2.1 pan as an R-expression

We start our empirical exploration by discussing uses of pan that do not differ from other lexical noun phrases. This discussion will form a baseline for other uses of pan. An example of pan as an R-expression is given in (1). Here we see pan in an argument position (direct object). The nominal is inflected for ϕ-features (feminine singular) and case (genitive). In this example, the nominal is modified by a demonstrativeFootnote ⁹ and can serve as an antecedent to a pronoun, suggesting that in this case pan realizes a DP structure.

Further evidence that pan in its R-expression use corresponds to a DP comes from the fact that it can be coordinated with a proper name, as in (2), and that it is modifiable by a relative clause, as in (3). Furthermore, this instantiation of pan can serve as the answer to an individual-questioning wh-question, as in (4).

In this guise pan behaves like any other regular noun in Polish. However, unlike other lexical nouns, pan exhibits additional syntactic behaviour, as discussed below.

2.2 pan as a nominal modifier

pan can also exhibit adjective-like behaviour. In this guise, the number, gender, and case of pan depend on the number, gender, and case of the root nominal. For example the feminine noun ‘female director,’ as in (5), requires the feminine form of pan. Since the ϕ-features of pan in the modifier use are dependent on the grammatical features of the head noun, they must come unvalued from the lexicon.Footnote ¹⁰ Note, however, that even though pan exhibits agreement, overt inflection is nominal, not adjectival.

The modifier version of pan has the syntactic distribution of a prenominal modifier.Footnote ¹¹ However, unlike other prenominal modifiers, it must be adjacent to the NP complex it modifies, as in (6).Footnote ¹² That is, pan is unlike other prenominal modifiers that can alter their order with respect to each other, as well as with respect to the head noun based on information-structure properties, as in (7).

The fact that pan must be adjacent to the root nominal and that further structural elements, such as adjectival modifiers, possessives and determiners, linearly precede pan suggests that pan is adjoined low in the structure, most likely within NP.

pan can, however, modify proper names as well, as in (8). Assuming that proper names are DPs, pan can function as a DP-modifier as well. pan as a DP modifier gets its features valued by agreement with the ϕ-features of the proper name, demonstrated in (9).

2.3 pan as a pronoun-like element

The previously discussed distribution of pan has given us a hint of its somewhat unusual syntactic properties, but it is the pronominal guise of the nominal that fully demonstrates its structural fluidity. Nominal structures based on pan can also be used to denote an interlocutor. First, observe that pan, as other R-expressions, can be used as a vocative, as in (10). In Polish, vocatives are marked by a specialized inflectional morphology and are external to the event structure of a clause.Footnote ¹³ However, as examples (11)–(12) demonstrate, pan can also be used as an addressee-oriented element (akin to a 2^nd person pronoun) when syntactically integrated as an argument of a clause. In this case, the nominal denotes the interlocutor and has the syntactic distribution, including case marking, of a DP argument.

However, the syntactic property that most clearly reveals the hybrid nature of pan within the nominal system is that in its pronominal use, pan systematically triggers variable agreement on predicates. As can be seen in (13), an interlocutor-interpreted panFootnote ¹⁴ either triggers a nominal-like agreement, that is, 3^rd person agreement on the predicate, as in (13a), or it triggers 2^nd person agreement, as in (13b). As far as we know, this agreement variation does not correlate with any interpretive differences.Footnote ¹⁵

As pointed out by an anonymous reviewer, kinship nouns also allow variable agreement, as seen in (14), suggesting that the pattern we see with pan is an intrinsic property of the system, not a lexical exception.Footnote ¹⁶

To our knowledge, Polish does not allow variable person agreement in any other domain. Polish does allow semantic agreement,Footnote ¹⁷ but the switch between grammatical and semantic value never appears within a local syntactic domain, such as in subject-predicate agreement. Instead, the switch to a semantic value only occurs across a sentence boundary and is mediated by a pronominal (a pronoun or a pro). The example in (15) shows semantic agreement in gender. In this example, a grammatically neuter noun that denotes a female triggers an obligatory neuter agreement on the predicate, but the agreement can switch to feminine in a following clause (here, a subject pro mediates the predicate agreement). The example in (16) demonstrates a switch in number. Numeral phrases like ‘five boys’ trigger neuter singular agreement, but the predicate of the following clause agrees in the plural, reflecting the semantic plurality of the numeral phrase. As with number, the switch to semantic agreement is not local.

