2.1 Three Arguments Supporting the Lexical Wordhood of Wh-Compound Questions

There are three arguments showing that wh-expressions like the boxed ones in (1–3) indeed form compounds. Firstly, it is well-known that in Japanese, the process of sequential voicing, or rendaku, voices the initial obstruent of the second member of a compound. This morphophonological process is illustrated in (5a, b).

In (5a), the initial voiceless consonant of the noun kasa ‘umbrella’, [k], changes to its voiced counterpart, [g], when it is compounded with another noun ame ‘rain’ to yield amagasa ‘umbrella’. The same process is responsible for the change from [s] to [z] in nekozita ‘sensitive tongue’ in (5b), which results from compounding neko ‘cat’ and sita ‘tongue’. Importantly for our present purposes, what we have dubbed wh-compounds may also undergo sequential voicing, as shown in (6a, b). Here, the initial voiceless consonant [k] of karami ‘related’ and kayu ‘porridge’ changes to its voiced counterpart, [g], when they are compounded with wh-words, doko ‘where’ and nani ‘what’. Note that the phrasal combinations that are closest approximations to the two compounds, shown in (6a’) and (6b’), do not undergo sequential voicing.

Secondly, what we call wh-compounds are subject to the same accent rule as bona fide compounds in Japanese. Kubozono (1995) observes that compounds abide by the Compound Accent Rule defined in (7).

Let us see how this rule works, using (8a, b) for illustration.

In (8a), we have the noun phrase sararíiman-no síntoo ‘a new party for office workers’ whose phrasehood is diagnosed by the presence of the genitive case marker no. The phrase in question has phrase accent in the sense that the lexical accent nuclei of the two input nouns are maintained in the noun phrase. In (8b), by contrast, we have sarariimansíntoo ‘office-workers’ New Party’ without the genitive case marking. This expression exhibits compound accent in the sense that it has only one accent nucleus, an accent pattern suggestive of the application of the Compound Accent Rule in (7).

With this accentual difference between phrases and compounds in mind, consider now examples in (9a, b).

(9a) illustrates the phrasal accent pattern of dóko-kara-no zyóohoo ‘information from where’, which retains the original accent nuclei of the two input nouns. This pattern is to be contrasted with the compound accent pattern of doko-zyóohoo ‘where-from information’, which behaves on a par with (8b), not with (8a), with respect to accent placement; it only has one accent nucleus on the second member. This observation lends further support to the conclusion that what we have deemed wh-compounds are indeed compounds and hence form a word-level unit in the lexicalist sense.

Finally, wh-compounds obey the ban on word-internal modification. That is, compounds in general do not accept modification into any component part, unlike phrases. To illustrate, consider examples in (10a, b).

In (10a), the nominal compound aozyasin ‘blueprint’ consists of aoi ‘blue’ and syasin ‘photo’. Its compoundhood is independently confirmed by the presence of sequential voicing, which changes [s] of the head noun to [z]. The ungrammaticality of (10a) shows that the degree adverb hanbun ‘half’ cannot modify the adjectival member of the compound alone. This ban on compound-internal modification is not observed with phrases, however, as shown by the grammaticality of (10b), where the same adverb can modify the adjectival aoi ‘blue’ contained within the NP aoi syasin ‘a blue photo’.

Now, the ill-formedness of the examples in (11a–c) indicates that wh-compounds, such as dare-toku ‘who-benefiting’, nani-nabe ‘what-porridge’, and doko-zyoohoo ‘where-from information’, all exhibit the ban on word-internal modification, for no degree modifier may intervene between the two members of the relevant compounds.¹

It is clear from the above that wh-N expressions in the examples introduced so far constitute a subspecies of genuine compounds in Japanese grammar and hence a word-level unit from the standpoint of the lexicalist theory of the syntax-morphology interface, according to which compounds are formed in the pre-syntactic lexical component and enter the syntactic derivation as an unanalyzable atomic operand. In the next section, however, we will introduce novel data pointing to the opposite conclusion that wh-compounds do allow certain syntactic operations to have access to their internal structure.

