3.1 Full syncretism (AAA pattern)

First, consider the data from English once more. In English, all three functions (non-specific, specific unknown and specific known) are represented by only one lexical item some-, which means that English non-specific, specific unknown and specific known markers are fully syncretic:

This pattern is quite common and is attested in a large number of languages, for example: Spanish, Latvian, Dutch, Icelandic, Bulgarian, Kazakh, Hungarian, Hindi, Maltese and Hebrew. Consider some examples from Polish, Japanese and Korean, in which the three types of markers are also fully syncretic:⁹

3.2 No syncretism (ABC)

As mentioned in Section 2, Russian is a language with no syncretism between the three markers, which are realized as separate lexical forms (see examples in Section 2).¹⁰ Therefore, the paradigm of the three indefinite markers in Russian is as follows:¹¹

Another language where no syncretism is observed is Lithuanian (cf. Haspelmath 1997: 276; Kozhanov 2015; Pilka 1984: 97, 29). The three markers are -nors, kaž- and kai-:

Just as Russian, Lithuanian uses three separate markers for the non-specific, specific unknown and specific known indefinite functions:

3.3 ABB syncretism

Another pattern of syncretism is observed in Ossetic, Yakut, Georgian and Nanay. These languages lexically distinguish only specific and nonspecific markers. In other words, the specific known and unknown markers are syncretic to the exclusion of the non-specific marker. First consider the data from Ossetic (Haspelmath 1997: 281; Kulaev 1958: 52), in which the morpheme -dær is used in the specific unknown/known functions and is- appears in the non-specific function:¹²

The same kind of pattern is attested in Yakut (Afanas’ev and Xaritonov 1968; Haspelmath 1997: 290; Ubrjatova 1982). The forms used in Yakut are -ere (specific unknown/known) and -eme (non-specific):

Georgian is another language where markers used in the specific known and unknown functions are syncretic (Haspelmath 1997: 304; Sharahsendize 2018). The morpheme -ɣac appears in the specific unknown/known functions, while -me is used in the non-specific function:

The ABB pattern can also be observed in Nanay. The morpheme -daa is used in the non-specific function, while -nuu appears only in specific contexts. It should however be noted that the data on this language presented in Haspelmath (1997) are described as incomplete. The author did not have detailed information about the specific functions (known/unknown). I will therefore omit Nanay in the summary below:

To summarize, the specific unknown and specific known markers are syncretic in Ossetic, Yakut and Georgian:

3.4 AAB syncretism

Latin represents a pattern, where the non-specific and specific unknown markers are syncretic to the exclusion of the specific known one. The non-specific and specific unknown functions are represented by ali- (Haspelmath 1997: 254; The Bible: Acts 3: 5, Luke 8: 46, Mark 9: 38):

The specific known function is expressed with the morpheme -dam:

Therefore, the lexical exponents used in Latin are as follows:

3.5 Syncretism – summary

As shown in this section, the non-specific, specific unknown and specific known indefinite functions can correspond to a varying number of lexical items, from just one to as many as three. In other words, the non-specific, specific unknown and specific known indefinite markers (understood as morphemes used in particular indefinite functions) can be syncretic. Table 1 shows the patterns of syncretism attested in the studied language sample:

The data shown in Table 1 are arranged according to the map of indefinite functions (see Figure 1.), where the non-specific, specific unknown and specific known functions form a sequence of adjacent items. This ordering stems from a generalization made in Haspelmath (1997: 4), which says that a series of indefinite pronouns may express multiple indefinite functions, but only if those functions are adjacent on the map. In the context of syncretism, this generalization means that syncretism between indefinite markers should target only adjacent functions on Haspelmath’s map. This claim seems to be confirmed, since there are no attested cases in which the specific known marker is sycretic with the non-specific markers to the exclusion of the specific unknown marker.