As for semantic agreement with person, there are no clear examples parallel to the aforementioned semantic agreement with number and gender. Instead, we briefly discuss several constructions that display some similarity to pan in their dual person behaviour, yet they differ in other syntactic properties. We start with imposters, namely, constructions in which a 3^rd person noun phrase is used to refer to the speaker or the interlocutor. According to Collins and Postal (Reference Collins and Postal2012), these DPs contain a pronoun in their structure, and their structural and interpretive properties follow from this additional prononominal. Polish imposters, like their English counterparts (Collins and Postal, Reference Collins and Postal2012), require grammatical person agreement, as in (17). That is, the embedded pronoun cannot trigger local subject-predicate agreement.

Proper names, when used to address an interlocutor and when morpho-syntactically marked as a vocative, can only trigger 2^nd person agreement, as in (18).Footnote ¹⁸ Note that unlike in English, a vocative can be linearized inside a clause; that is, it does not have to be at the left periphery, but even then, it displays vocative case and vocative distribution.

Other honorific and title-denoting nominals, such as doktor ‘doctor’, also do not display variable agreement, as seen in (19). Finally, the structure cannot be an appositive with a pro because in comparable appositive constructions the pronoun must be overt, and predicates, instead of displaying variable agreement, obligatorily agree with the features of the pronoun, as in (20).

Finally, in very colloquial contexts when the subject is salient, the interlocutor version of pan can be dropped, while retaining the 3^rd person agreement. For example, (21) is reported as felicitous in a context when the interlocutor is part of a discourse situation in which someone suggests inquiring about cigarettes.Footnote ¹⁹ We believe these examples constitute a case of pro-drop when a subject with pronominal properties can remain unpronounced (Polish is a canonical pro-drop language), thus further highlighting the pronominal dimension of the structures based on pan.

To summarize, pan is unlike other R-expressions in that it denotes an interlocutor and the interlocutor interpretation is accompanied by variable agreement. However, pan is also unlike a pronoun in that it can function both as an NP and a DP modifier and it can be the root of a regular R-expression. Table 1 summarizes the syntactic distribution of pan.

Table 1: A summary of the properties of PAN

The data pattern presented in this section raises several questions. First, are the three syntactic distributions discussed in section 2 based on one or more representations stored in the lexicon? Second, what mechanism underlies the variable agreement attested with the interlocutor uses?

In the next section, we will argue that despite syntactic variability, all the syntactic uses of pan are based on a single representation in the lexicon and further, that this representation does not differ significantly from other regular lexical nouns in Polish. Furthermore, we will argue that different uses correspond to different sizes of syntactic structure being projected from the same root, and that the critical variation in agreement patterns only arises at spell-out.

3.1 Feature manipulation at the interfaces

Work on the execution of phase theory (Chomsky Reference Chomsky and Kenstowicz2001 and subsequent work) has emphasized the role of the interfaces in the process. It has raised the theoretical possibility that the interfaces may alter feature composition, and the valuation of narrow syntax features, at the syntax-morphology interface. For example, in the Distributed Morphology approaches to late insertion of roots, root insertion effectively resets the zero features of the root position. Works that utilize some notion of post-syntactic agreement (e.g., Bobaljik Reference Bobaljik, Harbour, Adger and Béjar2008, Arregi and Nevins Reference Arregi and Nevins2012, and Norris Reference Norris2014) rely on morphological feature copying. When we adopt the conception of spell-out as a derivational window during which interface operations can affect the feature composition of heads, we expect features to be altered not only at the syntax-morphology interface but also at the syntax-semantics interface. We argue that the syntactic variation attested in the different uses of pan stems from such interface feature interactions. That is, when pan projects a DP structure, unvalued syntactic features on D as a phase head can be altered by the syntax-semantics interface, as shown in Figure 1.

