Guide to the Linguabase data tables

Every term ranked by familiarity, with single-word vs. multi-word and character counts, continuously to ~2M. Core is a curated set of 400,000 words — not simply the top 400K by rank: about a quarter are included or excluded by other criteria. Past the full ranking’s 400K mark, words simply get rarer; well past ~1.5M, the mix shifts toward compounds and technical terms — real English, but of little use for games and most purposes outside research or error detection.

Columns familiarity_score is 0–1 · in_core and core_rank reflect curated core membership, not a familiarity_rank cutoff · is_mwe flags a multi-word expression

Per-word signals for filtering by audience: iconability, game-suitability, thirteen accessibility dimensions (US-centricity, pop-culture dependence, sensitivity, religiosity, and others), and a single G–X content rating.

Columns iconability (0–1) · game_suitability (1–7) · content_rating (G–X) · + 13 accessibility dims scored 1–5: kid_appropriate, kid_familiar, cultural_scope, americana, time_stability, sensitivity, concreteness, register, domain_specificity, emotional_valence, geographic_english, pop_culture, religiosity

Roughly 100 weighted associated words per term, balanced across senses. Shown here formatted, as a word and its list of associates; the underlying file is a simple word-then-list row.

The same row as it sits in the file — word, a tab, then the pipe-delimited list:

elephant	tusk|pachyderm|trunk|elephant in the room|hippopotamus|mammoth|African elephant|Asian elephant|ivory|savanna|giraffe|Ganesh|lion|proboscidean|poaching

Columns associates — related words, strongest first; pipe-delimited in TSV, native list in Parquet

Note: Ganesh, proboscidean, and the elephant in the room rarely surface from an LLM prompted for the same word.

Each word mapped to its distinct meanings. A compact index for checking which senses exist before joining against their word pools.

Columns sense_labels — the list of senses the word has

The associations table split by meaning: a separate pool for each sense of each word.

Columns sense is the meaning this pool belongs to · associates is the related words for that sense

Note: senses and sense_associations were generated independently and use different sense descriptors — they are not aligned one-to-one.

Each word linked to its inflections, derivations, and etymological relatives. In a game these prevent trivial moves — reaching frost shouldn’t count as also reaching frosty. The curated set is additionally cleaned of false cognates.

Columns relatives — family members: inflections, derivations, etymological kin

Opposites per word, organized by axis — because a word can be opposed along more than one dimension. Each opposite carries a strength rating from 1 to 7.

Columns axis is the dimension along which the opposite holds · opposites is a list of term:rating, rating 1–7

The same opposition, but at the category level: a topic pool set against its opposite pool — a Cold category and its opposite Hot pool — again organized by axis.

Columns axis is the dimension of opposition · opposite_names is the opposing category names · opposite_members is member words of the opposing pool

Curated, symmetric opposite sets: two member pools opposed along a named axis, with scores describing how cleanly they separate. Shown here as two facing pools.

Columns score (overall quality of the pairing) · topic_a, topic_b (the two opposed topics) · axis (dimension of contrast) · topic_a_members, topic_b_members (the two member pools) · differentiation (how cleanly the pools separate) · member_strength (how strongly members fit their pool)

One readable paragraph per word, roughly 45–65 words, covering the major senses in plain prose. Written for display, not dictionary lookup, and independently authored — not drawn from any dictionary. The {curly braces} convention marks text to be set in italics.

Columns definition — one ~45–65-word paragraph; {curly braces} denote italics

Short hints for gameplay, from 26 angles — one-word hints, synonyms, categories, functions, contrasts, figurative reads, and learner/expert tiers. Trailing stars (*, **, ***) mark author favorites; n/a appears where a clue type doesn’t apply.

Columns clue_W1–clue_W5 (1–5 word clues) · 15 angle clues: clue_SYN, clue_DEF, clue_CAT, clue_FUN, clue_ASS, clue_PRO, clue_CON, clue_FIG, clue_EX, clue_ACT, clue_NEG, clue_SIT, clue_LOC, clue_SND, clue_OBV · clue_L1–clue_L3 (learner) · clue_E1–clue_E3 (expert)

A curated topic→members index: each category names a concept and lists the words that belong to it, with classification (scope, suitability, sensitivity) attached.

Columns members_core, members_loose (tighter / looser cuts) · game_suitability (1–7) · category_strength · sensitivity_topic, sensitivity_innuendo · alt_names · familiarity, relatedness

Built for the grouping-style game OtherWordly: short headwords paired with a sense, a one-line description, a member set, and an audience classification.

Columns description · quality_score · + audience signals: americana, obviousness, trivia_knowledge, age_threshold, pop_culture, known_us_adult, valence, register