NOTE-065: Bruening 2018, lexicalism: wrong and superfluous

Contribution

The lexicalist hypothesis says that a separate word-formation component, encapsulated and feeding syntax one way, builds words out of reach of phrasal syntax. Bruening argues point by point that it makes false predictions. He then argues that every generalization it was credited with explaining follows anyway from constraints that do not mention “word”. The new constructive piece is a formal definition of X0 as a segment-based node type ((62), p. 25). With it, phrases inside words (a phrase adjoined to a null N0, (63)) and complex heads are both X0s. Restrictions on extraction and coordination can then be stated over X0 vs XP, and they cover both more cases (whole words cannot be extracted either) and fewer cases (ellipsis and focus are unaffected) than lexical integrity does.

Key insight

The facts that looked like evidence for a word/phrase boundary are really facts about heads vs phrases, and about what individual morphemes select. Heads and phrases occur on both sides of the orthographic word boundary. A principle stated over “words” is therefore either false (where syntax reaches in) or redundant (where something else already blocks it).

Assumptions

• What is being attacked. The target is the lexicalist hypothesis in the specific sense of distinct, encapsulated word and phrase components with one-way interaction. The target is not “lexicalist” in the sense of theories with rich lexical entries: “This article has absolutely nothing to say about this meaning” (p. 2).
• Parsimony. A one-component grammar is preferred “assuming that they are equivalent in their empirical coverage” (p. 1). The conclusion is that “all other things are equal”, and in fact the one-component grammar does better (p. 36).
• Words are granted for the sake of argument (fn. 2).
• A generative, broadly Distributed-Morphology picture of X0s.
  • Feature bundles are drawn from a universal set and bundled language-specifically. “Any bundle of features that can be spelled out as a morpheme … is an X0” (p. 24).
  • X0s combine by adjunction, following the formal definitions of Rogers 1998.
  • Adopting “something close to” the DM view of X0s is “not … an endorsement of other aspects of distributed morphology” (fn. 21).
• A syntagmatic approach with null heads. Examples are a null N0 that turns a phrase into a noun (fn. 6; (63)) and phrasal passives (96). He regards empty elements as unproblematic, “entirely a matter of opinion” (fn. 6).
• Judgments. The data are the author’s own English judgments, attested web and media examples ((4), (5), (9), fn. 10), and cited literature for other languages.