Figure 1: Phase head features can be targeted by the CI-interface

Crucially, as shown in Figure 2, morphological spell-out targets only the complement of a phase head. That is, the edge of the phase – the locus of features targeted by the syntax-semantics interface at Transfer – remains accessible to subsequent syntactic operations and can be utilized in post-syntactic operations targeting structurally higher syntactic structures.

Figure 2: Interface impacted features remain in the derivation

3.2 How to derive different pans

In this section we provide detailed derivations for the different uses of pan, focusing on the distinction between the R-expression and the addressee-oriented use. The grammar fragment is constructed to highlight which parts of the derivations drive the distinct uses and in what structural configuration such a variation can come about. The first steps of the derivation are the same for all uses of pan, including the modification use.

For concreteness we assume that the first step of the derivation creates a derivational workspace, that is, an empty set as a placeholder for late insertion of a root (see, e.g., De Belder and van Craenenbroeck Reference De Belder and van Craenenbroeck2015 and Kučerová and Szczegielniak Reference Kučerová and Szczegielniak2019). This placeholder merges with a nominalizing functional head (n) which creates a nominal root structure. We assume, following much work on gender (see, e.g., Kramer Reference Kramer2015 and Kučerová Reference Kučerová2018) that n contains an unvalued number and gender feature. These unvalued syntactic features project in syntax, creating a nP structure. In languages like Polish, gender is an inherent property of a root. For concreteness, we implement this inherent property of a root as an indexical feature on the root, following Acquaviva (Reference Acquaviva2014).Footnote ²⁴ This indexical feature is mapped onto n (and any other instances of gender that enter into an Agree chain with n) when the root is late inserted. The first derivational steps are exemplified in Figure 3.

Figure 3: Indexical gender on the root values n

For concreteness, we assume that the valued number feature is merged as a separate functional head (Num, as in Ritter Reference Ritter1993, Reference Ritter1995), although nothing in the proposed derivation hinges on this derivational decision. The number feature could have been part of n, or associated with the root.Footnote ²⁵ By Agree, as matching and valuation, the unvalued number feature on n gets valued by the valued number feature on Num. This part of the derivation is schematized in Figure 4.

Figure 4: Num merged as a valued number-feature

Finally, the D head is merged in the structure (Figure 5). We assume that D is a bundle of unvalued ϕ-features and an unvalued person feature (Ritter Reference Ritter1995, Béjar and Rezac Reference Béjar and Rezac2003).Footnote ²⁶ The unvalued ϕ-features trigger Agree with the valued number feature on Num and the gender feature on n. By matching and valuation, the ϕ-features on D get valued (with the number feature getting valued immediately, and gender only as part of late insertion of the indexical root).

Figure 5: D merged as a bundle of unvalued ϕ-features and an unvalued person feature

Crucially, the person feature on D remains unvalued. Since D is a phase head, the structure is spelled out with an unvalued person feature. As we outlined in the opening of this section, we adopt a model of grammatical architecture where spell-out creates a derivational window during which interface operations can impact feature composition of phase heads (see, for example, Kučerová Reference Kučerová2018, Reference Kučerová2019, Reference Kučerová, van Craenenbroeck, Pots and Temmerman2020 for further discussion). In the case of nominals, this amounts to the feature composition of D. Specifically, we argue that the syntax-semantics interface (CI, using Chomsky's terminology) can manipulate unvalued features on D. This reasoning has a precedent in Distributed Morphology approaches where, for example, root insertion resets an empty set to a valued set, or morphological operations reset unvalued syntactic features to a morphological default. In our case, since only the person feature on D remains unvalued – that is, it does not receive any value from narrow syntax – the syntax-semantics is able to provide a semantically-licensed value, as schematized in Figure 6. Alternatively, the syntax-morphology interface realizes the person feature as a default.