2.2 Three Arguments Supporting the Internal Syntactic Accessibility of Wh-Compounds

We will now present three arguments showing that the internal structure of wh-compounds is accessible to certain syntactic processes and constraints, contrary to the conclusion reached in section 2.1.

2.3 The Janus-Faced Profile of Wh-Compound Questions and the Ordering Paradox

Let us take stock of our findings thus far. On one hand, we have presented data on sequential voicing, the Compound Accent Rule and the impossibility of word-internal modification to show that wh-compounds are to be recognized as a real word-level unit in the Lexicon according to the central tenets of the LH. On the other hand, we have shown that this type of compound not only gives rise to a genuine wh-question licensed by the interrogative C head, yielding partial fragment answers targeting the wh-part of the compound alone, but also exhibits robust movement-sensitive restrictions such as island effects, intervention effects and additional wh-effects (albeit directly influenced by distinct prosodies affecting the varying range of acceptability of the relevant cases).

It is clear now that wh-compound questions simultaneously exhibit lexical wordhood and syntactic accessibility. This mutually conflicting mixture of the various properties of this type of question poses a serious challenge to the lexicalist conception of the syntax-morphology interface. To see theoretical implications of the above findings more clearly, let us consider below a sample of some representative definitions of the Lexical Integrity Principle proposed in the lexicalist literature.

Given what one may call the ‘feed-forward’ view of the morphology-syntax interface within which the LH is traditionally defined, the acceptability of partial fragment answers as illustrated in (12A) and (13A) would be unaccounted for in a lexicalist theory, for a wh-compound, being a word-level unit in the lexicalist sense, should be atomic for syntax, a point clearly shared in the three definitions of the Lexical Integrity Principle above. For the same reason, the left member of a wh-compound with its interrogative/existential/universal/negative polarity reading licensed by its local quantificational particles should not be able to serve as input for compounding, for this word formation is a lexical process in the pre-syntactic lexicon, but our findings above indicate otherwise. For these reasons, wh-compounds raise a real architectural ordering paradox for any version of the LH-driven conception of the syntax-morphology interface.

Note that the paradox occurs because the Lexicon is postulated in the lexicalist framework as an independent pre-syntactic module specifically dedicated to certain types of word formation such as compounding, thereby maintaining some version of the LH as a trafficking condition on the syntax-morphology connection. Capitalizing on this point, in the next section, we will seek for an alternative analysis of wh-compound questions within the DM framework, which attempts to dispense with the lexicalist Lexicon as an autonomous module feeding the computational component of syntax.

3.1 Sato’s (2010) DM-Based Analysis of Phrasal Compounds based on Renumeration

Sato (2010) proposes a syntactic analysis of phrasal compounds in line with the basic tenets of the DM framework which draws on a combination of Johnson’s (2004) conception of renumeration with Uriagereka’s (1999) multiple spell-out model of syntax. Sato proposes that a spelled-out structure is returned to the syntactic derivational workspace as a derived simplex lexical item or “giant lexical compound” (see section 3.2 for a more detailed exposition of Uriagereka’s model). To see how his analysis works, consider his DM-style derivation of the phrasal compound, an I-drank-too-much headache, depicted in (34a–c).

In (34a), the CP is assembled by successive applications of Merge (and Move). The CP so assembled then undergoes early spell-out and is returned to the numeration as a syntactic treelet, as show in (34b), following Johnson’s (2004) theory of renumeration. We assume here that the Spell-Out operation in (34b) is triggered by the C phase head, as standardly assumed (Chomsky 2000; 2001; 2004; 2008).⁶ The renumerated item is now plugged into the syntactic workspace as a derived lexical item – call it α – as depicted in (34c). The resulting syntactic object merges with N and D in that order to yield the phrasal compound, as desired, without causing any ordering paradox. It is crucial to note that this non-lexicalist ‘single-engine’ analysis allows for a unified treatment of phrasal compounds and regular nominal compounds such as a nurse shoe, as indicated in its syntactic derivation in (35).

