At first sight, PNs seem to refer to objects: the name Susan simply points to a referent and does not provide a description of that referent (cf. Mill 1843/1973; Kaplan 1989; Jeshion 2015; Saab & Lo Guercio 2018, among others). This reference remains constant across possible worlds, and for this reason, PNs are singular, rigid designators that do not interact with quantified items (cf. Kripke 1980: 1971). In this sense, they contrast with common nouns (CNs). However, PNs in examples like (1)a–b, from Burge (1973), and (2), quoted by Matushansky (2008: 598), do not seem to be rigid designators: they appear with overt quantifiers and look more like common nouns (CNs).

For this reason, other researchers have suggested that PNs are predicates involving naming: Ernestico means ‘the ‘X’ named ‘Ernestico’’ (cf. Kneale 1962; Burge 1973; Geurts 1997a; b; Matushansky 2008; Gray 2012, among others). In Burge’s (1973: 428) words: “a proper name is a predicate true of an object if and only if the object is given that name in an appropriate way.” Although these approaches share the notion of a naming predicate, they differ in important ways, such as whether PNs are rigid designators or not.1

Proponents of the referential analysis of PNs note that PNs can be used in contexts where the predicative analysis is not so straightforward (cf. Jeshion 2015). Andrés Saab (p.c.) notes examples such as (3), from Saab & Lo Guercio (2018).

These examples involve deferred interpretations (Jeshion 2015), in the sense that the literal referent of the PN (the painter Rembrandt in (3)a) is connected to the object by some meaning extension (‘the painting by Rembrandt’). Crucially, however, these examples do not involve a direct naming function, so that (3)a in the deferred interpretation does not mean ‘I saw the object named Rembrandt in the museum’. We will return to these examples below.

PNs also raise important syntactic questions. First, they tend to be bare singulars in languages that do not generally allow bare singulars, as illustrated in (4)a vs. (4)b–c. Whereas Ana can be bare in the first example, hermana ‘sister’ cannot appear without a determiner in the second one.

Longobardi (1994) has suggested that this is an illusory contrast: both (4)a and (4)c involve a DP, but (4)a has obligatory N-to-D raising, whereas (4)c has an overt determiner. N-to-D raising results from an independent requirement that arguments be DPs headed by an overtly realized D head. This head may be overt or it may be realized as the raised PN. Longobardi mentions languages with overt PN determiners as evidence for his analysis. For example, in Greek and in Chilean Spanish, the definite determiner appears with PNs, as seen in (5). In this case, the determiner is expletive, in the sense that it is only required to fill the D position, but it does not contribute additional meaning.

The notion that D is an expletive would explain languages where it is optionally present, or restricted to a certain gender. For example, in Castilian Spanish, el ‘the.MAS’ tends to be optional, but la ‘the.FEM’ tends to be more frequently overt, as seen in (6). In addition to Castilian Spanish, oral registers of Latin American Spanish, as well as the regional Italian spoken in Trentino (Jan Casalicchio, p.c.) and several other Gallo-Romance varieties have this pattern, as noted by an anonymous reviewer.

A third set of languages have specialized determiners for PNs as seen in (7)–(8). In (7), the genitive marker is different for buhay ‘life’ (genitive: ng) than for Maria (specialized genitive ni) in Tagalog.

In Balearic Catalan, the PN determiner is en/na ‘D.MAS/D.FEM’ as illustrated in (8)a–b, from Bernstein, Ordóñez & Roca (2016). The CN definite determiner in this variety is es/sa.

Languages that have a determiner with PNs tend to use the definite one if they do not use a specialized one (cf. Chierchia 1998: 397; Matushansky 2008: 593). Chierchia specifically states: “if a language assigns a proper name to the semantic category pred, its only option for turning the proper name into an argument will be to project the category D. Furthermore, the only choice of determiner is the.” This generalization is stated in (9) and illustrated with the Spanish examples in (10) and the Greek examples in (11). The alternative to drafting the definite determiner for PNs is to have a specialized PN determiner, as we above saw for Catalan and Tagalog.