Key results

• Error 1: phrasal syntax feeds word formation (§2).
  • Phrasal compounds. Examples are a don’t-you-dare! look and a sweet I-love-you cake ((1)–(5)). The first member obeys phrasal syntax (the ill-formed variants in (3)), is productive, and is attested on the web and in ads.
  • Zero-derivation from phrases. Examples are know-it-all, has-been, how-to, a fine how-do-you-do ((6)–(8)), and verbs such as Bonnie and Clyded (9). These can feed further derivation (now-nowing, (10)). Bruening argues the pattern is as systematic as the suffix -age, which has ~150 forms (p. 6).
  • Overt derivation of phrases. Examples are do-it-yourselfer and unputdownable, pre-/post-/ex- attached to phrases, and Dutch -achtig/-loos (p. 6).
  • The “quotation” defence (Wiese 1996, Pafel 2015) fails. Anaphora crosses into the phrase: coreference, VP ellipsis, one-anaphora, NP ellipsis and e-type anaphora ((12)). A mechanism that turns any phrase into a noun stem is itself the thing the hypothesis forbids. The encapsulation it provides is just strict cyclicity (Mascaró 1976), which applies to every stem: Reaganomics, he-man, squirrel food, and word-internally Trump out-Trumps himself ((15)–(18)). Non-constituent strings are excluded (gimme-the), so the input is a phrase, not a string (p. 8).
  • German resultative nominalization (Leerfischung, (22)). Arguments of nominalizations are never obligatory (Williams 2015). So the obligatory resultative AP is explained if a phrase is nominalized, and it must be stipulated under Müller’s (2006) lexical rule. Bruening calls this “not decisive” but “suggestive” (p. 12).
• Error 2: phrasal operations access sub-word units (§3).
  • Word-part ellipsis (infra e ultrasuoni, Freund- oder Feindschaft, pre- and post-revolutionary, (23)). It cannot strand non-morphemes (bi- and ma-son, (24)–(29)), so it sees morphological structure. It is the same process as right-node raising and nonconstituent coordination on phrases ((30)–(31); Chaves 2008, 2014). The Japanese causative embeds a disjunction of VPs ((32); Kuroda 2003).
  • Focus targets morphemes (a-moral, not im-moral, (33)) and is sensitive to constituency (focus projection, (34)–(35)).
• Error 3: the claimed differences in principle fail (§4).
  • Locality (§4.1). Affixal self- allows delayed resolution (self-serving, (38)). Kichaga reciprocal skips an applied argument (41) where Chichewa does not (40).
  • Adjuncts (§4.2). Affixes do scope over adjuncts: the Venda causative (43), Japanese -sugiro, and self- with for-adjuncts and emphatic reflexives ((47)–(48)). Williams’s generalization comes from re-, which is idiosyncratically more constrained ((49)–(51)) but takes particles in its scope (52).
  • Headedness (§4.3). There are head-initial words (enrage), dvandva compounds (bittersweet) and head-final phrases (two years ago). The suffixing preference is referred to Hawkins & Cutler 1988 (fn. 16).
  • Idiosyncrasy (§4.4). Irregularity is pervasive in phrases too: idioms, particle verbs, collocations. Fn. 17 cites Jackendoff 1997 and Sag et al. 2002, who estimate that multiword expressions are at least as numerous as single words. Listedness explains nothing by itself. Where combinations are recorded is a question for every theory: either in root entries or in structured lexical entries (p. 22).
• Superfluous: each residual fact has an independent account (§5).
  • Extraction (§5.1). A- and A-bar extraction target only XPs. That rules out extracting word parts, and also whole words and non-phrasal parts of phrases ((55)–(58)), which lexical integrity does not cover. X0s are islands because extraction cannot see non-XPs (p. 27; (64)).
  • Coordination (§5.2). Apparent X0 coordination is phrasal. The but did not test ((66), (69)–(70)) answers Borsley (2005) and Abeillé (2006). This yields a typology of processes: extraction and coordination target phrases only; some morphemes select X0 only; ellipsis and focus target anything. “This typology makes no reference to the notion of a word” (p. 29).
  • Inbound anaphoric islands (§5.3). They are explained by the lack of content in pro-forms (Bresnan & Mchombo’s own account) or by pro-forms being maximal projections (Sproat 1988). Phrasal compounds allow pro-forms (73), and Georgian allows referential pronouns in derivation (Harris 2006).
  • Chomsky’s internal-structure argument (§5.4). It has lost force because LFG and HPSG treat gerunds lexically as well. The claim that derived nouns lack verbal structure is disputed (Fu et al. 2001; instrumental PPs, (80)–(82)). Bruening concedes “the matter requires more research to be settled” (p. 31).
  • Wh-how inside words (§5.5). How is a specifier of a phrase, so it cannot modify a subpart of completeness (84).
  • The no-phrase constraint (§5.6) (*[AP quite happy]-ness). Affixes differ in selection. -ness selects an X0 and does not preserve complements ((89)–(93)). Passive and nominalizing affixes select phrases and do preserve them ((94)–(95)).
• Conclusion (§6, p. 36). “If some hypothesis is both incorrect and does no work, the obvious action to take is to discard it.”