Before closing this section, let us address a potential mechanical issue regarding our proposed approach to phrasal compounding. An anonymous reviewer observes that renumeration, as it is, is such a powerful mechanism that it has the risk of potential overgeneration. For example, the reviewer notes that examples like I had a [This-is-a-damn-fucking-end-of-the-world][I-drank-too-much] headache] would be freely generated by our system. As stated at the beginning of section 3, a phrasal compound is subject to regular syntactic rules in the sense that its non-head member itself must be a well-formed expression. The two clausal modifiers in the example above are grammatical expressions. We suspect that the example itself is grammatical, but is felt unacceptable due to independent extra-grammatical constraints imposed on phrasal compounds. One such constraint is that the non-head member of a phrasal compound typically has a quotative flavor (Wiese 1996; Harley 2011). This constraint presumably serves to ensure that only one quotative CP can be combined with a single compound head noun such as headache. This point is evidenced by the fact that the felt unacceptability of the original example can be removed in the modified version, I had a [This-is-a-damn-fucking-end-of-the-world-so-I-drank-too-much] headache], where the head noun is modified by a single clausal expression used as a quotative phrase. We will further show later in this paper that multiple spell-out and renumeration are triggered by a phasal unit – CP, vP and DP. For these reasons, we contend that potential cases of overgeneration created by renumeration are to be excluded/disfavored by independently known syntactic conditions (e.g., phasehood of non-head elements of a phrasal compound) as well as extra-syntactic semantic/pragmatic restrictions (e.g., the quotative nature of those elements).

3.2 Renumerating Wh-Compound Questions in Japanese

Having outlined the DM-based solution to the ordering paradox raised by phrasal compounds, we are now in a position to develop our analysis of wh-compound question formation in Japanese. We will show how our analysis can successfully accommodate syntactic accessibility/penetrability of wh-compound questions through regular syntactic processes while at the same time deriving their “lexical” effects through early spell-out and renumeration.

Consider the step-by-step derivation, shown in (37a–d), of the phrasal compound nani-nabe ‘what-hotpot’ in (13), repeated here as (36).

The DP nani-nabe ‘what-hotpot’ is assembled first, as shown in (37a). Suppose, following Kageyama (1993: 337, 338), that the [q] feature optionally undergoes percolation from the original D head onto its dominating DP, as indicated in (37b). The DP structure is subsequently spelled-out early and renumerated, as depicted in (37c). In this example, the entire wh-compound is nominal. Thus, it is reasonable to assume that it forms a DP phase (Svenonius 2004; Chomsky 2008) and undergoes early spell-out and renumeration for this reason. After this renumeration process, the syntactic treelet is returned to the derivational workspace as a derived terminal node α with the [q] feature and successively merged with √tabe, v, T, and C. This step is followed by q-agreement between α and the interrogative C. This series of derivational steps thus yields the regular wh-interrogative reading for (36Q).⁷

As shown in (36A), it is possible to omit part of the compound as a grammatical reply to the wh-compound question. As noted by an anonymous reviewer, it is expected that early spell-out and renumeration should make the whole compound atomic and indivisible with the wrong outcome that the truncated response just answering the non-head wh-part of the compound should be inaccessible. We agree with the reviewer that our system only allows the full-fledged answer kimuti-nabe ‘Kimchi hotpot’ as part of the renumeration/atomization process. We maintain that the shorter answer is derived through left-peripheral string deletion, a PF-deletion process independently developed for Japanese/Korean by Mukai (2003), An (2016; 2019) and Sato and Maeda (2018; 2019) (see also Weir 2012) which deletes a non-constituent part of an XP under the condition of string identity: we will come back to a detailed discussion of this analysis when we compare it with Kimura and Narita’s (2021; 2023) analysis in section 4. According to this analysis, the partial answer is derived as shown in (38).

Note that, as noted by Nishigauchi (1986; 1990), some such ellipsis process seems independently necessary to derive the shorter answer form in (39A2) from the basic response in (39A1) to the complex NP wh-question in (39Q), which Nishigauchi analyzes as involving large-scale pied-piping preceded by percolation of the q-feature from the wh-word nani ‘what’ to the nominal layer dominating the relative clause.

