1 Introduction

This squib aims to put limits on Nanosyntax and the idea of highly articulated feature decomposition. I argue that some coexpressions arise from semantic underspecification rather than from structural containment relations.

As an example of a mistaken nanosyntactic analysis of lexical items, I revise the findings of Dekier (2021) concerning the syncretism and containment of indefinites in light of their semantic implausibility and empirical inadequacy. I present an alternative semantic account of the indefinites’ map from Degano & Aloni (2025),¹ which is less problematic and more predictable.

In Section 2, I present the indefinites’ map and the syncretism patterns it gives rise to. In Section 3, I present Dekier’s analysis of these patterns. In Section 4, I discuss the problems with Dekier’s analysis. In Section 5, I present a semantic account of the indefinites’ map from Degano & Aloni (2025). In Section 6, I compare the two accounts and present additional predictions of the semantic account. Then, in Section 7, I discuss the implications of this comparison for grammar architecture. Finally, Section 8 concludes and discusses potential avenues for future research in Nanosyntax.

2 Indefinites’ map and syncretism patterns

Dekier proposes that the part of Haspelmath (2001)’s indefinites’ map illustrated in Figure 1 can be accounted for in terms of a nanosyntactic (Caha 2009; Starke 2009) approach to syncretism patterns.

Figure 1: Indefinites’ map from Haspelmath (2001).

Haspelmath’s map is based on a cross-linguistic survey of indefinites and can be read as follows. The nodes represent different uses/functions of indefinites, and only contiguous nodes can be coexpressed by the same form in a given language.

Our main concern is with (1) specific known, (2) specific unknown, and (3) irrealis non-specific indefinites, which are illustrated in examples (1) from Haspelmath (2001: 5–7).

The first two usages—specific known and specific unknown—are usages where the DP in question has a specific referent in the world, but the speaker either knows who this referent is (1) or does not (2). The third usage—non-specific irrealis—is a usage where the DP in question does not have a specific referent in the world, and its use in Haspelmath (2001) is typically in imperatives.

In English, these three usages are coexpressed by the same form some; however, cross-linguistically, this is not always the case. As illustrated in Haspelmath’s map, any coexpression pattern is possible as long as it is contiguous on the map. That is, the pattern where (1) and (3) are coexpressed, but not (2), is not attested. This is illustrated in Table 1.

In the next section, I present Dekier’s analysis of these patterns.

3 Dekier’s analysis

Dekier considers such syncretism patterns as an instance of a broader *ABA generalization (Bobaljik 2012),² which states that the ABA pattern is unattested in hierarchically ordered sequences. For the indefinites, Dekier proposes the structure in (2).

This tree states that the specific known indefinite contains the features of both the specific unknown and non-specific indefinites, while the specific unknown indefinite contains the features of the non-specific indefinite.

Dekier explicitly presents only one semantic argument for this hierarchy, as opposed to the reverse one, which would also comply with the *ABA generalization. It is based on the semantic properties of the indefinites. The specific known usage is the most complex one, since its semantics involves both the existence of a specific referent in the world and the speaker’s knowledge of this referent. The specific unknown usage is less complex, since it only involves the existence of a specific referent in the world, but not the speaker’s knowledge of this referent. Finally, the non-specific usage is the least complex, since it does not involve a specific referent in the world at all.

For the hierarchical ordering, Dekier’s work provides additional supporting evidence. Lexical gaps in indefinite pronouns start from the specific known indefinites and expand to the non-specific indefinites, as illustrated in Table 2.

Dekier says that the gaps are expected if the hierarchy in (2) is correct, since if a less complex indefinite is not lexicalized in a language, the more complex indefinites would not be lexicalized either. Otherwise, the lexical entry that spells out the more complex indefinite could also spell out the less complex indefinite due to the Nanosyntactic Superset Principle.

In the next section, I discuss the problems with Dekier’s analysis.

4 Problems with Dekier’s analysis

The main shortcoming of Dekier’s hierarchy, as noted by Dekier (2021) himself, is that no actual morphological containment pattern is attested in the indefinites. However, the nanosyntactic approach suggests that there should be such patterns if the features F₂ or F₃ from the hierarchy in (2) are spelled out by distinct exponents. This raises the question of why they never are.³

The semantic argument for the proposed hierarchy is not particularly compelling; it appears to be largely stipulative and only gains plausibility if one presupposes a hierarchical structure whose orientation remains to be determined. As will be discussed in Section 5, the semantics of indefinites can be, and arguably are better, accounted for within an alternative framework in which the specific unknown indefinite is the most complex, while the other two types are less complex, albeit in distinct ways.