Figure 6: Person feature on D at the interface

The rationale behind having the person feature interact with the syntax-semantics interface is based on two insights: (i) person is a syntactic feature that requires semantic licensing, and (ii) since the phase head remains available to subsequent syntactic computations, features at the edge of the phase are accessible to the syntactic derivation even after the corresponding phase has been transferred to the syntax-semantics interface.

To elaborate on the first point: there is a line of syntactic work that proposes that person is a special syntactic feature in that it requires syntax-semantics licensing for purposes of event and participant anchoring (see, e.g., Ritter and Wiltschko Reference Ritter and Wiltschko2014, Zubizarreta and Pancheva Reference Zubizarreta and Pancheva2017, Pancheva and Zubizarreta Reference Pancheva and Zubizarreta2018). Under this view, syntax-semantics licensing is a derivational process that maps a purely syntactic feature onto semantic structures such as [±participant, ±speaker].Footnote ²⁷ That is, the person feature has the potential to be altered by the syntax-semantics interface, and in turn, to become the locus of agreement and interpretive variation attested with pan.Footnote ²⁸ To further elaborate on the second point, the theory states that spell-out is in its nature asynchronic. The theory assumes that the part of the structure that is sent to morphology (i.e., the complement of the phase head) is no longer accessible to further syntactic operations, but the edge of the phase remains in the derivation, even when the whole phase has been transferred to the syntax-semantics interface. This point is typically not made explicit in current work on phase theory. However, it is an inherent consequence of the assumption that phases are semantically complete units, but only the complement of the phase head is sent to the syntax-morphology interface. Consequently, features at the edge of the phase can be altered by the syntax-semantics interface and remain accessible to further syntactic derivation because they are not spelled out immediately. This allows, in turn, for the person feature to become a potential locus of agreement variation without necessarily yielding differences in semantic interpretation, simply because it is located at the phase edge.

To summarize the critical point: the person feature remains unvalued in narrow syntax. When the feature reaches the interfaces, the syntax-semantics interface can link the narrow syntax feature to a [±participant, ±speaker] representation. The syntax-morphology interface can then either refer to the semantically informed value of the person feature, or it can map the unvalued syntactic feature onto a default morphological realization.

With this background assumption clearly spelled out, we can now return to the derivations. As demonstrated in Figure 5, after D is merged, the narrow syntax derivation returns a structure with the gender and number feature valued, but the person feature unvalued. At spell-out, the syntax-semantics interface associates the unvalued syntactic feature with a [±participant] semantic feature, where the [−participant] value will return the R-expression use of pan, and the [+participant] output corresponds to the addressee type of pan. Let us first consider the derivation of pan as an R-expression, as in (1), repeated below as (22).

In this case, as schematized in Figure 7 for tej pani ‘this lady’, the unvalued person feature from narrow syntax is associated with the [−participant] semantic value. When the DP enters an agree relation with a predicate, the morphology component can either refer to the unvalued syntactic default, or it can take into account the semantically informed value. In both cases, the output converges on what is descriptively characterized as 3^rd person agreement.

Figure 7: tej pani ‘this lady’

The case that provides insight into the question of the categorization of pronouns versus R-expressions arises when the unvalued person feature is associated with the [+participant] value. In such environments, we observe an agreement alternation between the expected 3^rd person agreement and the structurally unexpected 2^nd person agreement, as in (13), repeated below as (23). Recall that the predicate agrees with the structural subject, that is, pan in nominative, not the vocative.