The prediction that syncretism has to target adjacent items in a sequence does not however apply solely to indefinite markers and connects to a broader generalization known as *ABA. As argued in Bobaljik (2012), the ABA pattern will not appear in ordered sequences (paradigms) where items form a hierarchy of structural containment. The absence of the ABA pattern is exactly what we notice when indefinite marker forms are arranged in a paradigm according to the ordering found on Haspelmath’s map of indefinite functions. Whenever syncretism between indefinite markers is observed, it always targets contiguous cells in the proposed paradigm. Hence, if the *ABA generalization is correct, the patterns shown in Table 1 lead us back to the initial proposal of this paper, namely that features corresponding to the non-specific, specific unknown and specific known indefinite markers form a cross-linguistically universal hierarchy of syntactic structures that are syntactically contained in one another. At this point, it should however be mentioned that the analysis of syncretism reveals only the relative order of elements in a sequence and does not inform us about the direction in which a syntactic hierarchy grows. This means that two orders of derivation should be considered:

While the specific unknown structure will inevitably be in the middle of the hierarchy, syncretism does not show which marker should correspond to the smallest layer of the sequence (F₁). A criterion that could be applied to establish the least and the most complex items in the hierarchy is morphological complexity. This means that we could expect that more complex indefinite structures will be phonologically realized as a greater number of morphemes, or that we will observe morpheme stacking. The morphological complexity criterion is however not useful in the case of indefinite markers. In the analyzed data sample, there are no cases of marker stacking or examples that can clearly indicate what types of markers are structurally more complex than others.¹³

I will therefore use another criterion to establish which marker corresponds to the smallest structure in the hierarchy, namely functional compositionality. The idea is that the three types of markers can be arranged in a sequence on the basis of their growing functional complexity. Non-specific markers can be considered to correspond to the simplest syntactic structure, since they only introduce indefinite entities but lack other properties such as specificity or knowledge of the speaker. In contrast, specific unknown markers introduce an additional property, specificity of the referent. Lastly, specific known markers have yet another additional property, knowledge of the speaker. Thus, the functional properties of the three marker types reveal the relative complexity of their underlying syntactic structures; non-specific markers correspond to the smallest subset of the hierarchy, and specific known markers spell out the whole hierarchy.¹⁴ The ordering proposed on the basis of the functional complexity of non-specific, specific unknown and specific known indefinite markers not only matches the relative order of the markers established on the basis of syncretism but also suggests that (37-a) is the correct ordering for the proposed hierarchy. For this reason, I will adopt (37-a) in the analysis.

The observed patterns of syncretism and the cross-linguistic absence of the ABA pattern, as predicted by the *ABA generalization, constitute strong evidence to support the claim that non-specific, specific unknown and specific known indefinite markers correspond to subsets of a structural containment hierarchy. However, it remains to be explained how exactly the hierarchy is lexicalized so that it gives rise to the observed patterns of syncretism. A clear and coherent answer to these questions follows from the methodology provided by Nanosyntax. Mechanisms introduced by the nanosyntactic framework, such as cyclic phrasal spellout and spellout-driven movement, allow us to explain the derivation of indefinite markers and the emergent patterns of syncretism.

4.1 The nanosyntactic lexicon and spellout (Starke 2009, 2014, 2018)

Due to the fact that terminal nodes in Nanosyntax contain only one feature, lexical items will not correspond to terminal nodes but to phrasal constituents. In consequence, the nanosyntactic lexicon constitutes an organized list of well-formed syntactic structures that a particular speaker is familiar with (Starke 2014: 1–2). Each piece of structure stored in the lexicon corresponds to a particular phonological exponent, for example:

When spellout is triggered, it will involve accessing the lexicon in search for a lexical entry (a lexically-stored piece of structure) matching the structure built in syntax. If a matching lexical entry is found, its corresponding phonological exponent can be inserted into that structure (a syntactic constituent). This kind of spellout system makes Nanosyntax a late-insertion model of grammar (cf. Caha 2020: 1). No lexical information is carried by features or provided in any way before spellout takes place.

The Nanosyntactic lexicalization process in not only phrasal but also cyclic, in the sense that spellout is triggered each time a new feature is merged. In other words, with every feature merge, spellout will immediately access the lexicon in an attempt to find a lexical entry that matches the derived structure. Obtaining a matching entry will provide a phonological exponent for the phrase derived by the last merge operation. The outcome of the previous lexicalization cycle, i.e an exponent corresponding to a subset of the current structure, will be overriden by the newly inserted exponent.¹⁶

Now consider the following examples which illustrate the basic principles of the nanosyntactic spellout system:

To lexicalize the structure in (40), the spellout mechanism has to access the lexicon and find a lexically-stored tree that matches the derived structure, for example:

Once a matching lexical entry is found, its corresponding phonological exponent is inserted into the phrasal node projected by the last successfully merged feature (F₃P), and the system can proceed to merge the next feature in accordance with the fseq. The merger of a new feature will trigger another spellout cycle.