A defender of the lexicalist theory might say that a similar analysis like ours could be replicated in a pre-syntactic lexical component if certain properties of the components of a wh-compound could be inherited by the derived word through mechanisms like percolation. Such an analysis would work for simplex wh-compounds as in (36Q) whose constituent part consists of a word-level unit such as nani-nabe ‘what-hotpot’, but it would not work for complex wh-compounds which involve what is clearly the product of the syntactic derivation and hence cannot be analyzed as a word-level lexical expression. The expression marked by the wavy line in (40) is a case in point, for this structure involves not only a relative clause structure but also conjunction indicated by the continuative copula de. Multiple wh-compound questions as shown in (3Q), repeated here as (41), illustrate the same point.⁸

Our current percolation-based system can be extended to account for multiple wh-compound questions like (40) and (41) as well. Consider the relevant part of the syntactic derivations for the complex wh-compound in (40), shown in (42).⁹

In this derivation, the whole compound headed by the noun densha ‘train’ takes the coordination structure as its lefthand member, which in turn is composed of two wh-compounds, nani-eki hatu ‘leaving what-station’ and nani-eki iki ‘bound for what-station’. The [q]-features originate in the respective interrogative D-heads and are pumped up successively from the heads all the way up to the upmost DP headed by densha ‘train’, as indicated in (42). The DP undergoes early spell-out and is returned to the main derivational workspace as a derived lexical item with the q-feature as its topmost label. The multiple wh-interrogative reading is correctly obtained from this derivation when this item enters into q-agreement with the interrogative C head no. The reader can verify that a similar analysis holds true for the triple wh-compound shown in (41) via successive q-feature percolation.

3.3 The Janus-Faced Nature of Wh-Compound Questions

Let us now see how the hybrid nature of wh-compound question formation can be accommodated in our present system. On one hand, the cluster of syntactic properties noted in section 2.2 manifest themselves because this type of question is derived in the syntax in accordance with regular syntactic principles that also regulate regular wh-question formation, including island/intervention/additional wh-effects.

On the other hand, the ‘lexical’ or “word-like” properties associated with this type of question noted in section 2.1 can be derived as a side effect of the derivational step depicted in (37c) where the compound constituents undergo spell-out mid-derivationally and renumeration before they are returned to the computational workspace as a simple derived item. The set of syntactic objects generated this way, we argue, corresponds to the lexical notion of ‘word’. Let us elaborate on this hypothesis below as it is directly relevant to our position on how to derive word-level effects from syntactic objects.

Within Uriagereka’s (1999) conservative version of the multiple spell-out system, once a complex syntactic object is spelled-out, the internal structure of the object is destroyed with its terminal nodes already linearized, attaining the status akin to a ‘giant lexical compound’ by the time when it re-enters the syntactic derivation. The derivation thus manipulates it not as an internally complex object but as an atom, though its syntactic terms are interpretable at the semantic/phonological components.¹⁰ To quote Uriagereka himself:

The conservative proposal is based on the fact that the collapsed Merge structure is no longer phrasal, after Spell-Out; in essence, the phrase marker that has undergone Spell-Out is like a giant lexical compound, whose syntactic terms are obviously interpretable but are not accessible to movement, ellipsis, and so forth…

In the conservative version, the spelled-out phrase marker behaves like a word, so that it can associate with the rest of the structure; this means it must keep its label after Spell-Out. Technically, if a phrase marker {α, {L, K}} collapses through Spell-Out, the result is {α, <L, K>}, which is mathematically equivalent to {α, {{L}, {L, K}}}. Since this object is not a syntactic object, it clearly can behave as a “frozen” compound. As a consequence, we need not add any further stipulations: the collapsing procedure of Spell-Out itself results in something akin to a word.

