2.1 A classifier language with D

Nuosu Yi, a Sino-Tibetan language spoken in the Liangshan Yi Autonomous Prefecture of Sichuan, China, is typologically unique. Unlike most classifier languages, such as Mandarin Chinese, Japanese and Korean, Nuosu Yi features an overt definite determiner, i.e., su³³ (Jiang 2012; 2018; 2020). In Nuosu Yi DP constructions, su³³ linearizes after the obligatory classifier, turning an indefinite DP (i.e., N-Clf, (4a)) into a definite one (i.e., N-Clf-su³³, (4b)).³

Importantly, su³³ is not a demonstrative as Nuosu Yi has the proximal demonstrative cy⁴⁴ and the distal demonstrative a³³zzy⁴⁴, which occupy distinct positions from su³³ (4c). Specifically, the demonstrative precedes the classifier whereas the definite determiner su³³ follows it. Furthermore, both su³³ and the demonstratives cannot combine directly with the head noun, and the absence of a classifier leads to ungrammaticality, as indicated in (4b) and (4c).

Additionally, demonstratives and su³³ cannot co-occur within the same nominal domain, see (5).⁴

Furthermore, I present two diagnostics providing evidence that su³³ is a definite determiner. The first evidence is from the consistency test (Löbner 1985; Dayal 2004), which is used to distinguish demonstratives from definite determiners (though see Moroney 2019 for potential challenges to this diagnostic). As seen in (6), the definite determiner su³³ leads to a contradiction (6a) while the demonstrative cy⁴⁴/a³³zzy⁴⁴ does not (6b).

Another diagnostic for definiteness comes from Milsark’s (1974) Definiteness Restriction, which states that definite expressions are not allowed in existential sentences. Indeed, the definite DP [N-Clf-su³³] is excluded from the existential jjo³³ ‘have’ construction in Nuosu Yi whereas the indefinite DP [N-Clf] is not, as exemplified in (7).

Given these two diagnostics, I take su³³ to be the definite determiner in Nuosu Yi (to be revised).

On the other hand, similar to typical classifier languages, bare nouns in Nuosu Yi are ambiguous between singular and plural. This is demonstrated in (8a), where the head noun vo⁵⁵ ‘pig’ does not change its form regardless of whether the numeral is singular or plural.

In the definite DP construction, when the numeral is cy²¹ ‘one’, it must be absent from the DP construction (8b). By contrast, when the numeral is larger than one, there is no such restriction (8c).

In addition, the classifier and the definite determiner form an inseparable bundle and no element can be inserted between them, see (9a) (cf. (9b)).⁵

To summarize, Nuosu Yi is typologically rare in that it has both obligatory classifiers and a definite determiner, namely, su³³. Moreover, the definite determiner immediately follows a classifier and nothing can be inserted between them.

2.2 Definite [N-Clf] phrase and tone change

Nuosu Yi is a tonal language with four tones. There are three lexical tones (i.e., /55/ (H), /33/ (M), /21/ (L)), along with a sandhi tone (i.e., /44/ (MH)) (Lama 1991; Edmondson et al. 2017; Lee & Shen 2024). The minimal pairs shown in Table 1 demonstrate the contrastive nature of these four tones.

Lama (1991) proposes that the sandhi tone 44 results from a long-term dissimilatory process involving the 33 and 21 tones in natural speech. Moreover, it has been proposed that the sandhi tone 44 is derived from the 33 tone or the 21 tone in syllable combination (Edmondson et al. 2017). Generally, if the 44 tone appears as the first syllable of a two-syllable word or phrase, it is likely that this sandhi tone originally came from a 33 tone, as seen in zza³³ ‘food’ + zze³³ ‘to eat’ → zza⁴⁴ zze³³ ‘to eat food’. Conversely, if the 44 tone appears on the second syllable of a two-syllable word or phrase, it might originate from a 21 tone, as in the example jjy³³ ‘recp’ + ndu²¹ ‘hit’ → jjy³³ ndu⁴⁴ ‘beat each other’.

However, in a few cases, the sandhi tone 44 also holds the status of a contrastive tone as in xi⁴⁴ ‘what’, lo⁴⁴ ‘after’, and jju⁴⁴ ‘nmlz’ (shown in the fourth column of Table 1), all of which are monosyllabic functional morphemes.

Crucially, Lama (1991) claims that the 44 tone largely has grammatical use and observes that when a 33 or 21 tone classifier takes the 44 tone, it can denote definiteness on its own, as demonstrated in (10) and (11).

I gloss the definite use of ma⁴⁴ and zzi⁴⁴ as ‘clf:def’, representing ‘definite classifier’. Specifically, the underlyingly 33 tone classifier ma³³ and the underlyingly 21 tone classifier zzi²¹ adopt the 44 tone, leading to a definite interpretation in the absence of su³³. As can be seen in the translations above, the indefinite reading is impossible in these cases.

