📄 Derivation in a word-based morphology: on the origin of Old English verbs of perception, cognition and emotion

Summary

Díaz Vera traces how Old English built causative verbs in three semantic fields: physical perception, cognition and emotion. He also asks whether each field follows its own derivational "grammar of causation." The study is qualitative and etymological. It reconstructs Proto-Germanic sources for selected verbs and reads them against cognitive accounts of causation (Lakoff and Johnson's view of cause as forced motion, and Carruthers' medieval seal-in-wax image of perception). The central claim concerns the formative *-jα-, which the author links to an Indo-European root meaning roughly "set in motion." Attached mostly to nouns, it produced the new class of weak verbs. Perception verbs come from nouns for external sense organs, with the stimulus reshaping the organ. Memory and some emotion verbs come from internal organs such as heart, liver and mind, framed as possession or location. Abstract emotions, lacking an organ, derive from the emotion noun itself. The paper matters because it shows that morphology can preserve a culture's folk models of causation, and that these models differ by semantic domain.

Key points

• The suffix *-jα- first formed agent nouns, for example OE hierde "herdsman" and dēma "judge." It later spread to verbs and gave rise to the weak verb class, as the old ablaut system declined.
• Perception verbs are denominal from sense-organ nouns: hieran (ear), ME nese (nose), fēlan (hand/palm), byrgan (belly/body). Seeing is the exception, and scēawian has only a loose resemblance to ēage "eye."
• Feeling and memory verbs come from internal organs: hiertan "encourage" (heart), and gemynan and gemanian "cause to remember" (mind).
• lufian "love" (from liver) is a Class II weak verb. It is younger than Class I, since its root shows no i-umlaut, and it combines causative with intensive/iterative meaning. The paper notes that the same causative–intensive link appears in Indo-European, Old Chinese and Amerindian languages.
• Abstract emotion verbs are built on the emotion noun itself: eglan (sadness), lystan (pleasure), scēamian (shame), hreowian (regret).

Interesting ideas

• I1. The paper distinguishes three folk schemas of causation: pushing an object, a stimulus altering the experiencer's organ, and an emotion taking possession of a body part. Different verb fields thus come with different built-in causal pictures.
• I2. Verbs in one lexical domain tend to share a derivational pattern, while verbs in different domains diverge. Grouping words by meaning domain before studying their morphology therefore exposes regularities.
• I3. Where no concrete organ is available, the derivation falls back from the organ noun to the name of the emotion. This is a graded choice of base, ordered from concrete to abstract.
• I4. One morpheme can carry both causative and intensive/iterative meaning, possibly because forcing something implies effort.
• I5. The position of a verb in morphological history serves as evidence. For example, the absence of i-umlaut is used to date Class II verbs as younger.

For Mova

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

• I1 (frames): Perception and emotion frames could tag the stimulus as a causer that changes the experiencer's state, rather than treating these verbs as plain transitive actions, which would give clearer semantic roles.
• I2 (world): Verb-class knowledge from VerbNet/FrameNet could record per-domain derivational links (noun → causative verb), so the world builder can connect "heart" or "shame" nouns to the related emotion states.
• I3 (tokens): Lemma and derivation analysis could link a verb to its base noun through an ordered fallback list (body part, then the abstract noun), keeping each choice explainable.
• I4 (frames): Frame building could allow a single morphological cue to license both a causation reading and an intensity/repetition reading, and let context choose between them instead of fixing one sense.
• I5 (learn): Evaluation could use known diagnostic features as negative controls, checking that learned derivation rules do not misclassify forms that lack the expected marker.