Claims

id	claim	strength	support
C1	Phrases productively feed word formation in English (compounds, zero-derivation, overt affixation)	strong	many attested and constructed examples, §2.1 (1)–(10). Cross-linguistic cases are cited, not shown (§2.3)
C2	The quotation analysis does not save the lexicalist hypothesis, because its encapsulation is just strict cyclicity, which holds of all stems	moderate	informal argument with diagnostics (12), (15)–(18). Fn. 9 concedes that not all phrasal compounds may be alike and leaves this “to future work”
C3	The German resultative nominalization favours a phrasal source	weak	a single case, explicitly “not decisive” (p. 12). Directly disputed in Müller 2018 (from the abstract)
C4	Word-part ellipsis is sensitive to morphological structure and is the same process as phrasal RNR / nonconstituent coordination	moderate	contrasts (24)–(29), and reliance on Chaves 2008, 2014
C5	Focus targets sub-word units and is constituency-sensitive	moderate	(33)–(35), with focus literature cited
C6	Word formation and phrasal syntax do not differ in locality, adjunct scope, headedness or idiosyncrasy	moderate	counterexamples to specific claims by Williams (2007) and Newmeyer (2009), §§4.1–4.4. This refutes those formulations, not every possible difference in principle
C7	Extraction targets only XPs, and this subsumes the extraction part of lexical integrity	moderate	(55)–(59) and the X0 definition (62). Müller 2018 argues X0 can be extracted (from the abstract)
C8	Apparent X0 coordination is always phrasal	moderate–weak	the but did not substitution argument, (66)–(70), fn. 24. Contested by Borsley, Abeillé and Müller
C9	The no-phrase constraint follows from affix-specific selection	weak	a proposal. The author says “a better analysis” may exist (p. 35). Supported by the complement-preservation correlation (89)–(95), shown for English only
C10	The residual constraints never refer to “word”	moderate	follows from how §5’s analyses are stated. It depends on C7–C9 being right
C11	”Word” probably has only a prosodic meaning	weak	assertion, “It is probable …” (p. 26)
C12	Lexicalist theories cannot answer where root–affix combinations are recorded any better than syntactic ones can	moderate	informal argument, §4.4
C13	One component is preferable to two on parsimony, given C1–C10	moderate	methodological argument; the parsimony is conditional on equal coverage (p. 1)

The abstract says the hypothesis is “wrong” and “superfluous” without qualification. The body is more careful in places: C3 is “not decisive”, §5.4 “requires more research”, and §5.6 and fn. 9 leave things open. The headline is therefore stronger than the least-supported sub-arguments. It is not stronger than the conjunction of the English evidence in §§2–3, which is the paper’s strongest part.

Concepts

• lexicalist hypothesis: the word-assembling system is separate from the phrase-assembling system, uses different principles, and interacts one way: word output feeds syntax, never the reverse (p. 1). Usually attributed to Chomsky 1970, and also called the lexical hypothesis (fn. 1).
• lexical integrity (Bresnan & Mchombo 1995): word parts are inaccessible to syntax. For Bruening it is a consequence of the lexicalist hypothesis, answered in §§5.1–5.3.
• X0 (62): “A node B is an X0 iff B is a segment of a syntactic node that includes nothing but a single bundle of morphosyntactic features”. Inclusion (61): A includes B iff every segment of A dominates B (after Chomsky 1986).
  • Complex heads and phrase-plus-null-N0 are X0s.
  • An XP is any node that is not an X0.
  • Most words are X0s. Not all X0s are words, and prosodic words can span non-dominated X0s (shouldn’t’ve, p. 26).
• strict cyclicity (in morphology): word formation cannot see inside the stem it attaches to (Mascaró 1976, Halle 1978). It is the real source of stem encapsulation (pp. 8–10).
• no phrase constraint (Botha 1981): phrases may not be input to word formation. Bruening treats it as affix-specific X0-selection (§5.6).
• delayed resolution (Williams 2007): an operator relates an argument to something non-local. Bruening shows it occurs in both morphology and syntax (§4.1).

Connections

Who cites whom.