The tone change shown in (10) and (11) will be discussed in more detail in Section 5: there, I will provide evidence that it is not a case of tone sandhi, as (i) it occurs only in specific grammatical contexts—namely, definite contexts; and (ii) the tone change from 21 to 44, as shown in (11), is not phonologically general. Instead, I argue that this is best analyzed as an instance of grammatical tone, where the tone change on the classifier results from definite tone docking. The point in this section is simply that classifiers with 21 and 33 tones shift to 44 to mark definiteness.

When the classifier has an underlying 55 tone, however, the classifier cannot take the 44 tone and the absence of su³³ only leads to an indefinite reading, as illustrated in (12).⁶

Furthermore, it is well-known that definiteness can be divided into anaphoric and unique definiteness (Schwarz 2009; 2013). The former type of definites requires previous linguistic mention (i.e., discourse anaphora) in order to refer, while the latter does not. To further investigate Nuosu Yi’s definite [N-Clf] pattern, I observe that it is also attested in anaphoric and unique definite contexts, as illustrated in (13) and (14).

In the anaphoric definite example (13), the definite expression si³³hni³³ ma⁴⁴ ‘girl clf:def’ in (13b) refers to the very girl that entered the classroom with a boy in the antecedent sentence (13a). In the unique definite example (14), hlo²¹bbo²¹ ma⁴⁴ ‘moon clf:def’ is understood as referring to the unique entity that has the relevant property of moon in the world. Additionally, the definite [N-Clf] construction can occur in both subject (13) and object (14) positions.⁷

Moreover, this definite construction is also observed in donkey anaphora, producer-product bridging and part-whole bridging contexts.⁸ The relevant examples are presented in (15)–(17). Note that it is impossible to replace the 44 tone of the definite classifier with its original tone in these contexts.

A final note is that the definite [Clf-N] phrase is only used in informal speech, while in formal speech, su³³ is obligatorily used. Example (18) is a news title from Yizu Wang ‘Yi People’s Daily Online’. Although the reference of the subject is clear to all readers—there is only one New Chengdu-Kunming Railway in China—and there is a tone change on the classifier ji⁴⁴, the presence of su³³ is strongly preferred in this formal context.

2.3 Presence of numeral and lack of tone change

In Nuosu Yi, numerals occur between the head noun and the classifier. When a numeral is present, the [N-Num-Clf] phrase is exclusively interpreted as indefinite (19a). However, when this indefinite [N-Num-Clf] phrase is combined with the definite determiner su³³ or a demonstrative, the whole expression clearly becomes definite, as shown in (19b) and (19c). Additionally, there is a difference in word order between the two cases: [N-Num-Clf-D] vs. [N-Dem-Num-Clf].

It is worth noting that in the presence of a numeral in the DP construction, the tone change on the classifier does not occur, as illustrated in (19b). Specifically, the classifier ma³³ keeps its original 33 tone.

Moreover, the classifier cannot take the 44 tone, regardless of whether su³³ is present or not, see (20).

Specifically, in (20a) su³³ is present while in (20b) su³³ is absent. The key generalization is that when a numeral is present, the classifier cannot take the 44 tone, regardless of the existence or presence of su³³ (cf. (19b)).

In summary, although definiteness can be overtly marked on classifiers via tone changes in Nuosu Yi, the presence of a numeral always blocks the definite form of the classifier. That is to say, in the definite [N-Num-Clf-D] construction, tone changes on the classifiers are never allowed. The co-occurrence of definite [N-Clf] and indefinite [N-Num-Clf] constructions is not unusual cross-linguistically. In the next section, I will present similar phenomena in other classifier languages, which can be effectively accounted for under a Clf-to-D head movement analysis.

3.1 Definite bare classifier phrase as a result of Clf-to-D head movement

The Clf-to-D head movement analysis was first proposed by Simpson (2005) to explain definite bare classifier phrases in several Southeast Asian languages and later expanded by Wu & Bodomo (2009) to account for similar phenomena in Cantonese. The central claim is that a determiner phrase (DP) projection exists not only in languages with overt definite determiners, such as English, but also in languages that lack them, like Mandarin Chinese and Cantonese. Specifically, Simpson observes that bare classifier phrases in Vietnamese, Hmong, and Nung—three article-less classifier languages—are often associated with definiteness. Examples from each language are provided in (21)–(23).

Following Longobardi (1994), Simpson (2005) posits that the head D is the locus of definiteness cross-linguistically. In article-less classifier languages, the definite D is null and must be overtly instantiated by some lexical element to be licensed. As a result, the [+definite] feature on D triggers Clf-to-D head movement, as illustrated in (24).

Another piece of evidence supporting the Clf-to-D head movement analysis comes from the Vietnamese “extra cai construction”. Specifically, Vietnamese allows a second general classifier, cai, to occur before the regular classifier, resulting in a [cai-Clf-N] sequence with a clear definite interpretation, as shown in (25).

