Source
Haspelmath (2011), LIT-063 (Folia Linguistica 45(1), 31–80), read in full in NOTE-073 from the author’s deposit of the publisher’s PDF.
What was actually shown
The target. A criterion succeeds only if it picks out “words as identified by written spaces in the normal spelling used by speakers and/or by linguists” (p. 38). Four other bases are set aside first (§2): speaker intuitions, semantic non-compositionality, orthography itself and the phonological word.
Criterion by criterion (§3). Each of the ten criteria in (2), p. 38, meets at least one counterexample in each direction it would need to hold. Dalabon allows pauses inside a word, and clitics cannot be preceded by one. Bloomfield’s “minimum free form” makes firewater a phrase and take the money a word. Mobility depends on telling “same construction” from “different construction”, for which there is no criterion (fn. 18). Incorporation interrupts words (Pawnee). Edge inflection and English possessive ‘s are promiscuous without being words. Affixes are ellipted and coordinated (pro- and anti-war, Turkish suspended affixation). Anaphoric islandhood tracks non-referentiality, not wordhood, and Georgian allows inbound anaphora. Extraction targets referential phrases, so nonextractability says nothing about words (p. 52). Morphophonological idiosyncrasy occurs in clitic combinations. All four kinds of deviation from biuniqueness occur in syntax, (31)–(37).
The batteries (§4, Table 1, p. 60). Nine studies that combine criteria use different subsets. No criterion is used by all nine; non-selectivity and non-coordinatability are used by six, biuniqueness by one. The method “is not rigorous, because the criteria can be selected opportunistically by the author” (p. 59).
The consequence, as the paper hedges it. (41), p. 65: “Linguists have no good basis for identifying words across languages, and hence no good basis for a general distinction between syntax and morphology”. Morphosyntax is “for theoretical purposes … currently best viewed as a unitary domain” (p. 72), and the paper leaves open that a cross-linguistic word could still be found (p. 33; (42a), p. 66).
What could have come out otherwise. Any one criterion could have survived the counterexample search, or the batteries could have converged on a shared set. Neither happened.
What this does not say
- That the word cannot be defined in a given language. §5 and §10 say “word in language X” can often be defined consistently, even so as to predict the spelling (p. 71). The finding is about comparison across languages.
- That some non-orthographic intermediate unit is not universal. The target is orthographic practice by design (p. 38). A critic can accept every counterexample and still posit such a unit. Haspelmath’s reply is that it would not deserve the name “word” and that there are too many candidates to privilege one (§5).
- That every combination of criteria fails. The claim about combinations (C2 in NOTE-073) is argued from disagreement among nine studies, not from a search over weightings. A weighted or fuzzy word would rescue the distinction if units clustered on a boundness scale, and that is untested (§6, Fig. 1; fn. 24 says why a test is hard).
- That morphology and syntax are shown to be one domain. That step needs the further premise that both are defined through the word, which the paper asserts (p. 32, fn. 2) and does not survey. It is the Proposed half, and belongs with THEORY-643’s independent route.
- That sub-word units, or any tokenization, are privileged. Morphs fail biuniqueness in syntax as in morphology (§3.10). The paper says nothing about tokenization, and its practical advice, computational linguists included, is to keep the spaced word as a convention without theoretical status (§9). NOTE-073 corrects the seed that said otherwise.
- That the counterexamples are uncontested. Several rest on cited analyses that are disputed (Lezgian -ni, Japanese (19), Romance object markers as affixes). One per direction is enough to defeat “necessary and sufficient”, but the paper does not say how often each criterion fails.
Connections
- Bruening (2018; LIT-071, NOTE-065) cites this paper in fn. 2 and then grants, for the argument, that words can be identified. His case against a separate word component (THEORY-643) therefore does not rest on this one, and the two reach the same negative verdict on lexical integrity by independent routes. Haspelmath’s §§3.6–3.8 and Bruening’s §§3.1, 5.1 and 5.3 use the same diagnostics.
- Krauska and Lau (2023; LIT-031, NOTE-029) cite the 2017 reprint for “there are no good criteria to empirically define wordhood” (pp. 2, 15), and carry the conclusion into production models (THEORY-588).
- The author’s constructive sequel, “Defining the word” (WORD 69(3), 2023), is not held or read.