To the extent that (9) holds, it is generally related to the meaning of PNs or the meaning of DPs containing PNs. Thus, for Ghomeshi & Massam (2009), DPs with a PN are headed by D_{DEF, PROPER}. Consequently, in Chilean Spanish, the morphological exponents of this determiner would be el or la depending on gender.

Similar explanations for (9) follow from the properties of PNs in Borer’s (2005) and Matushansky’s (2006) analyses, although they rely on very different overall assumptions. For Borer, insertion of a PN determiner follows from a <def-u> feature in D, which is definite (in addition to having a specific kind of content associated with PN interpretation). In Matushansky’s proposal, the type of determiner depends on language-specific morphological spellout rules, but the determiner is specified as D_DEF.

The generalization in (9), and the principles that derive it suggest a connection between the nature of PNs and definiteness. On the other hand, Longobardi (1994; 2005; 2008) analyzes PN determiners as expletive, in the sense that they do not seem to contribute any semantic content that is not already present in the PN.2 Thus, the optionally of determiners in varieties such as the one illustrated in (6) above does not seem to correlate with obvious semantic differences. As a result, we seem to have the paradox illustrated in (12).

The analyses proposed by Borer, by Ghomeshi & Massam and by Matushansky build definiteness into the PN determiner, explaining why it often overlaps with the regular definite determiner, and account for its apparent optionality through morphological rules, whereas Longobardi removes any content from D_PN, and justifies its presence for independent reasons, namely for the requirement that arguments must be DPs.

A third related issue addresses the licensing of PNs and DPs in general. In Longobardi’s (2008) and Bernstein’s (2008) analyses, N raises to D attracted by a [person] feature. Chierchia (1998) and more recently Longobardi (2008) point out that nouns must appear with an overt determiner in Italian, but pronouns void this requirement, as seen in (13). (13)a presents a sentence with a regular D (i, ‘the’), and (13)b shows one with the pronoun noi ‘we’. As (13)c shows, it is not possible to have both at the same time. This leads him to conclude that “the category D minimally consists of the person feature” (Longobardi 2008: 200).

In sum, the three questions in the literature that frame this paper are the following: first, is there a naming function as part of the meaning of PNs? Second, is the definite determiner the result of the PN’s meaning? Third, what feature licenses PNs?

The determiner-like phrase un tal is specialized for PNs, as illustrated in the following examples (from the Spanish corpus CREA, cf. Real Academia Española):

Un tal ‘a certain’ has the following properties: 1) it only appears with PNs, 2) the PN is animate/human, 3) un tal ‘X’ is singular, 4) a prenominal adjective is not possible, 5) un tal ‘X’ does not have the properties of an indefinite, 6) it is not interpreted as definite and 7) tal conveys an evidential (‘alleged’) or evaluative, distancing meaning.4

The first property is illustrated in (15), where CNs are ungrammatical:

A corpus search of CREA for the sequence un tal yielded 934 cases. Of the first 300 consulted, all of the ones with the relevant meaning had a PN except for the one in (16). This example includes a nominalized relative clause in place of the PN, but it is interpreted as if it were a PN, as in the English phrase what’s-his/her-name.

Un tal cannot appear in the deferred interpretation contexts presented in (3), as seen in (17). The first example can only have the literal PN interpretation ‘an ‘X’ allegedly called ‘Y’’. The other two are ungrammatical because they do not fulfil some of the other properties: (17)b applies to a plural and (17)c applies to an inanimate object.

We can draw the conclusion from this paradigm that deferred interpretation cases do not involve literal uses of PNs (cf. Fara 2015).

The second property states that only human/animate PNs can appear with un tal, as illustrated in (18): river names, cities and towns do not appear with un tal ‘a certain’, regardless of the syntactic structure of the name of the locality. In (18)b, for example, Tunja is a bare NP, whereas La Unión is a DP with a definite determiner in (18)c, and Aguascalientes is a plural NP in (18)d. By contrast, a PN like Chu-Lín that refers to a famous zoo panda, is acceptable, as seen in (19).

All of the grammatical examples presented so far illustrate that un tal appears with a singular PN, the third property. The example in (20)a shows that a plural PN is ungrammatical with unos tales, unlike the counterpart without tal shown in (20)b.

