/free/papers/gamon-2002-machine-learned-contexts-linguistic-operations/))
- **to try** — Pivot counting for the translator's lexicon: for each English chunk that the Hebrew source aligns, count the English words aligned to the same source chunk, and rank the merges by the product of the two conditional counts. This gives a count-based ranking of candidate concept merges, which the residual then judges (open problem 2). ([the paper](https://research.elamur.ai/free/papers/ganitkevitch-2013-ppdb-paraphrase-database/))
- **to try** — Release the lexicon in tiers by score, as this paper does. The translator can then measure the reader and encoder bits at each tier and keep the largest tier where the residual still falls. ([the paper](https://research.elamur.ai/free/papers/ganitkevitch-2013-ppdb-paraphrase-database/))
- **to try** — Data size against quality. Vary the number of verse pairs used to learn the interlingua lexicon, and plot the meta-world's residual against it. The paper's trend, more data helps the pipeline, can then be checked for Mova's lexicon. ([the paper](https://research.elamur.ai/free/papers/gao-2026-modular-neural-machine-translation-semantic/))
- **to try** — Out-of-book generalisation. Learn the lexicon without one book, then translate that book. The generalisation gap is the Mova analogue of the paper's generalisation test. ([the paper](https://research.elamur.ai/free/papers/gao-2026-modular-neural-machine-translation-semantic/))
- **to try** — Schema annotations as lexicon data: for each class relation, record the connective preposition ("in" for a person in a company). This is a table, and it fixes the preposition errors the paper counts (68 of 100 samples). ([the paper](https://research.elamur.ai/free/papers/garimella-2021-domain-aware-dependency-parsing-questions/))
- **to try** — Role markers in : a preposition is a dependent of its left-hand word with a case-style role, and its head is the noun to its right. Try this as a lexicon-driven attachment rule and measure LAS/UAS on UD EWT dev, run twice for determinism. ([the paper](https://research.elamur.ai/free/papers/geary-2006-factored-functional-dependency-transformation-english/))
- **to try** — Generate MoNLI-style pairs from an open hypernym table (check which lexicon Mova can use), split PMoNLI and NMoNLI, and require 100% on both. The paper's below-chance NMoNLI result is the failure a rule system should not show. ([the paper](https://research.elamur.ai/free/papers/geiger-2020-neural-natural-language-inference-models/))
- **to try** — Give a deterministic interpretation step after detection, built from the paper's strategies, so that an irony candidate also gets a means line with its reason: replace a strong evaluative word by its antonym from the polarity lexicon (\*lovely\* → \*unpleasant\*); otherwise negate the main evaluative predicate and drop intensifiers (\*so kind\* → \*not kind\*); a rhetorical question becomes the opposite statement; ironic \*thank you\* / \*glad\* about a narrated bad event becomes «the speaker wishes it had not happened». Measure agreement with the teacher's means on the silver sentences marked ironic (the lion's «singing» example in is one). ([the paper](https://research.elamur.ai/free/papers/ghosh-2020-interpreting-verbal-irony-linguistic-strategies/))
- **to try** — Pair the antonym list with the polarity lexicon in (~65 positive, ~70 negative lemmas): each strong positive word gets its opposite, written by hand from public-domain sources (WordNet antonyms are open). This is the resource the interpretation step needs and is small. ([the paper](https://research.elamur.ai/free/papers/ghosh-2020-interpreting-verbal-irony-linguistic-strategies/))
- **to try** — The paper compares three ways of generalising over head words (clusters, WordNet, bootstrapping) and finds them about equally accurate where each applies. Mova's dictionary relations play the WordNet role; measure how many unseen head words they cover before adding anything learned. ([the paper](https://research.elamur.ai/free/papers/gildea-2002-automatic-labeling-semantic-roles/))
- **to try** — Keep the lexicon with a reachability count: the paper's top-three rule (ranked by p(s\|t) times p(t\|s)) leaves some sentences unreachable. For each merge in the translator's lexicon, report how many verses lose their reading when it is dropped, instead of only the merge count. ([the paper](https://research.elamur.ai/free/papers/gimpel-2009-feature-rich-translation-quasi-synchronous/))
