Source
Krauska and Lau (2023), LIT-031 (Frontiers in Language Sciences 2, 1125127, Hypothesis and Theory), read in full in NOTE-029.
What was actually shown
Two lexicalist assumptions (p. 2). (1) Syntactic and morphological processes differ in kind. (2) Lexical items are triads of sound, meaning and syntax. A model making either counts as lexicalist, so treelet models, which drop (1), still count by (2).
Where the models embed them (§2; C1, strong). Levelt’s lemma and Levelt and Indefrey (2000), the Consensus Model (Ferreira and Slevc 2007), incremental planning in “lexically sized units”, and treelets (Kempen, Vosse and Kempen, Ferreira, Matchin and Hickok).
Where they fail (§2.1; C2–C4, moderate).
- Inuktitut. For havauti-tuq-ti-taq-niaq-tara, “I’m going to give her medicine frequently”, a lemma model must either store a concept per word or load the “medicine” lemma with diacritics that are productive, rarely used and only linearly ordered, so they cannot express structure inside the word.
- Vietnamese idiomatic collocations (Noyer 1998). nhà cửa ‘house’ and vườn tược, where tược has no meaning alone, are single concepts whose parts syntax can separate. Superlemmas fail, since they require a literal lemma for every part.
- Hiaki suppletion (Harley 2014). Some verbs supplete for object number (me’a/sua ‘kill.sgObj/plObj’), so the verb’s form depends on an object that word-sized incremental planning has not yet reached (p. 8).
- Treelets keep the triad and word- or phrase-sized storage, so they still fail on Inuktitut and Hiaki.
Algorithms, not only representations (§2.2; C5, moderate). Inserting lemmas into slots as they are retrieved fails on Vietnamese. Separating content retrieval from structure building fails on Hiaki, since syntactic number is not available at morphophonological retrieval. Building structure after retrieval fails on verbs whose arguments are unplanned (devour). The resolution is “a single unified process of syntactic structure building” (p. 11).
What could have come out otherwise. A treelet model, which already drops the morphology–syntax split, could have handled the three cases with stored units alone. On the paper’s analysis it does not.
What this does not say
- That the paper’s own model works. The 8-stage non-lexicalist model is a two-page verbal sketch with no implementation, simulation or data (C7, weak). The full model is “in prep.” (p. 11). NOTE-029 says this paper would not source a THEORY about that model, and this account does not.
- That the aphasia predictions hold. Paragrammatism hiding syntactic errors and agrammatic inflection as a failure of cognitive control are hypotheses (C8, weak). The Kalaallisut data are five case studies cited as consistent, and the predictions are hard to falsify as stated.
- That no lemma-style encoding could work. The critique covers named models on selected phenomena. It concedes that Eberhard et al. (2005) can handle Hiaki object suppletion through feature sharing (p. 10).
- That the linguistic case against lexicalism is made here. The paper declines to argue it (p. 2) and does not mention Müller’s 2018 reply to Bruening. It takes the conclusion as given; see Connections.
- That stored units have no fixed size, or that “lemmatization” is an implementation-level optimization. Both are conjectures (C10), framed as “an interesting empirical question” (p. 8).
- That wordhood intuitions come from orthography. Speculation resting on one citation, Hoosain 1992 (C9). NOTE-029 corrects the seed that attributed it to Haspelmath.
Connections
- Presupposes THEORY-643. The paper’s assumption (1) rejected is Bruening’s single combinatorial component. It takes that conclusion from the linguistic literature (Bruening 2018 is in its §1 list) as a premise for a claim about processing models. Its assumption (2), the triad, comes from Distributed Morphology and Preminger (2021), which the record does not hold, and Bruening does not support it.
- It cites Haspelmath (THEORY-477) for the absence of empirical criteria for wordhood (pp. 2, 15), not for its production argument.
- NOTE-029 names ANTH-THEORY-021 and ANTH-THEORY-022 as touching the same theme, that the unit of storage is not the orthographic word. This paper is evidence for neither.