The fourth property states that prenominal adjectives are disallowed with un tal ‘X’. For example, simpático ‘charming’ or supuesto ‘alleged’ cannot appear prenominally, as seen in (21). This property is shared with bare PNs, which cannot be preceded by an adjective either, as seen in (22) (cf. Longobardi 1994).

The last three properties we have just seen strongly suggest that un tal ‘X’ behaves like a regular PN: PNs are singular when they have rigid scope, they are typically animate, with certain exceptions like geographical place names, and they do not allow prenominal adjectives.

Property five states that un tal ‘X’ does not have the typical properties of indefinites. In order to illustrate this, I note that PNs can sometimes be interpreted as CNs, and in such contexts, they interact with quantifiers. For example, Geurts (1997b: 323) points out that in (23), the modal takes scope over Leslie, so that this sentence can be paraphrased as ‘it may be the case that a person named ‘Leslie’ is a man or a woman’. In this particular context, Leslie loses the scope rigidity usually associated with PNs, although as an anonymous reviewer points out, this use may refer to the name Leslie itself, rather than to anyone named Leslie.

Un tal ‘X’ does not behave in the same way in similar contexts, even though it has an overtly indefinite determiner. For example, the modal cannot take scope over un tal Adrián in (24), which is not interpreted as a semantic indefinite.

This sentence is ungrammatical presumably because un tal ‘X’ refers to an actual person named Adrián, and not to any person who may bear that name.

Similarly, note that it is not possible to add u otro ‘or another’, or cualquiera ‘whatsoever’ to un tal ‘X’, as seen in (25)a–(26)a, whereas the counterparts of those examples without tal are possible (cf. (25)b–(26)b). These facts taken together strongly suggest that despite having an overt indefinite determiner, un tal ‘X’ is not interpreted as an indefinite DP.

The sixth property is somewhat surprising as well: despite designating a single, unique individual, un tal ‘X’ can appear in existential constructions (cf. (27)a), unlike bare PNs, seen in (27)b. PNs with just an indefinite article pattern in the same way as un tal ‘X’ (cf. (27)c).

Similarly, un tal patterns with bare un in the context of small clauses with a verb of naming, seen in (28) (cf. Matushansky 2008). In those environments, un tal ‘X’ becomes ungrammatical, as seen in (29)a, just like a regular indefinite article (cf. (29)b).5

In sum, un tal ‘X’ shares a similar distribution to bare PNs, in particular it shows no scope interactions (unlike indefinite PNs), but it is not interpreted as a bare PN in existential contexts. In several other contexts, however, it patterns the same way as a bare indefinite un X. In the following section, we describe the semantic contributions of tal.

In this section, I review the two possible meanings that tal contributes to un tal ‘X’.

Although it is possible to have definite el tal ‘X’, in that case, two things happen: the meaning tends to be much more clearly evaluative rather than evidential, and several of the distributional/interpretive restrictions above disappear. To begin with, el tal ‘the so-called’ can appear with CNs (cf. (42)), with inanimate nouns (cf. (43)), and with plural nouns (cf. (44)):

If the noun is a CN, then prenominal adjectives are more acceptable, as shown in (45).

As these examples show, el tal ‘X’ becomes a definite description that refers to an antecedent in discourse, and in this sense it is definite. Consequently, it cannot appear in existential contexts, as seen in (46).

The fact that the changes in the distribution of el tal ‘X’ correlate with a change in the meaning of tal suggests that evidentiality is crucial for the PN-like properties of un tal ‘X’.

The properties of the alternation between un/el tal X are summarized in (47).9

To summarize the relevant issues, the properties of un tal challenge the generalization in (9), which states that PN determiners tend to be definite if they are not specialized, and also the paradox in (12), which describes two competing views of D_PN as either definite or expletive. Evidential tal cannot appear with a definite determiner and must appear with the indefinite. Conversely, evaluative tal can appear with the definite determiner, and is not restricted to PNs.