The paradigm gaps argument seems to be the most solid one theoretically; however, the absence of the ABA pattern is based on just three languages with partial gaps and three languages with complete gaps. It is not clear how robust this generalization is and whether it can be considered a real argument in favor of the hierarchy in (2).

In the next section, I present a semantic account of the indefinites’ map from Degano & Aloni (2025) and show that it is empirically more adequate and semantically more plausible than the one from Dekier (2021).

5 Semantic account

Degano & Aloni (2025), following the less formalized work of Farkas & Brasoveanu (2020), propose that definiteness and specificity can be united (and thus the indefinites’ map can be explained) in terms of variation and constancy of the denotation of the DP in question. Their account is based on team semantics (Hodges 1997; Väänänen 2007), and in what follows I present their account in a simplified way and discuss its implications.

According to Degano & Aloni (2025), indefinites, in addition to existential quantification, also exhibit variation and constancy restrictions on their denotation. Variation and constancy are calculated across all worlds in the case of definiteness and across a given epistemic world in the case of specificity. Variation gives us indefinite and non-specific readings, while constancy gives us definite and specific readings. The variation and constancy are captured by two atomic formulas: var(y,x) and dep(y,x). The first one means that x varies with respect to y, that is there are at least two assignments where x takes different values when y is held constant; the second one means that x depends on y, that is, for each assignment x takes the same value when y is held constant.

To describe the distribution of indefinites, we need v (the designated world variable, the values of which correspond to the worlds, which are possible given the speaker’s beliefs) and ∅ as possible instantiations of y. dep(v,x) means that x has exactly one value within each epistemic world v, but can vary across the worlds, while var(v,x) means that x has at least two different values within at least one epistemic world v. Meanwhile, var(∅,x) means that x has at least two different values (calculated across all epistemic worlds), and dep(∅,x) means that x has exactly one value across all epistemic worlds. These four notions, including their combination and the absence of any restrictions, give us six different types of indefinites, illustrated in Table 3.

Crucially, this analysis considers the indefinites that can express several functions not as syncretic forms, but as a single form with a less restrictive semantics that allows it to occur in several contexts. For example, the unmarked indefinite (i) can occur in all three contexts because it does not impose any restriction on the denotation of the DP in question. The specific indefinite (ii) can occur in both specific contexts because it imposes a restriction on the denotation of the DP in question within each epistemic world v, but not across all worlds. The epistemic indefinite (iv) can occur in both indefinite contexts because it imposes a restriction on the denotation of the DP in question to have at least two different values across all worlds, but not necessarily within each epistemic world v.

The unattested indefinite (vi) is predicted to be unattested because the requirements it imposes are contradictory and can only be posited in a disjunctive way.

In the next section, I compare the two accounts and present additional predictions of the semantic account.

6 Comparison of the two accounts

The semantic account presented above offers clear advantages over Dekier’s nanosyntactic approach. Furthermore, the semantic account makes two additional predictions. First, the specific unknown indefinite (vii) is predicted to be attested but less frequent than the other indefinites because it imposes a conjunction of two requirements on the denotation of the DP in question, which makes it more complex than the other indefinites. This prediction is borne out, as noted in Degano & Aloni (2025: 973).

The second prediction concerns diachronic changes. Haspelmath (2001) describes several attested diachronic changes involving indefinites relevant for us.

Extension from SU to the right. This type of change is from specific unknown indefinites (vii) to epistemic indefinites (iv) (Haspelmath 2001: ch. 6.4.4), as illustrated in Figure 2. This change is evidenced in indefinites that are etymologically derived from expressions with meanings like ‘I don’t know’ and their original use implies that the speaker could know the referent (that is, they are specific), but does not (that is, they are unknown) (Haspelmath 2001: ch. 6.2.1). This type is exemplified by Lithuanian kaž-, Albanian di- and older German neiz- (Haspelmath 2001: 153).

Figure 2: Diachronic change from SU to the right.