• Bruening cites Haspelmath 2011 (LIT-063) twice:
  • fn. 2 (p. 1), for the non-existence of a universal word definition, which he then brackets;
  • §5.1 (p. 23), “cf. Haspelmath 2011:52”, for extraction targeting phrases.
• Bruening appears to predate Krauska & Lau (LIT-031) and does not cite them. They cite Bruening 2018 once, in their §1 list of works against lexicalism in linguistic theory (p. 2). They draw none of their production-model arguments from his specific analyses.

Agreement.

• With Haspelmath. Bruening’s §3.1, §5.3 and §5.1 are the syntactician’s versions of Haspelmath’s §3.6 (word-part coordination and ellipsis; both rely on Chaves 2008), §3.7 (Georgian imis-ian-i, Harris 2006) and §3.8 (extraction). Both conclude that lexical integrity is empirically empty. Haspelmath calls it partly circular (§8). Bruening calls it superfluous.
• With Krauska & Lau. Bruening’s single combinatorial component is their “assumption 1” rejected. His §4.4 point that idiosyncrasy and storage cut across words and phrases (fn. 17: multiword expressions are as numerous as single words) matches their §2.1.5 claim that stored units can be of any size.

Parting.

• From Haspelmath. Haspelmath says the word cannot be identified across languages. Bruening grants that it can and argues the hypothesis fails anyway. So Bruening’s case does not need Haspelmath’s, and would survive if Haspelmath were wrong.
  • Conversely, Bruening replaces the word with X0/XP and a universal feature inventory (p. 24). That commitment is theory-internal, and Haspelmath’s comparative-concept programme would not grant it without a cross-linguistic identification procedure. That last point is my inference. Neither paper says it.
  • Their constructive proposals also diverge: Haspelmath favours a realizational/constructional merge (his §3.10.5), Bruening a syntagmatic DM-like syntax (fn. 6).
• From Krauska & Lau. Their “lexicalism” includes a second assumption: the lexical item as a triad of meaning, syntax and form. Bruening explicitly declines to address lexicalism in the rich-lexical-entry sense (p. 2), and his paper says nothing about form–meaning–syntax mismatches as such. Krauska & Lau’s second argument is therefore not supported by Bruening. It rests on the DM/nanosyntax literature and on Preminger (2021).

The published reply. Müller (2018), Language 94(1):e54–e66, replies in the same volume. From its abstract: X0 can be extracted, some arguments of nominalizations are obligatory, coordination can affect lexical items, and non-English morphology may need a distinct component. Any filing of Bruening should link the reply.

Bearing on the record

• Nucleation. There are no THEORY documents to support or contradict. A THEORY stating that “the observed restrictions on syntax targeting word parts are restrictions on X0 vs XP and on affix selection, not on words” could be sourced from this paper as Proposed, with Müller 2018 named as the opposing reading.
• Anthology. There is no instruction for ML practice here, and nothing belongs in the Anthology of the SOTA.
  • ANTH-THEORY-022 (from ANTH-LIT-410, Goldberg & Shirtz 2025, Language 101(2)) describes the same construction as §2.1: a phrase, up to a whole sentence, in a slot reserved for a word (an ‘I’m not a witch’ moment). ANTH-THEORY-022 gets from it that the lexical unit is delimited by construction rather than frequency. Bruening supplies the linguistic-theory context: the construction is productive and obeys phrasal syntax internally ((3)). It is not a listed or “lexicalized” unit, contra Bresnan & Mchombo (p. 3). Its contents stay accessible to discourse anaphora ((12)).
  • That is consistent with ANTH-THEORY-022 and could be cited there as background for “productive, not listed”. Bruening says nothing about frequency or comprehension, so it is not evidence for ANTH-THEORY-022’s actual claim. Whether Goldberg & Shirtz cite Bruening is unverified.

Note of 2026-10-16: THEORY-643 now cites LIT-071 as a source.