PNs are generally interpreted as definite and, for this reason, they tend to appear with a definite determiner (cf. (12)a above). Two properties have been identified with definiteness in the literature: familiarity and uniqueness, among others. Heim (1982; 1983) argues that definiteness involves familiarity in the sense of presupposing an existing referent in discourse. Thus, if a speaker wants to refer to an individual named Ana, she automatically presupposes that the individual named Ana exits in discourse. The existence presupposition is not cancellable for PNs, unlike with ordinary definite descriptions: whereas the current king of France is false because it has no referent, Ana automatically presupposes an individual named ‘Ana’.

In addition to familiarity, definiteness induces uniqueness and maximality (cf. Chierchia 1998, among others). Specifically, Chierchia (1998: 346) derives uniqueness and maximality by generalizing the iota operator ‘ɩ’: when applied to a set of singularities, the operator “will yield a result only when the predicate has just one object in its extension.”

Next, we turn the contribution of evidentiality, and later to the interaction between (in)definiteness and evidentiality.

As mentioned above, un tal ‘X’ has two possible meanings, evidential and evaluative. We argue that the evidential meaning, in particular, is possible with PNs because it operates on the naming content of the PN. This means that the NP predicate Ernestico has the meaning {ɩX such that X is named ‘/ernestiko/’} and the evidential provides information about the source of this naming, yielding {ɩX such that X allegedly named ‘/ernestiko/’}. This account explains why un tal is not compatible with CNs: since a regular CN does not involve being named anything, the evidential cannot target that semantic content.

I adopt a view of illocutionary evidentials as contributing pragmatic and presuppositional content (cf. Potts 2005, among others). Specifically, if we assume that discourse participants share a common ground that contains the set of shared presuppositions, the role of evidentials is to restrict that common ground through not-at-issue content (cf. Murray 2010; 2011). Murray (2010: 94–5) illustrates this point with the Cheyenne direct evidential shown in (48)a. This sentence has the at-issue, propositional content ‘Floyd won’, similar to the translation in English, and not-at-issue content indicating that the speaker has direct experience about the at-issue content. Whereas at-issue content can be denied and is negotiable by saying ‘no, he didn’t win’, not-at-issue content is non-negotiable and cannot be denied, so that the continuation ‘you didn’t witness it’ is not felicitous. A similar account would apply to the hearsay evidential illustrated in (48)b.

Not-at-issue content is asserted and directly added to the common ground. For example, when a speaker uses the reportative evidential in (48)b, the initial common ground is the one represented in the left figure in (49). This common ground includes whatever presuppositions and implicatures the discourse participants initially share. Once the statement in (48)b is produced, reportative evidential –sėstse ‘I hear’, represented as RPT, automatically updates the common ground to those situations in which the speaker heard the content ‘Floyd won’. This update is represented as the intersection portion of the center figure in (49). If the listener accepts the propositional content ‘Floyd won’, then the common ground is further updated, as indicated by the darkest intersection in the right-most diagram in (49). This new common ground is restricted to the situations in which Floyd won, and the speaker heard that information.

However, if the speaker does not accept the propositional content, for example, if she replies: ‘No, she didn’t win’, then the common ground will not be updated with at-issue content. Thus, common ground updates can be automatic (as in the case of non-negotiable, not-at-issue evidential updates) or negotiable (as in the case of at-issue content updates).

The extension of this proposal to the case of un tal ‘X’ would be as follows: imagine the discourse in (50). The initial discourse context at the time of the embedded temporal clause in bold, is represented in the left-hand diagram in (51), and it includes the common ground updated by ‘we were sitting at a restaurant’.

Un tal automatically updates the common ground to the situations where the speaker heard that ‘someone is called ‘María’’, as shown in the middle diagram of (51) (recall that RPT indicates reported information). Finally, the proposed at-issue update would be that ‘someone (who is called María) arrived’, as in the right-hand diagram of (51). As before, the interlocutor can deny this last proposed update, for example by saying: no, era Marta ‘no, it was Marta’, or no, no llegó ‘no, she didn’t arrive’, but the reported source of María’s name cannot be questioned.

To sum up, first, tal represents an evidential operator, second, it takes structural scope over the naming predicate introduced by the PN; third, the evidential content is not-at-issue and it automatically updates and restricts the common ground. Because it is not-at-issue, the source of evidence cannot be contradicted.