How does this alternation arise? After the syntax-semantics interface associates the syntactically unvalued person feature with the [+participant] value, more specifically [+participant, −speaker], the complement of the phase head is spelled out, but the edge of the phase remains in the derivation. After the DP combines with the rest of the structure and the predicate (for concreteness, the T head) probes for the closest DP with matching ϕ-features, an agree link between T and DP is created. After the corresponding part of the structure is sent to the syntax-semantics interface, the morphology component matches the accessible DP features onto the predicate. However, unlike in the case of the R-expression use of pan, morphology can return two different values. When morphology targets the unvalued syntactic person feature, the person feature gets realized as the morphological default, that is, 3^rd person agreement. When morphology targets the syntax-semantics interface value, it returns the 2^nd person agreement as the best match for the [+participant, −speaker] value.Footnote ²⁹ The two agreement options are schematized in Figure 8.

Figure 8: pan 2^nd/3^rd person alternation

There remains one important aspect of the interpretive and syntactic properties of the addressee-type of pan to be addressed, namely, that pan can only be valued as [−speaker], and not as [+speaker]. We argue that the restriction has to do with the politeness nature of pan. According to Portner et al. Reference Portner, Pak and Zanuttini2019, politeness marking has two separate components: one is syntactic licensing,Footnote ³⁰ the other one is pragmatic licensing. The idea behind pragmatic licensing is that it maps a social distancing interpretation onto a context update that is anchored to a discourse participant structure.Footnote ³¹ For the politeness interpretation to go through, it is necessary for the inherently present speaker-anchoring to be spatially, and in turn socially, distant from the participant value encoded in the representation of pan. While social and spatial distancing between the speaker and the addressee is possible, a social and spatial distancing within a speaker is nonsensical.Footnote ³²

4.1 pan in structures without a D projection

The previous discussion highlighted the role of a person feature and its corresponding role in the interlocutor interpretations associated with pan. The consequence is that when D is not projected, no interlocutor interpretation is possible. As we have seen in (5), repeated below as (24), pan can also be used as a nominal modifier.

Unlike other prenominal modifiers in Polish, pan must be adjacent to the noun it modifies and it must agree with its ϕ-features, as seen in (6), repeated below as (25).

We propose that these modification uses are based on smaller nominal projections. Namely, in this case, pan corresponds to nP, and it adjoins in Spec, nP of the noun it modifies. We argue that a local relation with the modified noun is necessary for the semantic component to associate the politeness interpretation with the nominal. In turn, the syntactic position accounts for the adjacency restriction, as well as the modifier-like interpretation.

As for pan's ϕ-feature properties, for concreteness, we assume that the ϕ-feature valuation arises via concord.Footnote ³³ Such a concord operation effectively mutes the indexical gender features otherwise associated with the root of pan.Footnote ³⁴ Technically, we follow Norris (Reference Norris2014) in that concord is a result of two processes: syntax creates a feature percolation structure, and morphology then copies values onto the percolated features. We slightly modify Norris's account by assuming that for us, roots are inserted late. When morphological insertion targets the concord structure as a trigger for feature copying (see Norris Reference Norris2014 for a technical implementation), morphologically realized features are fully determined by the concord structure, instead of the indexical gender on the late inserted root.Footnote ³⁵ A schematic version of the derivation is given in Figure 9.

Figure 9: pani Dyrektorka pan.f.sg director.f.sg ‘Madame Director’

Another place where we observe nominal structures lacking a D projection to be incapable of providing an interlocutor interpretation is the use of pan as a possessor.Footnote ³⁶ In Polish, possessors are adjectives morpho-syntactically. That is, when pan is merged as a possessor it is embedded in an adjectival projection. Since there is no D in the structure, these uses never yield an interlocutor interpretation, unlike structures with possessive pronouns, as demonstrated in (26).