(Uriagereka 1999: 256–257)

Our hypothesis is that the set of atoms derived through this spell-out + renumeration procedure form candidate input for a speaker’s mental lexicon, a storage point that is unavoidable to postulate in some way regardless of whether we adopt some version of the lexicalist or non-lexicalist position of the syntax-morphology interface. Accordingly, all so-called ‘lexical’ effects associated with “word” observed in section 2 (sequential voicing, compound accent and the ban on word-internal modification: see section 3.4 for our clarification on the last profile) may be reanalyzed as the results of post-syntactic phonological processes applying to the renumerated atoms at the PF interface. This hypothesis is further substantiated by the observation that ‘lexical’ effects are always morphophonological; there is no such effect which would bring about changes in, say, word order or syntactic transformation, at the syntactic level. Take sequential voicing, for instance. Our current hypothesis leads to the view that this is a post-syntactic process that can apply to a spelled-out/renumerated items. Indeed, this view seems plausible, for a native Japanese speaker can correctly apply sequential voicing to the “off-the-cuff” type of wh-question compounds as in (43a, b) on the fly (see Kawahara 2012 for a relevant discussion on sequential voicing on nonce compounds), subject to known constrains on this morphophonological process, including Lyman’s Law, which states that it is blocked by a voiced construent within the second element of a compound, as shown in (44a, b).¹¹

More broadly, one implication of our proposal is that it launches a new DM-compatible system to derive ‘wordhood’ using technology of minimalist syntax without invoking the lexicalist notion of ‘word’. There is no denying that there are some non-eliminable morphological/phonological processes such as sequential voicing, compound accent, but at least we do not need a lexical component dedicated for forming words in the lexicalist sense. This point cannot be emphasized enough, for, to the best of our knowledge, the issue how ‘wordhood’ arises in such a framework like the DM, which otherwise aims to dispense with an autonomous module for word formation, has not received attention in the DM literature (see also section 4 for a related discussion on this point, where we review Kimura and Narita’s (2023) recent DM-analysis of wh-compound questions). According to our system, ‘word’ in a language L may be reconceptualized partially in terms of “possible spell-out domains” or phase units in L whereas ‘wordhood’ in L is an epiphenomenal consequence of certain characteristic morphophonological rules applying to renumerated items in L on a language-particular basis.

3.4 The Ban on Word-Internal Modification Revisited

In this section, we address one remaining question with our analysis of wh-compound question. Recall from section 2.1 that we have observed that wh-compounds obey the ban on word-internal modification, as illustrated in (11a–c), repeated here as (45a–c).

The question facing us now, then, is how our theory should block such examples. For instance, nothing in our system appears to block the sequence of syntactic derivation where the noun nabe ‘hotpot’ merges with the adjective oo ‘big’ and the resulting syntactic object, in turn, merges with the interrogative D head nani ‘what’. This sequence should yield nani-oo-nabe ‘what-big.hotpot’, a result that seems unacceptable as shown in (45b), as reported by Harada (2014).

We actually do not think that there is any need to block such examples from being freely generated in the present DM model, however. In fact, we venture that such examples are completely grammatical and that their alleged ban on word-internal modification are explained away by independent extra-grammatical encyclopedic knowledge of the roots involved in wh-compound formation. For example, nabe itself is ambiguous between a pan (a cooking utensil) and a cuisine (a type of food served), but once it is modified by some scaler adjectives such as ookii ‘big’, it may only yield the former reading because of its idiosyncratic selectional restriction on the type of its modifiees: a pan may or may not be big, but a cuisine certainly cannot. This observation is illustrated in (46a).

It follows then that merging nani ‘what’ and oo-nabe ‘big pot’ as a wh-compound and asking for the identity of the kind of cuisine results in semantic anomaly, as shown in (46b), though the compound itself is grammatical as far as its syntactic derivation goes.

Our position that the perceived unacceptability of (45b)/(46b) is due to encyclopedic knowledge of the lexically ambiguous root √nabe is further supported by the following observation. In a restricted range of cultural contexts, this root can actually be used exceptionally to specify the name of a local cuisine as long as the cuisine is widely known to be served using a large-sized pan. One such case is shown in (47), where oo-nabe ‘big pot’ combines with imoni ‘taro and meat soup’ to yield imoni-oo-nabe ‘[taro-and-meat-soup]-big-hotpot’.