Matching between syntax and the lexicon is regulated by the Superset Principle. If no lexical entry is an exact match for a structure, it can still be spelled out with an overspecified lexical entry. In consequence, an exact match is not always necessary to spell out a structure. The Superset Principle states that:

This can be illustrated with the following example where two lexical entries are available:

When F₁ is merged, lexical access is triggered and (43-a) becomes the matching entry. The phonological exponent α can be inserted:

The next merge adds F₂ to the structure and a new spellout cycle begins. At this point, even though the lexicon contains no entry corresponding to the sequence [F₁, F₂], the Superset Principle will allow (43-b) to spell it out because the tree in (43-b) contains the derived structure. Thus, β will be inserted as the phonological exponent of [F₂ [F₁]], which will also overwrite the result of the previous lexicalization cycle:

The lexicalization of [F₁,F₂,F₃] is again straightforward, since (43-b) is an exact match for the sequence:

A question that naturally arises at this point concerns the spellout of F₁P with (43-b). After all, according to the Superset Principle, (43-b) is also a proper match for F₁P, since F₁P constitutes a subset of the tree in (43-b). This problem is solved by the second rule that governs lexical insertion in the nanosyntactic framework, namely the Elsewhere Principle:¹⁷

The Elsewhere Principle specifies that the closest match will always win in case of a competition between multiple lexical entries. For this reason, (43-b) will never be selected over (43-a) as the matching lexical entry for F₁P, as the latter contains fewer superfluous features.

The Superset Principle and the Elsewhere Principle explain syncretism and the *ABA generalization (Bobaljik 2007, Bobaljik 2012). The Superset Principle makes it possible for a single entry to spell out multiple structures that exist in a containment relation, which gives rise to syncretism. At the same time, the Elsewhere Principle constrains lexical insertion, so that more specific lexical entries will always win against ones containing more superfluous features. This means that a single lexical entry will not be used to spell out two non-contiguous layers of a syntactic hierarchy. In the nanosyntactic system, the following spellout results for the derivation of [F₁, F₂, F₃] will be illicit (given the entries in 49):

4.2 Spellout-driven movement and subderivation (Starke 2018; Wiland 2019)

The Superset Principle makes it possible to lexicalize structures that do not have exact matches in the lexicon (as they can be spelled out with overspecified lexical entries), however, there may be cases where none of the available lexical entries match the derived piece of syntactic structure. This presents a major problem for the lexicalisation mechanism (as described so far), assuming that the nanosyntactic spellout mechanism may not leave a feature without a phonological exponent.¹⁸ In other words, all features have to be spelled out in each spellout cycle.

The nanosyntactic answer to this issue, and another important feature of the nanosyntactic spellout system, is a mechanism known as spellout-driven movement (cf. Caha 2011; Starke 2018). Whenever lexicalization is impossible (due to the lack of a matching lexical entry), the need to spell out the structure triggers syntactic movement to obtain a lexicalizable tree geometry and save the derivation from crashing. Below, I discuss an example of spellout-driven movement that is relevant to the analysis presented in this paper.

Consider a new example derivation where F₃ has just been added to the structure, triggering spellout. Additionally, assume that there is no entry in the lexicon that could be used to spell out the created sequence. As seen below, in the previous lexicalization cycle α was inserted as the phonological exponent of F₂P:

Now assume that the lexicon includes a lexical entry containing F₃, which however cannot be used because it does not match the sequence [F₁, F₂, F₃]. The entry matches a constituent containing only F₃:

To use this entry, it is necessary to obtain a tree geometry, where F₃ forms a separate constituent. This can be achieved by a roll-up movement that evacuates the complement of F₃ to a position above it:¹⁹

The result of the roll-up movement is a tree in which the sister of F₂P can be spelled out as /β/ in line with (51). This means that F₃P will now be lexicalized as a separate morpheme:

