1 Introduction

The constructions exemplified by the Finnish and Russian sentences in (1) are cross-linguistically common.

The constructions in (1) have a plural pronoun accompanied by another NP joined to the pronoun by a connector, meaning roughly ‘with’. The connector phrase is a sort of an annex (or a satellite) to the pronoun. The two can typically be either disjoint (as in (1a)) or adjacent (as in (1b)), forming a phrase together in the latter case (see, e.g., Lichtenberk 2000; Vassilieva & Larsson 2005; Holmberg & Kurki 2019). A surprising hallmark of the construction is that the plural pronoun can have a singular reading. Constructions of this type are found in many languages “from families that include (but are not limited to) Afroasiatic (Chadic), Algic, Australian, Austronesian (Malayo-Polinesian), Indo-European (Baltic, Celtic, Germanic, Romance, Slavonic), Mayan, Niger-Congo, Nilo-Saharan, Salishan, Siouan, Trans-New Guinea and Uralo-Altaic” (Vassilieva & Larson 2005: 102). They have been referred to as Plural Pronoun Constructions (Schwartz 1988a, b; Aissen 1989; Vassilieva & Larson 2005; Cable 2017) or as Inclusory Constructions (Lichtenberk 2000; Singer 2001; Haspelmath 2007).¹ Neither term is entirely satisfactory (as also pointed out by Holmberg & Kurki 2019), so we opt for the theory-neutral Plural Pronoun[x-NP/NP-x] Construction instead, Pro[x-NP] for short, where x stands for the connector.²

According to Schwarts (1988a: 242), the languages in her sample with Pro[x-NP] that also have a case system make a case distinction between the pronoun and the annexed NP, “as if the connector were functioning as a case-assigning preposition”. In the Russian example in (1b), for instance, the pronoun my is nominative while Petej is in the instrumental. Such a case split is also seen in the absence of a distinct connector in examples like Hungarian Mi elmentünk Jancsival “we went.1PL Johnny-with” = ‘I left with Johnny’, where the pronoun mi ‘we’ is nominative, while Jancsival is in the instrumental (Moravcsik 2003: 480). We take this to support the view that case affixes in Finno-Ugric languages are commonly adpositions (Asbury 2010). Accordingly, we subsume the Hungarian type under Pro[x-NP] constructions (including postpositional Pro[NP-x]).

There is also a related but a much rarer construction that has neither a case split nor an overt connector, which is illustrated in (2).

We refer to this construction as Pro[NP]. It lacks an overt connector, thus having a plural pronoun and a directly juxtaposed or annexed NP. Symptomatically, though, it shares a basic property with Pro[x-NP] constructions: the plural pronoun commonly has a singular reading. We refer to this reading as the Pro[(x-)NP] reading. We take the singular reading of the plural pronoun to be the most interesting and the defining property of Pro[(x-)NP] constructions.

We know of only a handful of Pro[NP] languages and of no detailed descriptions of any such language, with the sole exception of the description of Toqabaqita in Lichtenberk (2000). The absence of a case split between the pronoun and the NP annex in the Yapese, Mokilese, and Kpelle examples in (2), as also in Toqabaqita, is unspectacular, as none of these languages have case systems. However, Icelandic is a Pro[NP] language with a case system, and in this language the Pro[NP] construction is always case congruent. We illustrate this in (3) (for only nominative við ‘we’).

As seen in (3), the construction is ambiguous between a singular and a plural reading of the plural pronoun (and the plural reading, in turn, is ambiguous in the same way as we in general: ‘I and you.SG/PL’, ‘I and they’, etc.). The unexpected singular reading is the more neutral and the much more common of the two. The plural reading is usually difficult to get and often far-fetched, but it can be forced in certain contexts, at least for some speakers (as in “Have you seen Peter_i recently? Yes, I_k saw him yesterday. We_i+k Olaf are all going to the movies together tonight”). However, many speakers cannot get the plural reading at all (instead requiring coordination, “we_i+k and Olof”).³ We will thus focus on the singular reading for the most part.⁴

Icelandic Pro[NP] has been sporadically mentioned in the literature (see, e.g., Sigurðsson 2006; Thráinsson 2007; Wood 2009). However, it has never been described or analyzed in detail; to our knowledge, no case congruent Pro[NP] constructions ever have. In this article we aim at providing both a description and a comprehensive analysis of the Icelandic Pro[NP] construction.