Extension from NS to the left. Several indefinites (Haspelmath 2001: ch. 6.4.2), given their etymology, originally were used in ‘free choice’ contexts, which are connected to the specific known and specific unknown usages via the non-specific one, according to Haspelmath’s map, see Figure 1; thus, for us such indefinites’ extension to the left means extension from the non-specific indefinites (iii). Among relevant cases for us we have the French unmarked indefinite (i) quelque (Haspelmath 2001).⁴ Additionally, the change from non-specific indefinite (iii) is instantiated by German epistemic indefinite (iv) irgend- (Aloni & Port 2015). This type of change is illustrated in Figure 3.

Figure 3: Diachronic change from NS to the left.

The semantic account allows us to predict both the extension to the left and to the right. Aloni & Degano (2022: 16) note that the account predicts that the non-specific indefinite (iii) can develop into the epistemic indefinite (iv) because the requirement var(v,x) entails and can be weakened via diachronic change to var(∅,x). Diachronic weakening from var(∅,x) to no restriction, and from dep(v,x) to no restriction, is also possible. This yields, respectively, the changes from epistemic indefinites (iv) to unmarked indefinites (i) and from specific indefinites (ii) to unmarked indefinites (i); both changes are attested, as noted above.⁵ The semantic approach also predicts changes from specific indefinites (ii) to unmarked indefinites (i) and from specific known indefinites (v) to specific indefinites (ii), both of which are unattested. However, given that we have only a few clear cases of diachronic change, this may reflect a lack of data. The crucial point is that all attested changes instantiate semantic weakening, independent of their direction (which would be crucial for a hierarchical Nanosyntactic approach), as predicted by the semantic account.

Diachronic Nanosyntax is not yet developed, and we do not know what changes can lexical entries undergo, thus it is not a straightforward task to compare predictions of two approaches on diachrony. However, given that we know that words become obsolete, we can legitimately assume that the loss of a lexical entry is an option for diachronic change. Given this, the Nanosyntactic account could predict only a part of what the semantic account predicted, taking into account the Elsewhere and Superset Principles (Caha 2009; Starke 2009). Consider the following case, illustrated in examples (3)–(4). If a language initially had two lexical entries, one spelling out F₁ and the other spelling out F₂ and F₁, then if the entry spelling out just F₁ is lost, the entry spelling out both F₂ and F₁ could spell out just F₁ due to the Superset Principle. This would lead to a change from the specific unknown indefinite (vii) to the epistemic indefinite (iv).

Thus, Nanosyntax by default predicts that as a diachronic change, a lexical entry could gain the ability to spell out a subset of what it originally could, and Nanosyntax would not predict the attested change in the reverse direction.⁶ In this way, Nanosyntax would also predict a change from specific known (v) to specific (ii) indefinites, and consequently from specific (ii) to unmarked indefinites (i), both unattested. The semantic approach in fact mispredicts these changes as well, as noted above.⁷

The general point is that if we had a hierarchical structure for indefinites as proposed by Dekier (2021), we would expect to see a unidirectional diachronic change pattern, from more complex indefinites to less complex ones, as is the case for direction > recipient (see Caha (2017) for hierarchy and Haspelmath (2003) for diachrony), or comitative > instrument (see Caha (2009) for hierarchy and Narrog (2010) for diachrony). However, as we have seen, the attested changes for indefinites go in both directions, which is unexpected under the nanosyntactic approach.

These empirical advantages of the semantic account over the nanosyntactic one raise broader questions about the architecture of grammar and the role of syntax versus semantics in accounting for coexpression patterns.

7 Grammar architecture

I suggest that the plausibility of the semantic account demonstrates that it is not so that all coexpressions should be analyzed as spellouts of subconstituents of a complex syntactic structure, but rather sometimes they are just spellouts of non-hierarchically structured nodes. This is not to say that Nanosyntax cannot account for syncretism patterns, but rather that it should not be the default approach to all coexpression phenomena, and that semantics should not be neglected.

There are two options how to model the structure of indefinites in a proposed way. The first one is to posit that in syntax there are different terminal or non-terminal ∃ nodes not related to each other by any featural containment and coming with pairs of lexical entries, as in examples (5)–(7) for Russian AB(B/C)⁸ indefinites (Haspelmath 2001: 65).

Another option is to propose that there exists only one ∃ node, as illustrated in examples (8)–(9), which comes with lexical entries that are form–syntax–meaning triples.