(25) can be naturally accounted for under the Clf-to-D head movement framework, where the general classifier cai occupies the head D position, leading to definiteness, while the regular classifier con remains in situ.

Building on Simpson’s analysis, Wu & Bodomo (2009) extend the Clf-to-D head movement approach to account for definite [Clf-N] phrases in Cantonese, another classifier language that lacks overt definite determiners. An example of a definite bare classifier phrase in Cantonese is provided in (26).

Wu & Bodomo argue that the null D triggers Clf-to-D head movement, leading the classifier to combine with D and license the null definite D. Note that the definite [Clf-N] construction can also appear in object position, see (27).

Wu & Bodomo further point out that Clf-to-D head movement is not obligatory in Cantonese, as [Clf-N] can be interpreted as either definite or indefinite depending on the context. For example, (27) can also receive an indefinite interpretation, i.e., the bare classifier phrase gaa ce ‘clf car’ can mean ‘a car’, leading to the interpretation ‘He sold a car’ (Wu & Bodomo 2009: 497). In such cases, Clf-to-D head movement does not occur, and the classifier remains in situ, resulting in an indefinite reading as the head D is not licensed.

It is important to note that Clf-to-D head movement is not universally available in all classifier languages. For example, in Mandarin Chinese, bare classifier phrases never receive a definite interpretation and they are only allowed in object positions, as demonstrated in (28).

In summary, while Mandarin Chinese does not allow definite [Clf-N] phrases, Cantonese, Vietnamese, Hmong, and Nung do. This contrast suggests cross-linguistic variation in the availability of Clf-to-D raising.

3.2 Presence of numeral and blocking of Clf-to-D head movement

It has been observed that numerals have the effect of “undoing the definiteness” in classifier languages (Cheng & Sybesma 1999). Specifically, when a numeral is present in a nominal phrase, the entire phrase yields only indefinite interpretations. For instance, Simpson (2005) observes that in Vietnamese, [Num-Clf-N] phrases are consistently interpreted as indefinite, see (29).

To account for the indefinite nature of [Num-Clf-N] constructions, Simpson proposes that a numeral phrase (NumP) is projected above the classifier phrase (ClfP), with numerals occupying the head Num. This structure prevents Clf-to-D head movement, as the classifier cannot raise to D over Num due to the Head Movement Constraint (Travis 1984). The constraint states that a head X⁰ can only move to an immediately governing head Y⁰. Since the head Num intervenes, its presence blocks the classifier from moving to D, as illustrated in (30).

Similar indefinite [Num-Clf-N] patterns are observed in Cantonese (Wu & Bodomo 2009; Hall 2019). One example is shown in (31) where the [Num-Clf-N] phrase is not allowed to occur in the subject position because of a ban on indefinite preverbal subjects.

(31) shows that the [Num-Clf-N] phrase is exclusively interpreted as indefinite in Cantonese. The Head Movement Constraint provides a straightforward explanation for the unavailability of the definite reading in (31). Since the head Num intervenes between Clf and D, Clf-to-D head movement is blocked, resulting in an indefinite interpretation of the entire phrase.

3.3 Tonal marking of definiteness on classifiers

In this section, I will examine Wenzhou Wu, an article-less classifier language that marks definiteness on classifiers via tonal alternations. I will demonstrate that the Wenzhou Wu data can also be accounted for under the framework of Clf-to-D head movement and the Head Movement Constraint.

Building on Cheng & Sybesma (2005), Hall (2019) reports that Wenzhou Wu (Sino-Tibetan, China) uses tone to mark definiteness on classifiers. Wenzhou Wu has eight lexical tones, which can be divided into four groups (A, B, C, and D). Each group can be further divided into two subgroups based on register, high (hi) and low (lo). This is illustrated in Table 2.

In Wenzhou Wu, [Clf-N] phrases can denote both indefinite and definite meanings, with definiteness marked by a tonal change on the classifier to Group D. For indefinite [Clf-N] phrases, the classifier retains its underlying tone. However, when the tone shifts to a D-tone, the [Clf-N] phrase is interpreted as definite. Thus, classifiers with hi-A, hi-B, hi-C, and hi-D tones shift to hi-D to express definiteness, while classifiers with lo-A, lo-B, lo-C, and lo-D surface as lo-D when denoting definiteness. This tonal shift results in a minimal pair, as illustrated in (32), where the only difference lies in the tone of the classifier.

In addition, in Wenzhou Wu, indefinite phrases are prohibited from appearing in preverbal subject positions, similar to the ban observed in Cantonese (see (31)). As such, classifiers with an “indefinite” tone are disallowed in subject position (33a), while classifiers marked with a “definite” D-tone are permitted (33b).

Furthermore, classifiers with an underlying D-tone are ambiguous between indefinite and definite interpretations, as demonstrated in (34).

However, when a numeral is present, the classifier must retain its underlying tone, and the [Num-Clf-N] phrase is exclusively interpreted as indefinite, as shown in (35).