The spellout mechanism will always attempt to remerge a piece of structure if a matching lexical entry is not found (spec-to-spec movement and then roll-up movement).²⁰ In cases where movement cannot produce a lexicalizable structure, the system will backtrack to the previous merge cycle (undo the last merge operation) and apply transformations at that point. This operation, i.e. backtracking, may be attempted multiple times if necessary. However, if all of these steps (movement and movement after backtracking) fail to provide a structure in which all features can be spelled out, there is still one more option available, namely subderivation (Starke 2018).²¹ This last-resort operation spawns a parallel derivation in order to construct a lexicalizable constituent containing the feature that the system is trying to spell out. The subderived hierarchy will subsequently be spelled out and integrated into the main structure as a complex left branch. Consider the following set of examples illustrating the process. Assume that it is impossible to spell out F₃ through movement (and backtracking) in (54-a) and the lexicon contains an entry such as (54-b):

Since it is not possible to save the derivation in any other way, the syntactic system is forced to form a subderivation. Because derivations begin with a merge, the parallel structure will have two features at the bottom: a copy of the last succesfully merged feature from the main sequence and F₃, which the system wants to lexicalize. It should however be noted that there is no consensus at the moment regarding the first feature that has to be merged at the bottom of a subderivation. In the presented analysis, I will follow Caha et al. (2019), which states that the main sequence and the parallel derivation will overlap. The last succesfully merged feature of the main derivation will appear at the bottom of the subderivation together with the feature that the system is trying to lexicalize:

A completed subderivation is integrated into the main structure as a complex left branch and spelled out as a prefix with respect to F₂P:

The spellout-driven movement and subderivation mechanisms are of great significance to the nanosyntactic theory, since they alter tree geometry and facilitate matching between syntactic structure and lexically stored trees. Moreover, these mechanisms reveal that there is a clear structural difference between prefixes and suffixes. Spellout-driven movement will lead to the formation of a suffix, which is a remnant constituent with a unary foot, while subderivation will create a prefix, which is a complex left branch with a binary foot:

4.3 Nanosyntax – summary

As shown in this section, Nanosyntax introduces a new perspective on concepts such as the lexicon, the architecture of syntax and spellout. According to the nanosyntactic model of grammar, syntactic derivations arrange features according to the order specified by a cross-linguistically universal sequence (the fseq). Constituents formed by features are cyclically spelled out with lexical information provided by entries contained in the lexicon. An entry can spell out a piece of structure it matches or is overspecified for (the Superset Principle). In cases where two or more lexical entries are eligible to lexicalize a particular set of terminals, the one with the fewest superfluous features wins the competition (the Elsewhere Principle). If there is no lexical entry that can be used to spell out a structure, the syntactic system will employ syntactic transformations to obtain a lexicalizable configuration of features or spawn a subderivation that can be spelled out and integrated into the main structure as a complex left branch.²²

The nanosyntactic model of derivation successfully explains syncretism as a phenomenon which stems from the basic rules of lexicalization, that is cyclic phrasal spell-out and the Superset Principle. Syncretism arises whenever two or more phrasal nodes can be spelled out with the same lexical entry, as permitted under the Superset Principle. Furthermore, the nanosyntactic spellout system accounts for the *ABA generalization, as the ABA pattern is not possible under the Elsewhere Principle. As shown in example (48), the Elsewhere Principle guarantees that the spellout system will always choose the most specific lexical entry for each cycle, which rules out the possibility of a lexical entry matching non-adjacent phrasal nodes.

Lastly, the spellout-driven movement and subderivation mechanisms reveal how suffixes and prefixes are formed. Syntactic transformations that are applied when a piece of syntactic structure does not match any lexical entry in the lexicon result in the formation of suffixes and prefixes. Spellout-driven movement will create a suffix with a unary foot, while subderivation will form a prefix with a binary foot.

