📄 Trends in the development of vocabulary for emotion and cognition in English: A millennial perspective

Summary

The paper compares how English built its noun vocabulary for thinking and for feeling over roughly a thousand years, from Old English to the present. Earlier etymological work on medieval English focused mostly on technical or everyday-activity domains. Cross-linguistic borrowability figures from WOLD also gave a puzzling picture: emotion terms seemed less open to loans than cognition terms. The authors drew nouns from two Historical Thesaurus of English (HTE) categories, MENTAL CAPACITY and EMOTION, and excluded the technical subfields PSYCHOLOGY and PHILOSOPHY. They traced each noun's etymology in the Oxford English Dictionary. Each item was annotated for semantic subfield, source language of its root, and morphological structure, and the two lexicons were made equal in size. Principal Component Analysis and Correspondence Analysis then exposed the patterns. COGNITION nouns lean on Latin roots and compounding, while EMOTION nouns lean on Norse-derived roots and conversion. Middle English was a key period for MEMORY, FEAR, COMPASSION, PLEASURE and SUFFERING, and word-formation evidence suggests French loans were integrated early. The study matters because it adds a diachronic view of the "private" mental domain to contact-linguistics research.

Key points

• Data source: nouns listed under HTE "The Mind > Mental Capacity" and "Emotion", with etymologies checked against the Oxford English Dictionary.
• Exclusions and balancing:
  • PSYCHOLOGY and PHILOSOPHY terms were left out as technical, field-specific vocabulary rather than words for mental processes.
  • The EMOTION set was cut down to match the size of the MENTAL CAPACITY set.
• Three annotation dimensions were used:
  • semantic subfield;
  • root source language (e.g. Latin, French, Old Norse, native);
  • formation type, distinguishing a root borrowed whole ("R") from morphological formations (curly brackets).
• Context from WOLD (22 semantic fields, 41 languages):
  • COGNITION sits mid-cline in borrowability (11th of 22);
  • EMOTIONS AND VALUES ranks lower (16th), despite Weinreich's view of affective vocabulary as an "onomastic low-pressure area".
• Analysis used PCA and Correspondence Analysis, both for the whole history of English and separately for the Middle English period.

Interesting ideas

• I1. Annotating where a word's root comes from separately from how the word was built (whole loan vs. native derivation or compounding on a foreign base) reveals how deeply borrowed material has been integrated.
• I2. Before comparing two lexical domains, the authors downsample the larger one to the size of the smaller, so differences reflect structure rather than sheer size.
• I3. They point out that taxonomy boundaries are partly arbitrary (WOLD puts "clever" under emotions although it fits cognition), and that such placements can drive the headline results.
• I4. Technical subfields are removed from a general-domain lexicon so that specialist jargon does not distort the picture of everyday mental vocabulary.
• I5. Correspondence Analysis over purely categorical annotations (subfield × origin × formation) finds which feature combinations co-occur without any bag-of-words representation.

For Mova

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

• I1 (tokens): Morphology analysis in crate en could record, for each lemma, both the root origin and the formation process, which helps explain lemmas for derived or compound words that are absent from the dictionary.
• I2 (learn): Evaluation comparing parsers or frame extractors across genres or domains should equalize sample sizes first, so that a bigger domain does not masquerade as a better-handled one.
• I3 (world): When building the ontology from WordNet/FrameNet, Mova could flag concepts that plausibly belong to several top categories and run sensitivity checks rather than trusting a single placement.
• I4 (world): Compiled verb and noun classes for mental states could separate everyday cognition and emotion words from discipline terms, so belief and feeling states in fables are modelled on core vocabulary.
• I5 (learn): The same technique could be applied to categorical error annotations from witness parsers (relation type × POS × construction) to discover which combinations cluster as systematic failures worth a repair rule.