We are now in a position to derive the incompatibility between the definite determiner and evidential tal. In the preceding sections, I assumed that PNs must be unique and familiar, which results in using the definite determiner in the default case (more on this below). I also proposed that evidential tal contributes not-at-issue content that automatically updates the common ground. In this section, I will suggest that evidential content is incompatible with the familiarity presupposition.

Recall the generalizations established in section 2.3: whenever the determiner is definite, the meaning of tal ‘X’ is evaluative, not evidential, as shown again in (52).

Let us see how the common ground would be updated in those cases. Since the definite determiner indicates that the NP is familiar, this means that the proposed common ground update will only include contexts in which the referent of the PN is familiar. However, in those contexts in which the speaker is familiar with the referent of the DP, she should also have evidence as to the person’s name. In other words, the initial common ground already includes Ernestico, and therefore, his established name. When the speaker uses tal as an evidential, the common ground is automatically updated only to those situations in which the speaker heard that X is called Ernestico, as represented in (53). However, this newly updated common ground does not share anything in common with the familiarity presupposition encoded in the PN, so the final common ground cannot be successfully updated.

To put it slightly differently, because the evidential, not-at-issue content involves an automatic update, it prevails over the familiarity presupposition. The only available alternative in this case is to eliminate the familiarity presupposition, namely to use an indefinite determiner. Conversely, the definite determiner becomes possible when tal is interpreted as an evaluative, rather than as an evidential (cf. (52)). In this case, the evaluative is not evidential, although I have suggested that it is also not-at-issue. To account for the difference between evidentiality and evaluation, I offer two tentative possibilities: first, the evaluative not-at-issue content does not automatically update the common ground (and is therefore subject to negotiation, unlike evidential content), or, alternatively, if the evaluative common ground update is automatic, that evaluative content does not conflict with familiarity. I leave this issue open for further research.10

One immediate consequence of the analysis just presented is that definiteness/familiarity cannot be the only property licensing PNs. Rather, definiteness/familiarity is the default value, given the normal conditions in which a PN appears. These conditions can be systematically altered, as we saw with evidentiality. The familiarity presupposition is cancelled due to the presence of evidential content, therefore the definite determiner does not appear. However, this conclusion raises the question of why PNs should yield a familiar presupposition in the general case. As mentioned in section 1.2, Bernstein (2008) and Longobardi (2008) propose that DPs are licensed by a [person] feature in D, based on the complementary distribution of pronouns and determiners in certain contexts in Italian (cf. (13) above, repeated as (54)).

However, this paradigm does not extend to Spanish, which lacks a pronominal determiner in similar cases, as discussed in Jelinek (1984); Ordóñez & Treviño (1999) and Camacho (2013), see (55).

This contrast between Spanish and Italian suggests that the syntactic location of [person] may vary across languages (D in Italian, a higher head in Spanish). Be that as it may, it is also true that pronominal determiners are generally restricted to plural contexts (cf. (56)a), and for this reason, they are not productive with PNs (cf. (56)b–d). This suggests that [person] is not the only possible way of licensing D.

In order to see other possibilities, we turn to Speas’ (2004) account of evidentiality in connection to Harley & Ritter’s (2002) feature hierarchy for referential expressions. Speas (2004) proposes that evidentials are governed by constraints similar to those that govern referential expressions. Specifically, she extends the feature hierarchy for referential expressions developed by Harley & Ritter (2002) to a modal base. This extension yields an evidential feature hierarchy, as illustrated in (57).

When the feature hierarchy applies to a referential base, the result is the hierarchy of referential expressions we see in (57)a, when it applies to a modal base, we obtain the evidential feature hierarchy in (57)b. As Speas (2004: 261) argues, “evidential morphemes spell out an agreement relation between the discourse and the world(s) in which the sentence is to be interpreted”. For example, a direct-evidence evidential will encode a modal base restricted to the [+SPEAKER] node in (57)b. Hearsay evidence, on the other hand, involves a modal base restricted to [INDIVIDUATED].

This proposal offers a principled connection between evidentiality and referentiality (cf. also Rooryck 2001) that will allow us to account for important parts of the distribution of un tal ‘X’. We can now reformulate Longobardi and Bernstein’s insights about the importance of person in licensing DPs, as in (58).