The second option is contra Preminger (2021)’s non-semiotic approach, which, based on idioms, claims that forms should not be assumed to (directly) have meanings, but rather all lexical entries are either syntax-form or syntax-meaning mappings. Thus, this option (if to be aligned with Preminger’s findings) requires two different types of lexical entries, since idioms of the type discussed by Preminger (2021) would still need to be accounted for. Traditional Nanosyntax itself follows (or at least is compatible with) non-semiotic approach for functional elements, which are decomposed into features, but it utilizes semiotic entries for roots, which are assumed to have lexical meanings— as triples in the sense of Ramchand (2008); Vanden Wyngaerd et al. (2020); Vanden Wyngaerd & De Clercq & Caha (2021). If the indefinites data are to be analyzed in line with Nanosyntax, this suggests that the ∃ head should be treated as a root.

The data from indefinites provides evidence for the first option. There are cases of morphologically complex indefinites that, however, are not in morphological containment relations with other indefinites. This is exemplified by the Russian non-specific marker -nibud’, which consists of the negative marker ni and the subjunctive form bud’ of the verb byt’ ‘to be’. This suggests that the ∃ nodes are complex and complex in different ways, which means that they themselves are different from each other.

Although the choice between the two options for indefinites is probably the first one, it is unclear whether this can be generalized to other cases of coexpression phenomena of a non-syntactic nature. I leave this question for future research.

8 Conclusion

In this paper, I have argued against the universal applicability of the nanosyntactic approach to coexpression phenomena, using indefinite pronouns as a test case. I have shown that Dekier (2021)’s nanosyntactic analysis of indefinites faces serious empirical and theoretical problems: it fails to account for the absence of morphological containment patterns, relies on a weak semantic argument for feature hierarchy, and makes incorrect predictions about diachronic change.

In contrast, the semantic account of Degano & Aloni (2025), which treats indefinites as lexical items with varying degrees of semantic restrictiveness rather than as syncretic spellouts of complex feature structures, provides a more empirically adequate and theoretically sound explanation of the attested patterns. Crucially, this account better predicts the observed diachronic changes and the relative frequency of different indefinite types.

The broader implication of this comparison is that coexpressions cannot be reduced to PF alone: different contexts in which the same exponent is used need not be derived from distinct syntactic structures, but may instead be an epiphenomenon of semantic underspecification at LF.

Consequently, there are clear cases where Nanosyntax seems to be the best approach. Some elements of the research in question clearly exhibit morphological containment patterns and are thus best analyzed with Nanosyntax. For example, works on case (Caha 2009; 2017; Starke 2017; Caha 2018; 2019)⁹, locational case (Pantcheva 2011), comparatives (Caha & De Clercq & Vanden Wyngaerd 2019; Vanden Wyngaerd et al. 2020; De Clercq et al. 2022), numerals (Wa̧giel & Caha 2020; 2021), number (Caha 2022), persons (Vanden Wyngaerd 2018), verb actionality (Caha & De Clercq & Vanden Wyngaerd 2023), and ontological categories (Baunaz & Lander 2018b).

Other research in Nanosyntax explores paradigms, which, by definition, require featural combination and cannot be modeled otherwise; see, among others, (Taraldsen 2019; Caha 2021; Cortiula 2025).

Complementizers (Baunaz 2018; Wiland 2018; Baunaz & Lander 2018a), imperfective aspect (Starke 2021; Dikmen & Demirok 2025), and negation (De Clercq 2020) are not clear cases. As far as I am aware, there are no cases of morphological containment presented for complementizers and aspect and only partial morphological containment for negation markers, but this may be due to the limited data available. This cannot be said of Dekier (2021), who bases his analysis on large typological data from Haspelmath (2001).

Funding information

The results of the project “Linguistic and cognitive diversity in formal models, computer tools, and educational resources” (2025–2027), carried out within the framework of the Basic Research Program at the National Research University Higher School of Economics (HSE University), are presented in this work.

Acknowledgements

I am grateful to Alexander Sergienko, Daniar Kasenov, and other colleagues at the Laboratory on Formal Models in Linguistics, HSE University, for discussion, and to Pavel Rudnev, the anonymous reviewers, and the editor Guido Vanden Wyngaerd for their helpful comments on the paper.

Competing interests

The author has no competing interests to declare.