4.2 Left periphery feature valuation as an alternative?

The present analysis ties the interlocutor type of pan with the phasehood status of DPs. The analysis thus structurally differs from proposals that argue for a left-periphery functional projection that encodes the speaker/addressee orientation as a feature that must enter into an Agree relation with a person feature on a DP. Here we compare our own proposal with two influential proposals that argue for the presence of a left-periphery functional projection that enters into an Agree-like relation with the relevant DP, namely, the proposals spelled out in Sigurðsson (Reference Sigurðsson2004) and Portner et al. (Reference Portner, Pak and Zanuttini2019).

Sigurðsson (Reference Sigurðsson2004) argues for a [±speaker] feature introduced at the left periphery, which enters into an Agree-like relation with DPs within the corresponding clause. The proposal assumes that person-related features are logophoric, that is, they are not subject to locality restrictions. Further, the agree relation is selective: if there is more than one pronoun to be bound by the left periphery, these distinct pronouns are associated with an index that selectively identifies them for the purposes of Agree with the left-peripheral person representation. That is to say, the proposal addresses the question of how a pronominal, as well as a pronominal-like structure, gets bound by a [±speaker] feature, but it does not address the question of how a full lexical NP can obtain a pronominal-like interpretation. For the sake of argument, let us assume that once the unvalued person feature is established in the syntactic structure, the association of the person feature with a [±speaker] feature takes place only at the level of syntactic association with the left periphery, as opposed to the level of spell-out of the DP itself, as argued in our proposal. First of all, it is not immediately clear how such an account could distinguish between the interlocutor use of pan, which can trigger a subject-predicate agreement, and structures such as imposters that also obtain an interlocutor interpretation, but whose association with the interlocutor speaker does not translate into agreement relations.Footnote ³⁷

The issue is more general and points to the main difference between our account and that of Sigurðsson (Reference Sigurðsson2004). For Sigurðsson, the timing of structural relations is irrelevant because the interlocutor association takes place only at the very end of the derivation. Moreover, the phase status of the DP itself is also irrelevant. In order to account for the dual agreement pattern observed with the interlocutor type of pan, agreement would need to be established only after the complete clause, including its left periphery, has been spelled out. Furthermore, because the DP would in such cases be uniquely associated with an addressee interpretation, it is not clear how and why this dual agreement arises. That is to say, Sigurðsson (Reference Sigurðsson2004) might be a well-suited account for modelling an association of pronouns with interpretive [±speaker] features. However, his proposal does not extend to the type of person feature interactions we observe with pan.

Portner et al. (Reference Portner, Pak and Zanuttini2019) also assume that there is a designated functional projection at the left periphery that enters into an Agree relation with the relevant DP. In their proposal, the relevant feature pertains to politeness, not person per se. Crucially, this feature c strictly obeys locality, that is, in this respect it is different from the logophoric person feature of Sigurðsson (Reference Sigurðsson2004). Consequently, because c needs to enter into an Agree relation with the cP projection at the periphery, the agree relation is subject to intervention effects. As Portner et al. (Reference Portner, Pak and Zanuttini2019) report, intervention effects are indeed attested with the politeness nominals they investigated. They provide an example from French when an honorific title, grammatically a 3^rd person nominal, can in and of itself obtain a politeness interpretation via agree with cP. However, a structurally lower element (a bound pronoun in their case) must be bound by the grammatical 3^rd person nominal because the nominal behaves as an intervener between cP and the pronoun. Their original example (Portner et al. Reference Portner, Pak and Zanuttini2019: 25, ex. (53)) is given in (27).

This intervention configuration is exactly parallel to the configurations in which the interlocutor version of pan triggers 2^nd person agreement. If the interlocutor valuation of pan was indeed based on a syntactic Agree relation with a left-peripheral functional projection, the Agree relation between the subject and the predicate should be subject to the same intervention facts as the French example in (27). However, as we have seen throughout this article, pan is rather different. Its feature valuation is subject to locality restrictions, but the valuation is localized at the phasehood of a DP instead of being based on Agree with the left periphery.