- **to try** — Add a ScaleUp table to the frames: an adjective or noun after "what a" or "how" yields a predicate with a positive or negative valence (the paper's "What joy!" versus "What a disaster!"). Keep the valence lexicon in a BTreeMap, with "underspecified" as a third value, and test it on the examples the paper gives (What joy!, What a hero!, What a disaster!, What a tragedy!, What a goal!, What a painting!). ([the paper](https://research.elamur.ai/free/papers/ginzburg-2023-exclaiming-non-sententially-exclamative-sluices/))
- **to try** — Frequency table: report Mova's accuracy by how often each replacement word occurs in the lesson corpus, as the paper does. This shows whether the lexicon answers depend on coverage. ([the paper](https://research.elamur.ai/free/papers/glockner-2018-breaking-nli-systems-sentences-that/))
- **to try** — Implement the paper's WordNet baseline exactly as described (hyponym gives entailment, hypernym gives neutral, antonym or a shared hypernym within two edges gives contradiction, other gives other) in as a control for lexical-inference questions, and measure it beside Mova. ([the paper](https://research.elamur.ai/free/papers/glockner-2018-breaking-nli-systems-sentences-that/))
- **to try** — One derivation, several readings: keep the reader's derivation (the parse with its frames) as the single object, and give it separate interpretations: the meta-world, the English string, and the Hebrew-side gloss. The interpretations are separate tables over one tree, so a change in one does not touch the others. This follows the split in between tree and world. ([the paper](https://research.elamur.ai/free/papers/gontrum-2017-alto-rapid-prototyping-parsing-translation/))
- **to try** — : when a PTB-style tree is converted, group a dictionary MWE as one subtree before head assignment, which avoids the multiple-head case; measure the count of multiple-head and cycle repairs on PTB sections (test only) before and after. ([the paper](https://research.elamur.ai/free/papers/gram-2016-kato-construction-english-dependency-corpus-incorporating/))
- **to try** — and : a dictionary of compound function words, used as a tagger feature so that the whole unit gets one tag; measure UPOS and LAS on EWT test restricted to sentences with such words, twice to check determinism. ([the paper](https://research.elamur.ai/free/papers/gram-2016-kato-construction-english-dependency-corpus-incorporating/))
- **to try** — Use the warning flag: a lemma found by stripping or marked doubtful should not count in the lexicon counts or. Measure how much the tale register ranking changes when flagged lemmas are left and when they are removed. ([the paper](https://research.elamur.ai/free/papers/gram-2020-trips-lemmatising-verbs-middle-english-corpora/))
- **to try** — Verb frames as tree families:, give each verb a class from a small closed set (intransitive, transitive, ditransitive, ditransitive with PP, sentential complement) and show that the class is what the frame reader needs. Start with the verb chapter's classes, not the full tree set. ([the paper](https://research.elamur.ai/free/papers/group-1998-lexicalized-tree-adjoining-grammar-english/))
- **to try** — Names gazetteer: the paper's gazetteer came only from training data and hurt named-entity recall. For the translator's name lexicon, use an outside list and keep training-only names as a separate, measured class. ([the paper](https://research.elamur.ai/free/papers/gruzitis-2017-rigotrio-semeval-2017-task-9/))
- **to try** — A categorial-variation table for English, built from derivations and the lexicon (\*hunger\*/\*hungry\*, \*death\*/\*die\*). It lets two meta concepts that differ only in category (\*his death\* / \*he died\*) count as one concept with a category feature. This is a candidate for the residual-driven lexicon (open problem 2). Keep a merge only if reader and encoder bits go down. ([the paper](https://research.elamur.ai/free/papers/habash-2002-generation-heavy-hybrid-machine-translation/))
- **to try** — : write the rewrite rules as an ordered table in a deterministic normaliser, and combine them with a dictionary lookup. Weight each rule by its accuracy on dev data, as the paper's weighted vote does. Measure exact match on held-out letters, run twice. ([the paper](https://research.elamur.ai/free/papers/hamalainen-2018-normalizing-early-english-letters-present/))
