Look up distinct dictionary senses for a word.

Returns stable sense IDs, parts of speech, glosses, examples, IPA, etymology, translations, and source provenance. Use an ISO/BCP-47 language tag such as en, de, es, ja, or zh-Hant. limit is the number of senses. Each detail limit is an independent total response budget, fairly shared across returned senses; 0 disables that detail class. Every sense lists any detail fields truncated by these fixed budgets.

Find synonyms grouped by sense and part of speech.

Pass a sense_id from dictionary_lookup when context matters. Candidates without a source sense are returned only as explicitly unsensed groups. max_senses controls how many lexical senses are inspected; unsensed_limit allocates part of the total candidate limit to all unsensed results. Set 0 for strictly sense-scoped candidates. Allocations must not exceed limit. count is the number of sense groups; candidate_count is the total number of nested synonym candidates governed by limit.

Translate a word while preserving its source-sense association.

Use dictionary_lookup first for ambiguous words, then pass its sense_id. Results are grouped by source sense and never silently cross senses. max_senses bounds the source-native senses inspected independently of limit. limit is the total translation-candidate budget, fairly shared across matching sense groups. count reports groups and candidate_count reports nested translations.

Find one directed lexical or commonsense relation.

Supported relations include antonym, hypernym, hyponym, meronym, holonym, derivation, etymology, use, capability, location, and related. Each result states its direction and provenance. Direct results have relation_scope="direct" and distance=1. Hypernym and hyponym queries may also return distance-2 results explicitly labelled relation_scope="transitive", with both sourced edges in path. max_depth is the relation graph hop limit (v1 supports one or two), while transitive_limit allocates part of the total candidate limit. Set max_depth=1 or transitive_limit=0 for direct results only. Allocations must not exceed limit.

Find distributional semantic neighbours with an optional language filter.

This uses the installed memory-mapped Numberbatch ANN index. It returns an empty unavailable response when semantic artifacts are not installed. Omit target_language for global multilingual results, set it equal to source_language for monolingual results, or use another tag for cross-lingual results. Similarity is finite cosine similarity in the range -1 through 1.

Find exact or near English rhymes.

Results are English CMUdict-backed headwords. near means exactly one ARPAbet-token insertion, deletion, or substitution. This fixed behavior has no automatic or configurable edit distance. The query itself is excluded from results.

Find corpus-backed wordplay candidates of one requested kind.

kind selects exactly one relation: anagram (eligible English headwords with the same normalized letters), palindrome (stored corpus palindromes; the query's own status is reported as input_is_palindrome), spoonerism (exactly two whitespace-separated English headwords whose initial consonant clusters are exchanged), or pun (exact CMUdict homophones whose source-native senses differ, returned as labelled candidates, never jokes). context is only accepted for pun; without it pun results are labelled context_scope="uncontextualized". Every result carries provenance.