In the next section, I will explain in detail the derivation of all the patterns of syncretism attested in the studied language sample (AAA, ABC, AAB and ABB) using the analytical tools of Nanosyntax outlined in this section. It also seems worth noting at this point that the nanosyntactic framework proves to be useful not only in the case of indefinite markers. So far, the approach has been successful in analyzing the phenomenon of syncretism in many other grammatical domains such as participles (Starke 2006), case (Caha 2009), spatial adpositions (Pantcheva 2011), negation markers (De Clercq 2013), demonstratives (Lander & Haegeman 2016), personal pronouns (Vanden Wyngaerd 2018), wh-pronouns (Wiland 2018; Vangsnes 2013), complementizers (Baunaz & Lander 2018a), verbal prefixes in Slavic (Wiland 2012; Tolskaya 2018), class prefixes in Bantu (Taraldsen 2010; Taraldsen et al. 2018) and internal structure of verbs (Taraldsen-Medova & Wiland 2019; Wiland 2019).

5.1 English (AAA pattern)

As seen in Table 2, English has only one lexical exponent for all three of the indefinite markers, namely (some-):

This means that in English, there is only one lexical entry that is used to lexicalize the sequence [F₁, F₂, F₃] and all its subsets. In other words, under the Superset Principle, the entry matching the whole hierarchy will also spell out [F₁, F₂] and [F₁]:

The three indefinite markers in English appear as prefixes with respect to the categorical stem, which means that their structure is formed through subderivation. When the first indefinite layer (F₁) is merged on top of the stem, it will not be spelled out due to the lack of a matching entry in the lexicon. The lexical entry in (59) will not match the derived structure:

Subsequent attempts to lexicalize the structure through spellout-driven movement (spec-to-spec and roll-up) and then backtracking will also not provide a lexicalizable tree geometry, at which point the derivation will have to resort to subderivation. The merger of F₁ will be undone and a new parallel derivation will be created. This subderivation will begin with the merger of F₁ with a copy of the last succesfully merged feature from the main sequence to form the minimal derivation structure (Caha et al. 2019). The properties of this feature are not relevant to the analysis, which is why it will be labeled F_x. The subderivation will be spelled out with the lexical entry in (59) and integrated into the main structure in the following way:

With the subsequent indefinite layers (F₂ and F₃), the steps described above will be repeated.²⁴ After multiple failed attempts to spell out the next feature provided by the fseq (movement and backtracking), the derivation will eventually backtrack to the stem and spawn a derivation providing the required features.²⁵ Each time, the subderivation will be spelled out as a subset of the entry in (59):

Starke (2018) suggests an alternative to the derivation of prefixes described above. According to this proposal, subderive may be considered such a costly operation that a parallel derivation should be kept active as long as possible, instead of being integrated into the main spine immediately after providing the feature that the system wants to spell out in the current cycle. This means that it may not be necessary to repeat the spellout algorithm (stay and spell out, move, backtrack and finally subderive) to derive each layer of the indefinite structure. After the first indefinite layer is subderived (F₁P), the subderivation will be extended and also provide F₂ and F₃ (depending on the intended indefinite structure). This will generate the same results as the derivation process show above.

5.2 Russian (no syncretism)

The indefinite marker paradigm for Russian contains three separate forms. This is shown in Table 3:

This means that there are three separate lexical entries which lexicalize the indefinite hierarchy. Each lexical entry matches a different subconstituent of the hierarchy:

Since the first two indefinite markers (non-specific and specific unknown) are suffixes, they are spelled out through the displacement of the stem to a position above the indefinite projections, in line with the spellout-driven movement mechanism (see Section 4.2). The phonological exponent matched with the specific unknown structure will overwrite the exponent corresponding to the smaller non-specific structure:

In contrast with the non-specific and specific unknown markers, the specific known marker koe- is a prefix, which means that its underlying structure should constitute a subderived phrase containing all three layers of the indefinite hierarchy. The derivation of this structure will begin with the merger of (F₃) on top of the tree shown in (64-b). The resulting structure will not match any entry in the lexicon:

Since spellout-driven movement will also fail to produce a lexicalizable tree geometry, the spellout mechanism will force syntax to create a subderivation. However, a subderivation spawned after layers F₁ and F₂ have been assembled will not lead to the formation of the desired structure. To properly derive a left branch containing the sequence [F₁, F₂, F₃], the syntactic system has to resort to backtracking, undo the merger of all the previously derived indefinite layers (F₁ and F₂), and then create a subderivation. Layers F₁ and F₂ will be merged at the bottom of the subderivation to form the minimal binary structure and the parallel derivation will then remain active until the F₃ is provided. The resulting structure will be integrated into the main sequence as a complex left branch (a prefix):²⁶