- **to try** — Presupposition triggers as rules: a lexicon of trigger patterns (definite descriptions such as "the only", ordinals such as "the first", factive verbs), emitting presupposition frames in prag. Check each frame against the world and measure the precision on the school questions. ([the paper](https://research.elamur.ai/free/papers/han-2023-prewome-exploiting-presuppositions-as-working/))
- **to try** — Frame identification is the step to measure first. For predicates with no lexicon entry, the choice is among all frames; Mova should report such "unknown" and "unlinked" cases separately, as the paper does,. ([the paper](https://research.elamur.ai/free/papers/hartmann-2017-out-domain-framenet-semantic-role/))
- **to try** — Report Mova's frame accuracy both and without the lexicon filter. The paper recommends the no-lexicon number because it shows the frame reader's quality without the lexicon's coverage. ([the paper](https://research.elamur.ai/free/papers/hartmann-2017-out-domain-framenet-semantic-role/))
- **to try** — Lexical axioms from the dictionary: compile the relations in into the paper's fixed implication forms (hypernym gives one direction, synonym gives both). Keep the per-category accuracy table as the report format. ([the paper](https://research.elamur.ai/free/papers/haruta-2020-combining-event-semantics-degree-semantics/))
- **to try** — The ambiguity score as a count, not an SVM: for each letter n-gram in a respelling, count how many pronunciations it has in a dictionary, and reject a syllable whose n-grams have more than one reading. Deterministic and explainable; measure its agreement with the judges' decisions on the paper's examples. ([the paper](https://research.elamur.ai/free/papers/hauer-2013-automatic-generation-english-respellings/))
- **to try** — Load only the lexicon entries whose concepts occur in the input verse. Measure speed and check that the output is byte-identical and without the pre-selection, twice on the same input. ([the paper](https://research.elamur.ai/free/papers/haugereid-2012-extracting-semantic-transfer-rules-parallel/))
- **to try** — Keep three gender fields per noun lemma in the Torah dictionary: form gender (from the ending, singular and plural separately), agreement gender (measured from the trees), referent sex (from the lexicon); a noun like אָב with plural אָבוֹת then shows its mismatch explicitly. ([the paper](https://research.elamur.ai/free/papers/hbo-1994-kroeze-three-dimensional-approach-gender-sex/))
- **to try** — (to check) Discourse markers as a closed class with functions rather than glosses: one table (marker → function → our rule → count) in the discourse chapter. ([the paper](https://research.elamur.ai/free/papers/hbo-2004-merwe-overview-recent-developments-description-biblical/))
- **to try** — A loanword layer in the Torah dictionary: tag Egyptian, Akkadian and Persian loans (from BDB's notes) so that their distribution can be counted by book. ([the paper](https://research.elamur.ai/free/papers/hbo-2005-schniedewind-prolegomena-sociolinguistics-classical-hebrew/))
- **to try** — (to check) A politeness lexicon for the pragmatics rules: עַבְדְּךָ / עַבְדֶּךָ (speaker as "your servant"), אֲדֹנִי (addressee as "my lord"), נָא, אִם־נָא מָצָאתִי חֵן בְּעֵינֶיךָ, בִּי אֲדֹנִי; a quote with any of them gets hedge = polite. ([the paper](https://research.elamur.ai/free/papers/hbo-2007-warren-rothlin-politeness-strategies-biblical-hebrew-west/))
- **to try** — In the Torah dictionary, order the senses of אִישׁ as: each/any member of a group; husband/representative; a man (male) — with counts from our trees, using Stein's syntactic cues (distributive "אִישׁ … אָחִיו/רֵעֵהוּ", construct or suffix, apposition to a collective). ([the paper](https://research.elamur.ai/free/papers/hbo-2008-stein-noun-biblical-hebrew-term-affiliation/))
- **to try** — List the performative verbs, with Austin's class, as a flag in the Torah dictionary. ([the paper](https://research.elamur.ai/free/papers/hbo-2012-andrason-making-sound-performative-qatal-explanation/))
- **to try** — A note in the binyan list: hitpaʿel forms whose meaning is the reflexive/passive of the qal (not of the piel) may be old tG forms; the Torah dictionary can flag hitpaʿel lemmas whose root has no piel in the corpus, as candidates. ([the paper](https://research.elamur.ai/free/papers/hbo-2012-hutton-morphology-tg-stem-hebrew-tirgalti/))