Limitations

• The evidence is overwhelmingly English. Other languages (Dutch, German, Italian, Japanese, Chichewa/Kichaga, Venda, Georgian) are cited from secondary literature, often in a sentence. Müller’s reply presses exactly this. Fn. 28 leaves German how to future research.
• Several sub-arguments are explicitly provisional:
  • the resultative nominalization (p. 12);
  • derived vs underived nouns (p. 31);
  • the analysis of -ness (p. 35);
  • non-quotative phrasal compounds (fn. 9);
  • self- delayed resolution (“I do not have any account”, p. 17).
• “Superfluous” is shown case by case, for the arguments of Bresnan & Mchombo (1995), Williams (2007), Newmeyer (2009) and Chomsky (1970). It is not shown for every lexicalist framework. The author says criticisms “may not apply to certain other approaches” (p. 2).
• The X0/XP replacement carries its own cost. It needs null heads (fn. 6), adjunction-based segment definitions, and extraction that is blind to X0s. Whether that is simpler overall than a lexical component is argued by parsimony, not measured.
• Grammaticality judgments are mostly the author’s own. No systematic acceptability data are collected.
• Word-part ellipsis relies substantially on Chaves’s analysis, which is itself framed as operating on linearized strings (Chaves 2014). Bruening’s reply to that framing is the non-segmentable-morpheme contrasts in (24)–(29).

Open questions

• Can X0 be extracted in any language? Müller claims German evidence. If it can, the extraction account in §5.1 must become language-specific, and the paper itself anticipates “there may be languages where extraction can target an X0” (p. 26).
• Is apparent head coordination ever irreducibly X0-level? The substitution test (but did not) has to be checked in languages other than English.
• Do the affix-selection accounts (§5.6) predict which affixes accept phrases, cross-linguistically? The complement-preservation correlation in (89)–(95) is a testable prediction. The paper states it for English only.
• How do productive phrasal compounds differ from quotations (fn. 9)? A criterion is needed.

Corrections to the seeded skim

• Access: the udel.edu copy is a draft. The seed says the author’s own copy is at udel.edu/~bruening/Downloads/LexicalismSuperfluous4.pdf and treats it as the published version. That file is a draft dated 20 October 2016 (“comments welcome”), not the published text. The published version is the MUSE PDF. The MUSE PDF is 43 pp. because of a cover page. The article is 42 pp. (pp. 1–42), references included.
• Bruening does not build on Haspelmath. The seed says Bruening is “the theoretical-syntax side of c61’s typological argument”. Footnote 2 (p. 1) cites Haspelmath 2011 for the point that no universal definition of the word exists, then explicitly sets it aside: “even if we grant the lexicalist assumption that it is possible to identify words, the lexicalist position is both wrong and superfluous.” The two arguments are independent. Bruening’s does not depend on Haspelmath’s.
• The replies are confirmed, but only in part. The seed marks them “apparently”. The published reply is verified: Müller (2018), “The end of lexicalism as we know it?”, Language 94(1):e54–e66, in the REPLIES section. Its abstract disputes three of Bruening’s arguments: that X0 cannot be extracted, that coordination is always phrasal, and the phrasal analysis of resultatives. It concludes that “it may be reasonable to assume that there is a morphology that is indeed different from syntax”. I did not look for any rejoinder by Bruening, so whether one exists is unverified.
• §4 is not all against Williams and Chomsky. §§4.1–4.3 answer Williams (2007): locality of self- vs phrasal himself, adjuncts under re- vs again, and headedness. §4.4 answers Chomsky (1970) and Newmeyer (2009) on idiosyncrasy.
• Footnote 28 is narrow. It says a referee notes that German how behaves differently, and “I leave exploration of other languages to future research”. It is about one morpheme in §5.5, not a general deferral. Still, the paper’s evidence is overwhelmingly English (see Limitations).
• Word is not recast as nothing. The seed’s “no privileged word level” is right, but the paper makes a specific substitution. It defines X0 formally ((61)–(62), p. 25) as a segment of a node that includes only one bundle of morphosyntactic features. It adds that “It is probable that the notion of a ‘word’ has no meaning other than the prosodic one” (p. 26).