📄 Lexical Database Design: The Shakespeare Dictionary Model

Summary

The paper describes the logical design of the Shakespeare Dictionary (SHAD) database, built at Münster since 1983 on a PRIME 250-II using CODASYL network-model DBMS software. The source data combine Spevack's machine-readable Shakespeare concordance, including stage directions, quartos and variants, with Finkenstaedt's Chronological English Dictionary. This lets Shakespeare's vocabulary be set against all English words recorded up to 1623. Neuhaus argues that earlier lexical databases copied the layout of printed dictionaries, building around a typographic "headword" with tilde back-references. He calls this an imitation design that confuses presentation with lexical structure. His alternative makes the lemma the central entity, linked one-to-many to its inflected word-forms. Relations between lemmata (the "macrostructure") are stored explicitly. A sample morphology schema links lemmata, segments, allomorphs and morphemes, and a virtual record exposes morphological families as direct queries. The work matters because it treats a dictionary as a linguistic structure rather than a print artifact. It also proposes a standardized lemma record as the link between lexical databases.

Key points

• SHAD runs on CODASYL network-model software. The team chose it over hierarchical IMS because a standard model gives long-lived data a migration path, and because lexical relations are rarely naturally hierarchical.
• About 20,000 lemma records carry a full morphology covering derivations, compounds and inflected forms. They also hold etymology, a taxonomic classification by content, and more than thirty fields in total.
• The morphology schema has four base entities: lemmata, segments, allomorphs and morphemes, ranging from 2,500 to 40,000 records. A virtual record called MORPHEME-TO-LEMMA supports retrieval at the lemma, allomorph and morpheme levels.
• The Latin-derived prefix {IN-} links through allomorphs such as im-, il- and ir- to almost 200 Shakespearean lemmata. A figure shows the {SPEAK} family, with first-occurrence datings and frequencies.
• The first word-formation database was built in three steps, taking about 17 hours in total. Access is through FORTRAN and COBOL batch interfaces and the VISTA query language.

Interesting ideas

• I1. Model the lexicon on linguistic units such as the lemma and its paradigm, not on the presentation conventions of a source resource; copying an old medium's layout bakes accidental structure into the data.
• I2. Decompose words into segments, allomorphs and morphemes so that a morphological family becomes something you can query directly.
• I3. Attach a first-attestation date to each lemma, so a vocabulary can be cut off at any year and set against the remaining words that existed by then but went unused.
• I4. Store relations between lemmata, such as synonymy, sense links and etymological groupings, as explicit links rather than leaving them implicit in entry text.
• I5. Use a standardized lemma record as the single interface for connecting external lexical databases.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

• I1 (tokens): Keep the en crate's lexicon keyed by lemma, with an explicit one-to-many link to word-forms, so defective and suppletive paradigms are stored as data rather than handled as exceptions in code.
• I2 (tokens): A morpheme layer would let the lemmatizer and tagger explain guesses for unknown derived words, such as un- plus base plus -able, by pointing to the family members that share a morpheme.
• I3 (world): Dating lemmas in the compiled knowledge layer could help flag anachronistic or archaic senses when reading old fables and tales.
• I4 (world): The WordNet/FrameNet/Wikidata ontology should keep these links as typed edges on lemma nodes, so every lexical inference can be traced back to a stored relation.
• I5 (world): A canonical lemma ID shared by VerbNet classes, WordNet synsets and Wikidata items would make cross-resource joins reliable and explainable.