- **to try** — Gloss rule: do not translate זֶה/אֵלֶּה as "this/these" by distance; and read הַהוּא/הַהִיא adnominal as anaphoric ("the aforesaid"), resolving it to an earlier mention in the world, never to something in the speech situation. ([the paper](https://research.elamur.ai/free/papers/hbo-2014-kummerow-anaphora-deixis-tiberian-hebrew-semantically/))
- **to try** — A gloss rule: עבר + object/עַל with a law or covenant word (מִצְוָה, בְּרִית, פִּי יְהוָה) → "transgress"; with a river, land or border → "cross"; measure against JPS. ([the paper](https://research.elamur.ai/free/papers/hbo-2014-wilson-verbal-valency-biblical-hebrew-analysis/))
- **to try** — Build a valency frame table for עבר from the Torah trees: for each occurrence, the stem, the set of dependents (obj, obl with the preposition lemma), and the JPS verb used ("pass over", "cross", "transgress", "go through"). Clusters of frame + gloss give sense distinctions for the dictionary. ([the paper](https://research.elamur.ai/free/papers/hbo-2014-wilson-verbal-valency-biblical-hebrew-analysis/))
- **to try** — Generalize: a valency-frame column in the Torah dictionary for the 100 most frequent verbs, derived the same way and checked against BDB's sense divisions. This is cheap and makes the dictionary's senses explainable by syntax. ([the paper](https://research.elamur.ai/free/papers/hbo-2014-wilson-verbal-valency-biblical-hebrew-analysis/))
- **to try** — Keep the WLC text immutable and every layer (trees, prag, dictionary links) stand-off, keyed by word position; check that a rebuild never changes the base text (byte-identical gate). ([the paper](https://research.elamur.ai/free/papers/hbo-2015-roorda-hebrew-bible-data-laboratory-sharing/))
- **to try** — A default gloss rule from the sense profile: wayyiqtol → English simple past; qatal in narrative after a wayyiqtol chain or with X-qatal (subject or אֲשֶׁר first) → pluperfect ("had"); qatal in direct speech → present perfect. Measure agreement with JPS 1917 tense in the Torah, per text type (narrative vs speech). ([the paper](https://research.elamur.ai/free/papers/hbo-2016-andrason-complexity-verbal-semanticsan-intricate-relationship/))
- **to try** — Flag past-reference yiqtol without waw (archaic preterite) in poetry (Gen 49, Exod 15, Deut 32–33) as its own reading, so that the gloss rule does not render it future. ([the paper](https://research.elamur.ai/free/papers/hbo-2016-baranowski-biblical-hebrew-wayyiqtol-evidence-amarna/))
- **to try** — Build a small time field for the dictionary: יוֹם, חֹדֶשׁ, יֶרַח, שָׁנָה, בֹּקֶר, עֶרֶב, עוֹלָם, with dimension tags (objective unit vs cultic vs subjective/"remote time") as features, so that date expressions can be parsed into the world. ([the paper](https://research.elamur.ai/free/papers/hbo-2016-legnaioli-time-ancient-hebrew-lexicon-case/))
- **to try** — Split the Torah dictionary entry for חֹדֶשׁ into two senses by collocation: with an ordinal or number, or as "in the Nth month", → "month"; coordinated with שַׁבָּת/מוֹעֵד/חַג or in "head of your months" (רָאשֵׁי חָדְשֵׁיכֶם) → "new moon". Measure against the JPS rendering of each Torah occurrence. ([the paper](https://research.elamur.ai/free/papers/hbo-2016-legnaioli-time-ancient-hebrew-lexicon-case/))
- **to try** — A radial-sense layout for polysemous dictionary entries: core sense first, each extension tagged with its link (part-for-whole, medium-for-message, act, grammatical) so the entry says why the sense exists. ([the paper](https://research.elamur.ai/free/papers/hbo-2016-mylonas-speaking-one-s-heart-semantic/))
- **to try** — Sense rules for דָּבָר in the Torah dictionary, by construction: object of a verb of speaking or with לֵאמֹר → "word(s)/speech"; object of עָשָׂה, especially with הַזֶּה → "thing/deed"; indefinite under לֹא/אַל or in a question → "anything/nothing"; in construct before an abstract noun (דְּבַר הַמִּשְׁפָּט) → the kind named by that noun. Measure each rule against the JPS rendering of every Torah occurrence. ([the paper](https://research.elamur.ai/free/papers/hbo-2016-mylonas-speaking-one-s-heart-semantic/))
- **to try** — When דָּבָר stands for an utterance, the speech-act label of prag (command, request, report) can serve as its gloss; check whether the act of the quote that follows agrees with the English noun JPS uses. ([the paper](https://research.elamur.ai/free/papers/hbo-2016-mylonas-speaking-one-s-heart-semantic/))
