Bybee 1985 Ch 5: Dynamic Network Model of Lexical Representation #

@cite{bybee-1985}

This file formalizes the substrate of Bybee 1985 Ch 5 ("Two Principles in a Dynamic Model of Lexical Representation," pp. 111-135). It is a peer- framework directory alongside Theories/Morphology/{DM, PFM, WP, Nanosyntax, FragmentGrammars}/ — Bybee's network model is the usage-based competitor to the realisational and generative camps, and the audit (4-agent, 0.230.453) flagged its absence as the most important architectural gap in linglib's morphology layer.

The Two Principles #

Bybee abandons the structuralist/generative "is it in the lexicon or not?" binary (p. 116) and proposes two gradient + dynamic notions:

Lexical strength (§4 p. 117): each item entered in the lexicon has a strength that grows with each token of use ("etching it with deeper and darker lines each time"). Strength can also decline with disuse. Strength accounts for autonomy effects (Ch 3): high-strength items can be processed unanalyzed, retain irregularity, undergo inflectional split, develop fused/contracted forms.
Lexical connection (§5 pp. 118-119): each item bears multiple typed connections to other items. Three closeness factors: (a) number/nature of shared semantic features; (b) phonological similarity; (c) word frequency (high-frequency items form more distant connections — they're processed unanalyzed and don't lean on related items). The strongest non-identity relation is the morphological relation = parallel semantic + phonological connections. Identity (§7 p. 124) is the strongest possible relation: non-autonomous items map onto an existing representation rather than getting their own.

Five empirical phenomena lexical strength accounts for (§6) #

(p. 119-122): (i) maintenance of irregularity in high-frequency forms; (ii) high-frequency synthesis/fusion (English says/has/does; Spanish dar); (iii) lexical and inflectional split (wend/went split, supposed → spowsta); (iv) local markedness (Tiersma 1982 — a "derived" form becomes the basic one when more frequent); (v) development of contracted forms (don't, won't — Zwicky-Pullum 1983).

What this file does NOT cover #

Ch 5 §10 morphological classes as emergent from network connection density — deferred to a future SchemaInduction.lean sibling
Ch 5 §11 productivity from class size + variance — deferred
Cross-framework refutation theorem against DM (the bridge the cross-framework reconciler proposed: a paradigm where DM's subsetPrinciple predicts identical output but Bybee's strength predicts divergent retention) — deferred to a study file

Architectural commitments encoded here #

This substrate commits Bybee's central architectural claim: representations + connections form a graph where rules emerge from connection density, NOT a lexicon-rule binary. This is incompatible with DM's late-insertion-only architecture (HM 1993) and with PFM's realization-rule-only architecture (Stump 2001). Consumers that mix this substrate with DM/PFM should use bridge theorems in study files, not silently coexist.