To summarize, in Wenzhou Wu, the marking of definiteness on classifiers is blocked by the presence of a numeral. While [Clf-N] phrases can have both definite and indefinite interpretations, [Num-Clf-N] phrases are strictly indefinite.

Moreover, building on Gerner & Bisang (2008; 2010), Hall (2019) reports another classifier language that marks definiteness on classifiers: Weining Ahmao. Weining Ahmao (Hmong-Mien, China) is typologically exceptional as its classifiers exhibit overt morphology for definiteness (definite vs. indefinite), number (singular vs. plural), and size (augmentative, medial, diminutive). Crucially, Weining Ahmao also uses tones on classifiers to mark (in)definiteness. However, as one reviewer notes, the Weining Ahmao data are more complex than what is reported in Hall (2019). Specifically, (in)definiteness is realized through various forms in Weining Ahmao, including voicing, devoicing, breathy voice, and tone changes. Moreover, tone changes occur only in some of the classifiers and do not follow a uniform pattern (e.g., shifting exclusively to a single tone). Given these complexities, I have chosen to leave Weining Ahmao out of the present analysis.

I argue that the empirical facts in Wenzhou Wu are best explained through a Clf-to-D head movement analysis combined with the Head Movement Constraint, contra Hall (2019). Specifically, I propose that [Clf-N] can be interpreted as either definite or indefinite, depending on whether the classifier undergoes Clf-to-D head movement. When there is no numeral, the classifier can raise to D, valuing the [+definite] feature from D and yielding a definite reading, as in (32b) (also see Li & Bisang 2012 for a similar proposal for Wenzhou Wu definite [Clf-N]). When there is a numeral, however, Clf-to-D head movement is blocked because raising Clf over Num to D would result in violating the Head Movement Constraint (Travis 1984). Consequently, the classifier must take a default (indefinite) spell-out, as shown in (35).

On the other hand, Hall (2019) rejects the Head Movement Constraint and instead proposes a dual-structure account, as summarized in (36).⁹

For definite [Clf-N] phrases, the classifier first moves through # and then up to D through right adjunction, resulting in the realization of the complex head [D [#-Clf]], as in (37).

For indefinite [Num-Clf-N] constructions, the numeral and the classifier form a constituent to the exclusion of NP, as illustrated in (38). As a result, Clf-to-D movement is blocked by an independently motivated ban on head movement out of a specifier (Roberts 2010).

While the dual-structure account can also capture the facts in Wenzhou Wu, it requires more complex assumptions. I argue that the Clf-to-D head movement analysis better accounts for the Wenzhou Wu data for two reasons. First, the dual-structure account presupposes that classifiers and numerals form a constituent cross-linguistically (i.e., classifiers are needed to allow numerals to count, rather than allowing nouns to be counted), which remains a controversial topic. Second, it assumes that Clf moves to the right of #, which is non-standard, as head movement typically involves left adjunction (Kayne 1991; 1994). Given these considerations, I propose that Wenzhou Wu definite [Clf-N] constructions are best analyzed under the Clf-to-D head movement approach. Further exploration of the dual-structure account is left for future research.

In the next section, I propose a grammatical tone analysis to account for these data, in which the grammatical tone of the head D docks onto the classifier if they form a single morphosyntactic word (M-Word) via head movement. Thus, D is not null, but instead consists of a grammatical tone that realizes the feature [+definite]. Crucially, the [+definite] feature triggers Clf-to-D head movement. I will show that this approach can potentially offer a unified explanation for the tonal marking of definiteness observed in classifier languages such as Wenzhou Wu and Nuosu Yi.

4.1 The nominal structure and theoretical assumptions

Before presenting the main analysis, I first outline my assumptions for the syntactic structure of the Nuosu Yi nominal domain as well as the theoretical framework I adopt.

4.1.2 Theoretical assumptions

The current analysis follows the framework of Distributed Morphology (DM) (Halle & Marantz 1993; Embick & Noyer 2001; Embick 2007, etc.). In DM, syntax is the starting point for all complex linguistic structures. Morphology interprets the output of the syntactic derivation (e.g., merge, move, agree), i.e., terminal nodes with features.

The output of syntax is spelled out phase-cyclically at each category-defining head (n, a, v), working from the bottom up. Crucially, the output is subjected to an ordered set of postsyntactic PF operations, see Figure 1.

Figure 1: DM architecture (Embick & Noyer 2001).

At PF, Vocabulary Insertion adds phonological content to syntactic terminal nodes with features. To illustrate, consider a Nuosu Yi demonstrative phrase, such as the one exemplified in (39), which has the levels of projection as outlined in the syntactic structure (40).

Specifically, at the stage of Vocabulary Insertion, the terminal node Dem has phonological content added to it, which is specified in Nuosu Yi’s Vocabulary Items (43).

1. (43)
1. a.

1. Dem[+deictic, +proximal] ↔ cy⁴⁴

1.  
1. b.