5.3 Yakut (ABB pattern)

Yakut represents a group of languages in which the specific unknown and specific known markers are syncretic to the exclusion of the non-specific marker. The forms are shown in Table 4:

The observed pattern (ABB) results from the fact that the indefinite hierarchy is lexicalized by the exponents of two lexical entries:

The first entry, i.e. (67-a), will spell out only the smallest subset of the hierarchy. This entry will always be matched with F₁P, since the entry in (67-b) contains superfluous features and will be ignored under the Elsewhere principle. When F₂P and F₃P are assembled, (67-b) becomes the only matching entry. This means that the indefinite structure will be spelled out the following way:

Because all three indefinite markers in Yakut are suffixes, they will be spelled out as remnant constituents formed through the displacement (roll-up) of the categorical stem:

5.4 Latin (AAB pattern)

Latin represents the AAB pattern; the non-specific and specific unknown markers are syncretic to the exclusion of the specific known marker. Table 5 shows the indefinite marker paradigm for Latin:

The examples below show the two lexical entries that lexicalize the indefinite hierarchy in Latin:

The non-specific layer of the indefinite hierarchy is subderived in Latin and lexically realized as a prefix (ali-). The subderivation procedure is applied, since all other attempts to spell out F₁, i.e. movement of the stem and movement after backtracking to the previous merge cycle, will not produce a lexicalizable tree geometry:

The only way to obtain a lexicalizable tree is through subderivation where F₁ will be merged with a copy of the last successfully lexicalized feature from the main spine (F_x) to form a minimal strucutre. The resulting constituent can be lexicalized as a subset of (70-a) (ali-):

Next, to create the specific unknown structure, F₂ will be merged on top of F₁P. The resulting structure will however not match any of the available lexical entries. Any subsequent spellout operations (i.e. move and backtrack) will also not generate a tree that can be spelled out with (70-a) or (70-b):

To create a lexicalizable structure containing F₂, it is again necessary to backtrack to the stem and spawn a subderivation.²⁷ The subderived constituent will be formed with F₁ and F_x as the foot (which can be spelled out as a subset of (70-a)) and grown to contain F₂:²⁸. The resulting constituent will be spelled out with (70-b)

The specific known marker -dam is a suffix, which means that it is derived through spellout-driven movement. It is not possible to spell out F₃ immediately after merge, which will trigger the spellout algorithm (spellout-driven movement):

The simplest way in which the syntactic system can obtain a tree geometry where all the layers of the indefinite hierarchy form a constituent is to extract the stem constituent (the complement of F₂P) from the phrase projected through the subderivation of the left branch F₂P (the prefix). Note that the proposed movement is not in line with the spellout algorithm proposed in Starke (2018), which predicts only spec-to-spec and roll-up movements before the spellout system has to resort to backtracking and finally subderivation. However, since the extraction of a previously lexicalized constituent is the most straightforward way of deriving a suffix (-dam) from a prefix (-ali), it is reasonable to suggest that constituent extraction may constitute the last movement option that may be applied (after spec-to-spec and roll-up movements) before backtracking has to be used.²⁹ The suffix (F₃P) is lexicalized with the phonological exponent of entry (70-b):³⁰

Note that despite being a suffix, the constituent formed through the extraction of the stem will have a binary foot. This is due to the fact that the previously derived left branch (F₂P) will remain as a remnant constituent inside the newly created suffix.³¹

5.5 Analysis – summary

The nanosyntactic model of grammar can be successfully used to create a coherent analysis of the derivation of indefinite markers. Non-specific, specific unknown and specific known indefinite markers, which appear as either prefixes of suffixes on categorical stems, constitute phonological exponents corresponding to subsets of a hierarchy of indefinite features. Language-specific restrictions on lexicalization (the contents of the lexicon) are responsible for the fact that languages spell out the indefinite hierarchy in different ways. However, despite this variation in lexicalization of indefinite markers, the nanosyntactic spellout system, regulated by the Superset and Elsewhere Principles, guarantees that the *ABA rule is not broken.