A similar characterization applies to (45a) and (45c). Consider (48a, b).

(48a) illustrates a wh-compound headed by zyoohoo ‘information’ separated from a wh-word by an intervening modifier, soosu ‘source’. We did not manage to find a wh-compound akin to (45a), but there are some attested examples as in (48b), where the two members of the compound, kappuru ‘couple’ and toku ‘benefit’, are disrupted by a degree modifier, mettya ‘exceedingly’, but the result is completely acceptable.

4.1 The Adequacy of Generalization (51) and the Nature of In-Situ Deletion

The first issue concerns the empirical robustness of the generalization in (51) and its implications for the exact nature of the in-situ deletion process involved in deriving a fragment answer to a wh-compound question. Example (53) from Kimura and Narita (2021) shows that the wh-compound question nani-ya ‘what-store’ cannot be answered with hana ‘flower’ without the accompanying morpheme -ya ‘store’.

This pattern runs counter to (51) because the felicity of the fragment answer does not correlate with the availability of its full-fledged base compound. Kimura and Narita (2021:197) qualify regarding (51) that X cannot stand alone as a fragment answer when Y is a bound morpheme presumably because Y is morphologically too dependent on X so that focus marking on X must percolate onto the entire compound. Note, however, that the building blocks of compounds, being a bound root, are always bound morphemes by definition. Thus, konomi/gonomi ‘taste’ is treated as a bound root when positioned in a compound structure in the same way that ya ‘shop’ is. Thus, this bound-morpheme proviso cannot correctly capture the contrast between (49A1) and (53A) with respect to the grammaticality of the fragment answer option.

The same constraint is also called into question by the impossibility of the fragment answer in (54A2).

The question in (54Q) involves the wh-compound nani-zyoozu ‘good at-what’; its compound status can be verified by the fact that it exhibits a compound stress pattern with a single accent pitch, as opposed to two lexical accents. The examples in (54A1, 54A2) show that the fragment response home ‘praising’ is ill-formed even though the full-fledged compound answer, home-zyoozu ‘good at-praising’, is well-formed. Again, one might argue that zyoozu in a compound structure is a bound morpheme, but we counter that konomi/gonomi must be one by the same token.

Suppose, then, for the sake of argument that there is some well-defined criterion of “bound morpheme” that we can use in non-compound environments by which morphemes such as ya ‘shop’ and zyoozu ‘good at’ are bound whereas morphemes such as konomi ‘taste’ are not. Indeed, the former cannot stand on their own (e.g., *dare-no ya ‘intended: whose shop’) but the latter, in principle, can elsewhere (e.g., dare-no konomi ‘whose taste’). However, this view of boundedness is doomed to failure, considering the ungrammaticality of the fragment answer in (55A2), which involves daigaku ‘university’, a free morpheme in a non-compound context (e.g., dare-no daigaku ‘whose university’).¹³

These considerations suggest that the correspondence between the (un-)availability of a fragment answer and that of its full-fledged variant to a wh-compound question is not as robust as claimed by (51). This result, in turn, undermines Kimura and Narita’s version of the in-situ deletion analysis based on focus marking which is based on the generalization.

We hypothesize instead that the availability of the fragment answer X based on the [X-Y] compound is better predicted by whether X can grammatically occur in a gapped clause. Of significance here is that the acceptability/relative degree of the acceptability of the fragment answer option in (49A1), (53A), (54A2) and (55A2) corresponds to that of the surviving part of the compound after gapping has taken place in the initial clause in (56A), (57A), (58A) and (59A), in that order. For instance, the fragment answer option in (49A1) is grammatical when the non-head member of the compound can survive gapping as in (56A). Conversely, the fragment answer option in (54A2) is degraded when the non-head member of the compound cannot survive gapping as in (58A).