1. Dem[+deictic, +distal] ↔ a³³zzy⁴⁴

Likewise, through Vocabulary Insertion, the determiner morpheme is inserted at the head D and the classifier morpheme at Clf. In the next subsection, I will demonstrate that the tone change patterns in Nuosu Yi DPs can be captured if Clf is raised to D in syntax followed by definite tone docking in the PF component.

4.2 Clf-to-D head movement followed by definite tone docking

Recall that in article-less classifier languages like Cantonese, [Clf-N] phrases can denote definiteness without the presence of a demonstrative or a definite determiner (Wu & Bodomo 2009). Wu & Bodomo argue that a null D triggers Clf-to-D movement, which allows the classifier to combine with D and receive the [+definite] feature from D, thus the classifier behaves like a definite determiner. A similar approach has been proposed by Simpson (2005) for Southeast Asian languages such as Vietnamese, Hmong, and Nung. In a parallel way, Nuosu Yi can also express definiteness with [N-Clf] phrases without the presence of the definite determiner su³³, but only when the original 33 or 21 tone classifier takes the 44 tone. The key observations are repeated below:

In (44) and (45) the classifiers ma³³ and zzi²¹ take the 44 tone, yielding a definite reading while the indefinite reading becomes unavailable.

Following Wu & Bodomo (2009) and Simpson (2005), I argue that the definite [N-Clf] construction in Nuosu Yi can be best accounted for through Clf-to-D head movement, followed by definite tone docking. Specifically, I propose that the definite determiner carries a floating H tone that realizes the feature [+definite], which is essentially a grammatical tone,¹¹ as well as a determiner su³³.

The Vocabulary Insertion rule for the definite determiner is illustrated in (46). For clarity, I represent floating tones as circled superscript, following the convention of Rolle (2018).

In syntax, the classifier is triggered by the [+definite] feature on D and hence undergoes head movement to D. The derivation of the DP in (44) is shown in (47).

I follow the standard view of head movement as a head-adjunction operation (e.g., Baker 1988; Kayne 1991; 1994; Matushansky 2006; Roberts 2010), where the moved head X⁰ adjoins to the target head Y⁰. In particular, I follow Kayne (1991; 1994) in assuming that head movement must always be left-adjunction.

At PF, the floating tone ^Ⓗ docks leftward onto the preceding classifier if they form a single morphosyntactic word (M-Word) through head movement, as sketched below in (19). The docked tone is circled (e.g., Ⓗ).

This definite tone docking explains the tonal alternation observed in Nuosu Yi when classifiers express definiteness. It is the interaction between syntax (head movement) and phonology (tone docking) that results in the surface realization of definiteness through tone changes on classifiers.

Strictly speaking, I assume that the association domain of the floating H tone is within an M-Word (Embick 2007; 2015), a (potentially complex) head not dominated by a further head projection. Based on this definition, an M-Word could be (i) a complex head formed by head movement as in (49a), or (ii) each terminal head without head movement as in (49b) (each M-Word is represented in a box).

(49a) shows that syntactic head movement forms a complex head, which constitutes a single M-Word. What is crucial to the current analysis is that after Clf-to-D head movement, the classifier and the definite determiner are within the complex head [Clf-D], which is, in essence, an M-Word. Therefore, definite tone docking can take place. In contrast, when no head movement occurs, Clf and D remain as two distinct M-Words (49b), thereby preventing definite tone docking. Independent evidence supporting this M-Word association domain will be discussed in Section 6.1.

Recall that Nuosu Yi has four contrastive tones: 21 (L), 33 (M), 44 (MH) and 55 (H). Regarding definite tone docking, when the classifier has an underlying 33 (M) tone, the surface tone 44 (MH) arises as a simple association of the floating H tone to the base tone of the classifier, as illustrated in Figure 2a. However, when the classifier has an underlying 21 (L) tone, simple docking is prohibited, triggering further tone changes. Specifically, the docking of the floating H tone to a L tone classifier would result in an LH tone contour, which is unattested in the Nuosu Yi tone inventory. Hence, the change from LH to MH can be seen as a repair that preserves the structure of Nuosu Yi’s tone inventory, as illustrated in Figure 2b (see similar accounts for Cantonese tone changes in Chen 2000).¹²

Figure 2: Autosegmental sketch of definite tone docking.

In both cases, the definite H tone, as the realization of the feature [+definite], is overtly realized, hence su³³ can be omitted.

On the other hand, for classifiers with an underlying 55 (H) tone, the floating H tone applies vacuously because the tone of the classifier is already high. This can be seen in (50), where the classifier zzi⁵⁵ cannot undergo a tone change to 44, either when preceding the definite determiner (50b) or in its absence (50c).