- **to try** — Classify every אֵין in the Torah into the three types (suffix subject, NP subject, no subject + participle "no one who …") and gloss them differently ("he is not …", "X is not …", "there is no one who …"); check against JPS. ([the paper](https://research.elamur.ai/free/papers/hbo-2016-naude-contribution-qumran-historical-hebrew-linguistics/))
- **to try** — A small alias table of actors for the Torah (YHWH/Elohim; Israel/children of Israel/house of Jacob/the people; Moses/"the man"), written as data, applied after the grammatical steps, so the dictionary's names and the world share one actor id. ([the paper](https://research.elamur.ai/free/papers/hbo-2016-talstra-approaching-mountain-exodus-19-thou/))
- **to try** — Link Torah dictionary entries for concrete nouns (animals, plants, objects, places: תֵּבָה, גֹּפֶר, עֹרֵב, יוֹנָה, אֲרָרָט) to openly licensed images (public domain or CC BY on Wikimedia Commons) by a stable identifier (Wikidata QID), so the dictionary and the Torah course can show them; the link lives in the dictionary row, not in the text. ([the paper](https://research.elamur.ai/free/papers/hbo-2017-kofoed-corpus-driven-visualization-textual-worlds/))
- **to try** — Tag dictionary entries with a concreteness/semantic-field flag (animal, plant, artefact, place, person) as the hook for such links and for the world's object types. ([the paper](https://research.elamur.ai/free/papers/hbo-2017-kofoed-corpus-driven-visualization-textual-worlds/))
- **to try** — A rule for exhaustive exclusion: indefinite noun subject (no article, no suffix, not construct-bound to a definite) + לֹא + finite verb → read as "no X" in the English gloss and as negative universal in the world; check against JPS 1917 ("no razor", "none"). ([the paper](https://research.elamur.ai/free/papers/hbo-2018-merwe-fronting-exhaustive-exclusion-biblical-hebrew/))
- **to try** — For the English gloss side: when the rule picks the less usual subject, the JPS translation may still follow the usual one; log such disagreements instead of counting them as errors of prag. ([the paper](https://research.elamur.ai/free/papers/hbo-2018-naude-evolution-biblical-hebrew-linguistics-south/))
- **to try** — For the Torah dictionary: tag ע-roots with their Proto-Semitic value (/ʿ/ or /ġ/) where a cognate is clear, and split homonymous entries that the merger created (the paper's example is עֲוִילִים "unjust ones" vs "young boys"). ([the paper](https://research.elamur.ai/free/papers/hbo-2018-pietrzak-rhotic-phonemes-semitic-languages-case/))
- **to try** — Double-object verbs (מָלֵא, נָתַן with two bare NPs, הִשְׁקָה, הִלְבִּישׁ): list them from the dictionary's frames and check the trees give obj + iobj/obj consistently. ([the paper](https://research.elamur.ai/free/papers/hbo-2019-boulet-linguistic-reappraisal-biblical-hebrew-accusative/))
- **to try** — An Aktionsart field in the Torah dictionary, filled by the paper's yes/no tests (asked once per verb lemma, stored, reused), giving state / activity / achievement / accomplishment / semelfactive ± causative. ([the paper](https://research.elamur.ai/free/papers/hbo-2019-hojgaard-semantic-mapping-participants-legal-discourse/))
- **to try** — A sense split test for the dictionary: lemmas whose English alignment in JPS scatters over unrelated glosses (פָּנִים, בָּרָא) are candidates for homonym or idiom entries. ([the paper](https://research.elamur.ai/free/papers/hbo-2019-instone-brewer-tyndale-stepbible-data-development-machine/))
- **to try** — A two-pass metaphor finder for the Torah: first label noun animacy (binary) in the dictionary, then flag clauses whose inanimate subject governs a verb that usually takes an animate subject. ([the paper](https://research.elamur.ai/free/papers/hbo-2019-joode-problem-oriented-corpus-annotation-hebrew/))
- **to try** — A plain constituent display of a verse from our UD trees: one line per phrase, indented by depth, with the role (subject, object, predicate, oblique) and the MACULA English gloss, as an alternative view to CoNLL-U for readers of the manual and the course. ([the paper](https://research.elamur.ai/free/papers/hbo-2019-robie-adapting-digital-biblical-resources-translation/))
- **to try** — Make one base (the trees + dictionary) yield materials at several levels: a translator-level gloss line, a learner line with morphology, and the full tree, generated, not written by hand. ([the paper](https://research.elamur.ai/free/papers/hbo-2019-robie-adapting-digital-biblical-resources-translation/))