The new gapping/fragment answer correspondence above is exactly what we predict if the deletion process responsible for the short reply to a compound question is string-deletion, a process independently motivated for Japanese/Korean gapping (Mukai 2003; An 2016; 2019; Sato and Maeda 2018; 2019). String Deletion is a PF deletion operation that applies to a contiguous phonetic string in a phonological representation under identity with another occurrence of the same string. According to this analysis, the aforementioned correspondence is obtained because this process is involved in the generation of the fragment in both gapping and fragment answer contexts, as schematically represented in (60a, b) for the examples in (54A2) and (58A).

Our analysis highlights two discoveries regarding the nature of the in-situ deletion process involved in a fragment answer to a wh-compound question. Firstly, the deletion process involved is string-deletion. We believe this is an improvement, for Kimura and Narita (2021) do not make explicit what the deletion process is, except mentioning that it targets all given materials except for a focus-marked phrase within an ellipsis site; the same question remains even with their updated view of ellipsis as no-vocabulary-insertion within the DM model in Kimura and Narita (2023).¹⁴ Secondly, this new perspective, in turn, calls into question the validity of (51), which attempts to establish the link between the felicity of the fragment answer and that of the compound structure. We have argued that the right correlation must be sought instead between the availability of the fragment answer and the survivability of the non-head member of a compound in a gapped clause, that is, the clause derived through String Deletion.

4.2 The Origin of Lexicality/Wordhood within the DM Model

The second issue to be explored here related to comparison of our analysis with Kimura and Narita’s (2021; 2023) pertains to lexicality/wordhood and its origin. Recall that Kimura and Narita attribute the ungrammaticality of (50A2), repeated here as (61A2), to a lexical integrity effect.

However, one of the foundational assumptions of the DM framework is that there is no lexical component specifically dedicated for word formation processes such as compounding, to start with. In other words, to the extent that they adhere to the basic precepts of the non-lexicalist theory, there seems to be no way to block the generation of the answer form in (61A2) as well as (49A2), which is derived through it. This consideration indicates that there is actually no way to guarantee (51) (though it must be recalled that the generalization is inaccurate on empirical grounds as well, as we have already shown in the previous subsection). In fact, quite the opposite holds true; the phrasal compound structure as in (61A2) is actually grammatical, and the perceived ill-formedness is caused by some extraneous variable. Take the compound X-gonomi ‘X’s taste’, for instance. As its encyclopedic entry, this compound requires X to denote people with some salient individual level/kind denotation, such as hitori-musume ‘only daughter’, hahaoya ‘mother’, kodomo ‘child’, and dokusindansee ‘single man’. Thus, (61A2) can be rejected as easily by the non-kind denotation of the floating quantifier configuration underlying musume hitori independently of its underlying syntactic structure. Indeed, if we use a different compound head such as kurasi ‘living’, which does not impose this usage restriction, we can easily find acceptable cases of phrasal compounds as in (62a, b).

Kimura and Narita (2021:197, 2023:190) state that the phrasal compound structure as in (49A2)/(61A2) is ill-formed because of the lexical integrity effect or the opaqueness of the word unit to phrase-level syntactic derivations including quantifier floating. However, it is to be emphasized that such a statement is informulable in the DM framework, to begin with, which proposes to dispense with any pre-syntactic module responsible for word formations independent from phrases/sentences. There is no way to yield lexicality/wordhood in this single engine model.

By contrast, as we have argued in section 3.1., our proposal analysis, based as it is on renumeration and spell-out, advances a specific vision of the very origin of lexicality/wordhood by incorporating Uriagereka’s (1999) hypothesis that an internally complex phrase structure is periodically flattened for atomization before it is returned to the main derivational cascade as a derived terminal item. The second contribution of our analysis thus lies in our distinct perspective on the very origin of lexicality/wordhood. This point remains unaddressed in Kimura and Narita, which, on one hand, maintains a kind of the lexicalist stance on compound formation (to block the allegedly ungrammatical examples as in (61A2) as well as (49A2)) while, on the other hand, adhering to the DM view of the syntax-morphology elsewhere (to let ellipsis/deletion apply to an internal constituent of a compound). Relatedly, our perspective is important in that it leads one to question the robustness of their empirical generalization in (51).