In (50a), definite tone docking is vacuous when applied to the preceding classifier zzi⁵⁵, as it already bears a H tone. Consequently, the docking of the definite floating H tone has no effect, and the classifier surfaces with its original H tone. Notably, in this case, the definite floating H tone is not overtly realized, therefore, su³³ cannot be omitted. Taken together, the presence of the determiner su³³ is optional for classifiers like zzi²¹ and ma³³ when they surface with the 44 tone (as in examples (44) and (45)), but obligatory for classifiers with an underlying 55 tone (as in (50)), where the feature [+definite] has not been overtly realized through tonal changes.

Furthermore, recall that when an indefinite [N-Num-Clf] phrase is combined with the definite determiner, the entire phrase becomes definite, yet the classifier does not undergo any tone changes, as shown in (51) (cf. (52)).

The contrast between (51) and (52) critically shows that what plays a role in triggering tone changes on classifiers is not the linear word order, but the underlying syntactic structure. If tone change were determined solely by linear [Clf-D] order, we would expect it to occur on the classifier regardless of whether a numeral is present. However, this prediction is not borne out, as shown in (20) (repeated below as in (53)): when the numeral nyi²¹ ‘two’ is present, tone change on the classifier ma³³ is not allowed (cf. (51)).

Hence, I propose that a #P exists above ClfP, with the # head indicating singularity or plurality, and the numeral (NumP) occupies Spec #P (Li 1999). When a numeral larger than one is present, the # head is marked as [+plural]. I further propose that it is the presence of this [+plural] feature in the intervening # head that blocks Clf-to-D movement (see (54)). Consequently, the classifier and the definite determiner cannot form a single M-Word through head movement, preventing definite tone docking from occurring.

Another piece of evidence supporting the presence of a covert # head, marked as [+plural] when a numeral greater than one is present, comes from cross-linguistic patterns relating numerals and number marking. For example, in Turkish plural expressions, numerals cannot co-occur with the plural marker -ler, as shown in (55).

In (55a), the [+plural] feature is encoded in the # head due to the presence of the numeral iki ‘two’, and this plural meaning is interpreted semantically without the need of an overt plural marker on the noun. Likewise, I propose that in Nuosu Yi, when there is a numeral larger than one, the # head is marked with [+plural], even though no overt plural marker appears. Crucially, this covert head # with the [+plural] feature blocks Clf-to-D head movement.

That is to say, the covert # head, despite being silent, blocks Clf-to-D head movement just as overt # heads would. This is because the # head is not syntactically null. Since the numeral is in Spec #P, the # head must be marked with the feature [+plural].¹³

As a result of this structure (54), the classifier and definite determiner stay at the head Clf and the head D, respectively, thus forming two separate M-Words. Consequently, definite tone docking cannot take place. In (54), the floating H tone fails to find a host within its association domain, so it remains unhosted and does not get pronounced. This explains the absence of tone change on the classifier when a numeral is present. Nevertheless, the [+definite] D is licensed via the lexical insertion of ^Ⓗsu³³, leading to the definite interpretation. This further supports the idea that tone change on classifiers is structurally governed—specifically, by the presence or absence of Clf-to-D head movement.

4.3 Tone change as a signal for head movement: Evidence from the Q-det su³³

In this section, I show that the morpheme su³³ functions as a quantificational determiner (Q-det), when it appears without the definite floating H tone. Classical Generalized Quantifier (GQ) theory (Barwise & Cooper 1981) proposes that Q-dets occupy the head of a quantifier phrase (QP) and combine with a nominal argument (NP) to form a quantificational structure. I propose that, unlike its definite counterpart, the Q-det su³³ does not trigger Clf-to-D or Clf-to-Q head movement. An example of this quantificational use of su³³, glossed as ‘det’, is illustrated in (56), where it conveys a meaning equivalent to ‘each’ or ‘every’ in English.

As seen in (56), when su³³ functions as a Q-det, it is obligatory, and the classifier ma³³ retains its underlying 33 tone.

By contrast, when preceding the definite determiner ^Ⓗsu³³, the classifier ma³³ takes the 44 tone, as (57) illustrates. Moreover, the presence of su³³ is optional.

Recall that in Section 4.2, I proposed that the feature [+definite] of the definite determiner triggers Clf-to-D head movement, and that the floating H tone realizes this [+definite] feature at PF. The contrast between examples (56) and (57) follows naturally under this analysis. In (57), the definite 44 tone on the classifier is an overt realization of the [+definite] feature on D—realized via definite tone docking—which has triggered Clf-to-D head movement in syntax prior to PF. By contrast, in (56), the Q-det su³³ does not carry the [+definite] feature and therefore does not trigger Clf-to-Q head movement. I assume that the Q-det occupies the head of QP, with the underlying structure [_QP [_ClfP NP Clf] Q].

Furthermore, the universal reading of Q-det combined with a numeral is possible (58).¹⁴

Note that the numeral cy²¹ ‘one’ can occur in the Q-det phrase (58a) but it is not allowed in the definite DP construction (8b), repeated here as (59).¹⁵ This shows another difference between the Q-det su³³ and the definite D ^Ⓗsu³³.¹⁶

Moreover, as shown in (56) and (58), a Q-det phrase can freely occur in subject position. Nevertheless, its occurrence in object position is not allowed, see (60). Instead, another universal quantifier construction [N-cy²¹-Clf-zzi⁴⁴a²¹zzi³³] must be used, see (61).

