Bybee (1995): Regular morphology and the lexicon #
This file formalizes [Byb95]'s network model of productivity, in which schemas emerge
over stored forms and the productivity of a schema is determined by two things: its type
frequency, the number of distinct forms instantiating it, and the openness of its variables.
Token frequency does the opposite work, strengthening an individual form's own entry. Over a
corpus of tokens, a schema's type frequency counts the distinct forms it relates and its token
frequency counts every occurrence (typeFrequency_le_tokenFrequency), and a more general
schema has the larger type frequency (typeFrequency_mono). What a schema generates depends
on the corpus only through its types, since the lexicon of its roles is the corpus's set of
types, so repetition of a stored form leaves its generative capacity unchanged. Openness is the
other determinant: a schema generates every instance whatever is stored exactly when all of
its variables are open (Schema.isProductive_iff_forall_generates_iff), which is what makes
the English past in -ed fully productive (edSchema_isProductive).
Schemas are product-oriented: generalizations over the derived forms themselves, not over
pairs of base and derived form. The class of strung is the shape of its past tenses, and the
dialectal new members struck, snuck and drug fit that shape (struck_instantiates) while
their bases lack the vowel a source-oriented pairing from string would demand
(not_sourceOriented_pairs), the finding of [BM83].
Implementation notes #
- A corpus is a multiset of forms; the lexicon of a schema's roles is its set of types.
- The counts of Table 8.1, the French conjugations children overgeneralize by type rather than token frequency, and the German participle and plural cases are not formalized.
References #
Type frequency and token frequency #
The type frequency of a schema over a corpus: the number of distinct forms instantiating it.
Equations
- Bybee1995.typeFrequency s c = Multiset.countP s.Instantiates c.dedup
Instances For
The token frequency of a schema over a corpus: the number of occurrences of forms instantiating it.
Equations
- Bybee1995.tokenFrequency s c = Multiset.countP s.Instantiates c
Instances For
Types are at most tokens.
A more general schema has the larger type frequency.
Openness #
The English past in -ed, over a base slot and an affix slot: the base open, so totally
open in the sense of Schema.isProductive_iff_forall_generates_iff.
Equations
- Bybee1995.edSchema = ConstructionMorphology.Schema.productive ![⊥, ↑"ed"]
Instances For
The product-oriented class of strung #
The product-oriented schema of the strung class over onset, nucleus and coda: the nucleus pinned to /ʌ/, the onset open, the coda a closed variable filled from the class.
Equations
- Bybee1995.strungSchema = { body := ![⊥, ↑"ʌ", ⊥], opens := {0} }
Instances For
The stored members of the class.
Equations
Instances For
The dialectal new members fit the shape of the class.
Equations
- Bybee1995.instDecidableEqPairVar x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
A source-oriented schema pairing a base in /ɪ/ with a past in /ʌ/ over a shared onset and coda: the generalization over pairs such as string, strung.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The subscripting of the base's positions.
Equations
Instances For
The subscripting of the past's positions.
Equations
Instances For
string and strung are a paired instantiation of the source-oriented schema.
The new members are not: their bases lack /ɪ/, so no source-oriented schema pairs them, while the product-oriented schema admits them.