A matter that may be considered in terms of future research is the fact that indefinite markers also do not seem to violate *ABA when it comes to their morphological forms (prefixes/suffixes). In other words, there are no languages in the studied language sample in which the non-specific and specific known markers are suffixes (or prefixes) to the exclusion of the specific unknown marker. However, even if we are dealing with a genuine pattern in this case, at this point, it is not clear what its cause may be.³²

6.1 Indefinites syncretic with wh-pronouns

In a number of languages, we observe syncretism between interrogative pronouns and indefinite pronouns. Consider examples from Hopi, Dyirbal, Khmer, colloquial Dutch, colloquial German and Mandarin Chinese:³³

Examples of this kind of syncretism can be analyzed as cases where interrogative pronouns are spelled out as syncretic subsets of indefinite pronouns. In other words, the same lexical entry can be used to spell out the interrogative structure as well as the interrogative structure with indefinite layers on top of it. Consider, the following example from Mandarin Chinese, where only non-specific pronouns are syncretic with interrogatives (cf. Li 1992 and Lin 1998).³⁴ The non-specific indefinite (shénme) will be lexicalized with the same lexical entry as the interrogative pronoun (shénme)

In colloquial Dutch, pronouns for all three indefinite functions are syncretic with the interrogative pronouns. This means that for a particular category, the non-specific, specific unknown, specific known and interrogative pronouns are all spelled out as subsets of a single lexical entry:

6.2 Paradigm gaps

The studied data sample contains a number of languages in which we observe the absence of pronouns for one or more indefinite functions. The presence of such paradigm gaps makes these languages largely irrelevant in proving the proposed claim concerning the syntactic structure of indefinites. However, it should also be stressed that no language shown below can be used as an example against it. I discuss languages with paradigm gaps in an attempt to provide a possible explanation for this phenomenon:³⁵

There is yet another reason why it may be worth taking a closer look at paradigm gaps. There is a possibility that gaps in indefinite pronoun paradigms form a certain pattern. The currently available data seem to indicate that if only one indefinite pronoun type is missing, it is the most complex one (specific known), and if two indefinite pronoun types are absent, it should be the specific ones. However, since my data concerning languages with paradigm gaps is limited, I am not able to confirm the existence of an actual pattern in this case.

6.3 Paradigm gaps – possible explanations

The data in Table 6 can be accounted for in two ways. The first possibility that we may want to consider is that the indefinite hierarchy is smaller (reduced) in some languages. In other words, one or more layers of the indefinite hierarchy could be “cut off”. However, the main issue with this approach is that it does not seem to be consistent with the available data. In Irish, for example, despite the lack of pronominal indefinites, the three indefinite functions can be expressed with the noun modifier éigin ‘some’. Some other languages mentioned in this section, e.g. Swahili, Filipino and Chinese, also have indefinite modifiers corresponding to the indefinite functions that cannot be expressed pronominally. This shows that the indefinite hierarchy is unlikely to be absent or reduced in these languages. If so, then sequence reduction cannot be the correct explanation of the data.

A solution that takes into account indefinite noun modifiers is closely connected with the nanosyntactic lexicon. If paradigm gaps do not stem from variation in syntax, they may be caused by the lexicon.³⁶ In other words, a language does not have indefinite pronouns of a given type if it lacks a lexical entry (or entries) that is necessary to spell out the indefinite structure corresponding to that type. This, of course, means that features which cannot be spelled out as part of an indefinite pronoun may still appear inside other lexicalizable structures, such as indefinite modifiers. Additionally, this solution to the problem of paradigm gaps explains the regularity that can be seen in Table 6. If only one indefinite pronoun type is missing, it will be the specific known type. If two pronoun types are absent, it will be the specific types (specific known and unknown). This kind of pattern should be expected if paradigm gaps are caused by the unavailability of lexical entries. Under the nanosyntactic rules of lexicalization, a paradigm gap could only appear if a structure lacks a matching lexical entry and cannot be spelled out by a larger entry matching its superset. If a given structure can be spelled out, then all its subsets are also lexicalizable. For this reason we should never see a gap for an indefinite function if pronouns for a more complex function (or functions) are available.