On the other hand, there is no such restriction for the definite DP construction (i.e., [N-Num-Clf-D]) as it can occur in both subject and object positions (see Section 2.2).

In summary, su³³ (when not accompanied by the definite floating H tone) functions as a Q-det, and the universal [N-Num-Clf-Q] construction differs in several respects from the definite [N-Num-Clf-D] construction. The Q-det phrase is best analyzed under a non-movement structure, where the classifier and the Q-det remain in situ, whereas the definite determiner phrase is best explained through a Clf-to-D head movement analysis. These facts suggest that tone change is a signal of head movement, as it only occurs in the definite DP construction, where Clf-to-D movement takes place.

4.4 Comparison with non-movement analysis

As demonstrated in Section 4.2, the Clf-to-D head movement analysis and post-syntactic definite tone docking account for the tone change patterns on classifiers observed in Nuosu Yi DPs. At this point, it is worthwhile to compare the head movement analysis with other theoretical possibilities, namely, the non-movement analysis. Such an approach has in fact been discussed for Nuosu Yi in Jiang (2020: 242) who proposes that su³³ alone (i.e., without the floating H tone) functions as the definite determiner, assuming that classifiers occupy the Clf head position, as illustrated in (62).

Specifically, Jiang posits that the demonstrative and the numeral appear in the higher Spec ClfP and the lower Spec ClfP, respectively. Furthermore, NP is fronted to the initial position of ClfP. After NP-fronting, the ClfP merges with D, resulting in the surface word order [N-Num-Clf-D]. This non-movement analysis offers a simple way to account for the observed [N-Num-Clf-D] linear word order without assuming that the classifier moves to D when there is no numeral, while remaining in situ when a numeral is present.

However, the non-movement approach does not take the tone changes on classifiers into consideration. Under the non-movement analysis, one would need to propose the following morphophonological rule (63), which changes the tone of a classifier from 33 or 21 to 44 when followed by a definite determiner.

This rule would assume that the tone change on the classifier is determined solely by linear word order, which, as we have already seen, is not the case. In the remaining discussion, I will argue that the non-movement approach is less satisfactory than the head movement analysis for three reasons.

First, the non-movement analysis does not account for the absence of tone changes in the [N-Num-Clf-D] construction. Specifically, the morphophonological rule (63) would predict a tone change on classifiers in examples such as (51), repeated here as (64), as the 33 tone classifier ma³³ immediately precedes the definite determiner.

On the other hand, under the head movement analysis, I assume that the numeral occurs at Spec #P and the # head is marked with [+plural]. Clf-to-D head movement is blocked by the intervening # head due to the Head Movement Constraint (Travis 1984). Consequently, definite tone docking cannot take place as the classifier is not within the association domain of the floating H tone (i.e., within an M-Word).

Second, the non-movement analysis does not seem to shed light on the syntactic nature of the inseparability between Clf and D. Recall that in Nuosu Yi, D cannot combine with a head noun directly; rather, the mediation of a classifier is required, resulting in the [N-Clf-D] linear word order. Furthermore, nothing can intervene between Clf and D. Considering a non-movement structure such as (62), it remains unclear why Clf and D cannot be separated. On the contrary, under the Clf-to-D head movement analysis offered in (47), the restriction of the mediation of Clf follows automatically. In this framework, the inseparability of Clf and D arises from their formation as a [Clf-D] complex head through head movement, thus fully capturing the syntactic nature of Nuosu Yi’s stringent [Clf-D] order.

Moreover, in other languages with definite determiners like English and German, D can readily combine directly with the head noun. This raises a natural question, that is, why the [N-D] combination is prohibited in Nuosu Yi. The head movement analysis provides a compelling explanation for such typological patterns: in Nuosu Yi, it is the complex head D, i.e., [Clf-D]_D, that directly combines with the head noun. Hence, Nuosu Yi DPs are essentially consistent with languages such as English and German.

Lastly, the head movement analysis aligns well with cross-linguistic patterns. Clf-to-D head movement has been proposed to account for the definite bare classifier phrases in various East and Southeast Asian languages (Simpson 2005; Wu & Bodomo 2009; Li & Bisang 2012; Hall 2019). The definite [N-Clf] construction in Nuosu Yi fits into this broader typological pattern, supporting the claim that D is the locus of definiteness across languages, and that classifiers move to D to encode definiteness in classifier languages.

6.1 The role of tone in disambiguating syntactic structures

In head-final languages like Nuosu Yi, detecting head movement can be challenging because it is string-vacuous. The observation that classifiers and the definite determiner are always adjacent and uninterrupted can be explained by assuming that classifiers and the definite determiner occupy adjacent functional heads, with D selecting ClfP as its complement. This does not necessarily require postulating Clf-to-D head movement. However, I have shown that tone changes on classifiers in Nuosu Yi DPs play a crucial role in signaling syntactic structures, supporting the presence of head movement.