Let us assume that these features are hosted by a syntactic head ‘IND’, which will take a value from one of the nodes represented in the hierarchies in (57). For example, the DP el gato ‘the cat’ involves the value [INDIVID] for IND in the referential base in (57)a, whereas noi ricchi ‘we rich’ involves the value [PARTICIPANT] for that same node in that same base. In the case of un tal ‘X’, I propose that IND takes the value [INDIVID] from the modal base, as represented in the structure in (59). As in Longobardi’s (1994) analysis, the PN raises, but only to IND. This accounts for why no prenominal adjectives are possible. When IND is not spelled out as tal, as in Ana, the PN can raise further to D. The overt result is a bare PN. The partial spellout rules for D are presented in (60).11

Familiarity, in turn, is a feature that can optionally be associated with 3^rd person, individuated DPs, as seen in (61). In the first example, un peatón is indefinite (specific or non-specific). In the second one, it is definite/familiar.12

As I have argued earlier, PNs must be [INDIVID] and familiar, although the familiarity presupposition can be cancelled when [INDIVID] applies to a modal base.

An anonymous reviewer wonders whether the proposed analysis extends to other cases beyond un tal ‘X’. It is true that most DPs in Spanish, Italian or English are licensed via the referential base, not the modal one. However, some possibilities worth exploring involve epistemic adjectives, such as supuesto ‘alleged’, illustrated in (62), already discussed in section 2.2.1.

As suggested in that section, epistemic adjectives share some modal properties with un tal, but un tal has a source of evidence content that is not present in supuesto. This difference could be accounted for by proposing that supuesto applies to the root of the modal base in (57)b, whereas tal applies to the [IND] node. Following this idea, supuesto ‘alleged’ updates the common ground to those situations that are identical to the current one, but where the person who called is an impostor.

In other languages, evidentiality is more systematically expressed in nominals (cf. Imai 2003; Lecarme 2008). For example, according to Dixon (1972), nouns in Dyirbal appear with a noun marker with which they agree in case. This marker also indicates noun class and distance from the speaker, as illustrated in (63).

More importantly for our purposes, bayi ‘there.VIS’ and giyi ‘here.VIS’ indicate a visible referent, whereas ŋayi ‘there.NOT.VIS’ indicates a non-visible referent. Reference to visibility is also pervasive in other Austronesian languages like the Western Desert language (Wati) and in the Western Torres Strait language (Dixon 1972: 45). Thus, in terms of the hierarchy in (57), all three noun markers in Dyirbal would indicate distance with respect to the [+SPEAKER] node in the referential scale. Additionally, since they also indicate evidential content, they would also have values specified for the modal base: bayi ‘there.VIS’ and giyi ‘here.VIS’ indicate source of information, namely content related to the [+SPEAKER] node in the modal base, and ŋayi ‘there.NOT.VIS’ indicates source of information related to the [INDIVIDUATION] node in the modal base.

Danziger (1994) describes a similar pattern in the locative system of Mopan Mayan, which also involves divisions structured around participants and evidentiality (visible and non-visible): waye’ ‘deictic locative’ centered on the speaker, ta’kan ‘deictic locative’ centered on the interlocutor, tilo’ ‘deictic locative’ centered on a visible third person, and te’ ‘deictic locative’ measured around an non-visible third person’.

These paradigms from Dyirbal and Mopan Mayan provide further evidence of the interaction between the referential and the modal hierarchies in (57). They further illustrate the possibility that both bases simultaneously apply within the same nominal structure.

To summarize, I have proposed that PNs include a syntactic head IND whose semantic value [INDIVID] yields two possible results: when applied to the referential base, the PN is a person-licensed non-speaker, non-addressee, when it applies to the modal base, it is hearsay headed by tal. In the former case, the determiner is ∅ if N raises to D, or el/la if it does not. In the latter case, no raising is possible, and the determiner is un/una because hearsay is incompatible with the familiarity presupposition. I have also shown other cases in which the current analysis may be extended: epistemic adjectives like supuesto ‘alleged’, noun markers in Dyirbal and deictics in Mopan Mayan. Finally, these examples raise the possibility that both modal bases apply to the same nominal structure.