Recall that tonal marking of definiteness on classifiers is observed in both Wenzhou Wu and Nuosu Yi, which can be analyzed through Clf-to-D head movement. By contrast, the absence of tone changes on classifiers correlates with the presence of a numeral. Table 3 summarizes the patterns of definiteness marking with and without numerals in these two languages:¹⁷

The crucial point here is that when a numeral is present, the whole phrase is interpreted as indefinite and in both languages, the definite classifier (i.e., Clf_tone) cannot co-occur with a numeral. This restriction is not a result of linear word order at PF, as the presence of the numeral does not change the linear adjacency of the classifier and the definite determiner. Rather, it is the structural differences in syntax that lead to the restriction.

Building on previous studies on definite bare classifier constructions (Simpson 2005; Wu & Bodomo 2009; Li & Bisang 2012), I argue that tone changes on classifiers signal Clf-to-D head movement, as the tone change is led to by a grammatical tone of the head D (also see Jenks 2018 for the proposal that D has a phonological realization in article-less classifier languages). Crucially, this grammatical tone realizes the feature [+definite], which triggers Clf-to-D head movement. When a numeral is present, however, the classifier cannot move to D over Num or # because of the Head Movement Constraint (Travis 1984). Consequently, definite tone docking cannot occur in the presence of a numeral. This analysis thus provides a unified account for the marking of definiteness in these two languages through the interaction of tone docking and head movement.

Another example of tone disambiguating syntactic structures comes from [Pron-V] combinations in Nuosu Yi. Gerner (2013) observes that when a pronoun with an underlying 33 tone precedes a monosyllabic 33 tone verb, the pronoun takes a 44 tone if interpreted as the direct object, as shown in (74).

I extend this observation by noting that (i) the verb can be multi-syllabic, as in (75), and (ii) when two pronouns precede the verb, the object pronoun necessarily takes the 44 tone and it must be adjacent to the verb, as in (76).¹⁸

Here, the tone of the pronoun serves to disambiguate syntactic structures. Notably, no such tonal variation occurs in [(common) noun-verb] combinations. To account for this contrast, I propose that Nuosu Yi features both pronoun incorporation and anti-noun incorporation (Baker & Hale 1990). The direct object pronoun undergoes head movement to adjoin to the verb, forming a complex head [Pron-V], as sketched in (77).

I further propose that a floating grammatical tone (^Ⓗ) marks object pronouns when a pronoun and a verb form a single M-Word through head movement. In contrast, when the pronoun functions as a subject and moves from Spec vP to Spec TP, it remains in a separate terminal node from the verb, resulting in two distinct M-Words. This structural separation prevents the floating H tone from docking onto the pronoun, thereby blocking the tone change when the pronoun is interpreted as the subject.

Taken together, tone changes in Nuosu Yi’s [N-Clf-D] and [Pron-V] constructions can be attributed to head movement, suggesting that tone plays a crucial role in distinguishing between head movement and non-movement structures in the language.

6.2 Clf-to-D head movement in languages with overt D

Clf-to-D movement has been proposed to explain definite bare classifier phrases in several East and Southeast Asian languages (Simpson 2005; Wu & Bodomo 2009; Li & Bisang 2012), all of which are article-less classifier languages. In these cases, a null D is assumed to trigger movement of the classifier to D, allowing the classifier to acquire D-related features. The assumption behind this analysis is that a definite null D must be overtly instantiated by some lexical element to be licensed, thereby motivating Clf-to-D head movement. However, the data from Nuosu Yi show that Clf-to-D head movement can also occur in classifier languages with an overt D, suggesting that this movement is independent of whether D is overt or covert.

Moreover, regarding definite bare classifier phrases, the literature presents a long-standing debate on whether it is necessary to postulate a DP projection in article-less languages. For instance, Cheng & Sybesma (1999; 2005; 2012) argue against the Clf-to-D head movement analysis, proposing instead that there is no DP projection and that classifiers themselves function as the definite determiner, with Clf carrying a [+definite] value, as shown in (78).

However, the Nuosu Yi data contribute new insight to this debate by showing that classifiers act as definite determiners only when they move to D to acquire the [+definite] feature from D, as illustrated in (79). Crucially, when a numeral blocks Clf-to-D head movement in Nuosu Yi’s [N-Num-Clf-D] construction, the classifier can no longer convey definiteness.

In other words, classifiers in Nuosu Yi cannot denote definiteness on their own. To encode definiteness, classifiers must move to D to get assigned the [+definite] feature. Put differently, in Nuosu Yi, definiteness is encoded in D, not in Clf. Thus, the Nuosu Yi data offer strong evidence in favor of the Clf-to-D head movement analysis, supporting the necessity of a DP projection even in languages without overt definite determiners.