We connect our analysis to a broader understanding of the structure of plural pronouns and build on the idea that plural pronouns are constructed from sets of variables, {X, Y} (a key idea also pursued in Vassilieva & Larson 2005, although differently implemented; see also Holmberg & Kurki 2019). The primary X variable is fixed, as the speaker (X_SP), the addressee (X_AD), or a non-speaker/non-addressee (X_–SP/–AD): we = {X_SP, Y}, you.PL = {X_AD, Y}, and they = {X_–SP/–AD, Y}.⁵ The Y variable, in turn, is open to interpretation/reconstruction, depending on the context, but it can be constrained or specified in a number of ways. We assume that all plural pronouns have an NP annex, and that the specification of the Y variable is provided by the NP annex, silently in the case of bare plural pronouns, but overtly in Pro[NP]. Our approach is unorthodox, but we believe it is on the right track, and that it throws a new and an interesting light on the nature of plural pronouns.

What is special about the Icelandic Pro[NP] construction, then, is not the pronoun itself, but lies in the fact that the annex can be partly overt. Our analysis entails that pronouns do not have presyntactically specified semantics but spell out features whose interpretation is syntax- and context-dependent.⁶

The structure of the article is as follows. In Section 2, we provide a basic description of the Icelandic Pro[NP]. Section 3 discusses two constructions that are ostensibly similar to Pro[NP], the Preproprial Construction (“he/the Olaf”) and adnominal pronoun constructions like we linguists, and describes several ways in which they are distinct from Pro[NP]. In Section 4, we analyze the relation between the pronoun and the annex, proposing that it is an indirect head-complement relation, mediated by a silent element that heads the annex (bringing the construction (partly) in line with the Russian/Finnish type of Pro[x-NP]). We refer to this silent element as the lock. In Section 5, we defend our analysis of Pro[NP] and argue that it reveals more general properties of how plural pronouns are constructed and interpreted. Section 6 concludes.

2 Pro[NP] with singular annexes (including coordinated ones)

In this section we provide a fairly theory-neutral description of the Icelandic Pro[NP] construction with singular annexes (and a singular reading of the pronoun).

Pro[NP] is a productive and general construction in Icelandic. It is thus found for all three persons and all three genders, as illustrated in (4)–(6). As indicated in (5) and (6), third person plural pronouns in Pro[NP] normally require a linguistic antecedent (in contrast to the first and the second person pronouns). Pluralities of mixed third person genders are expressed by neuter, see (6).

In passing, we note that Pro[NP] is overwhelmingly more common in the language, in all persons and genders, than coordination of a name and a singular pronoun, such as ég og María ‘I and Mary’ and María og ég, as confirmed by searches (2019-10-15) in corpora: http://timarit.is/ and http://málheildir.árnastofnun.is. These are written language corpora, but, as far as we can tell, this also applies to the spoken language.

Case congruent Pro[NP] of this sort is found in all three persons in Old Norse as well,⁷ and for at least the third person in runic Swedish, as in þau kitilmuntr “they.N Ketilmund.M” = ‘Ketilmund and she’ (Wessén 1965: 60), suggesting that it is a Proto-Scandinavian or a Proto-Germanic inheritance. It has been completely lost in the modern Mainland Scandinavian languages (Danish, Norwegian, Swedish), but it is found in Faroese, the closest relative of Icelandic, albeit not as robustly as in Icelandic.⁸ We limit our description to Modern Icelandic.

Icelandic Pro[NP], then, is found for all persons and genders. It is also free to occur in all four cases (NOM, ACC, DAT, GEN) and in all grammatical functions, as subjects (nominative or quirky), direct objects, indirect objects, prepositional objects, possessives, or as dislocated DPs.⁹ This is partly illustrated in (7) (there is an ACC/DAT syncretism in the first person plural pronoun paradigm).

Pro[NP]s are also well-formed in ECM infinitives and as controllers of PRO (“she believed us Olaf be intelligent” = ‘she believed Olaf and me to be intelligent’, “we Olaf tried to win match-the” = ‘Olaf and I tried to win the match’).

The pronoun shows clear head properties in relation to the annex, controlling agreement at the clausal level. That is to say: the pronoun controls person (1, 2, 3) and number (PL) agreement of the finite verb, as seen in (4)–(6). In addition, it controls number (PL), gender (M, F, N), and case (NOM, ACC, DAT, GEN) agreement of adjectival and participial predicates. This is partly illustrated in (8).

As seen, the NP annex never affects or interferes with agreement. In addition, we note that plain first and second plural pronouns can be truth-functionally equivalent with Pro[NP]: plain við and þið can (obviously) refer to ‘Olaf and I/we’ and ‘Olaf and you’. Third person plural pronouns in Pro[NP], in contrast, normally require a linguistic antecedent, see (5) and (6). Given an appropriate context, though, plain þeir, þær, þau can of course refer to ‘Olaf and Peter’, ‘Mary and Anna’, and ‘Olaf and Mary’, respectively (in which case all individuals involved must usually have been overtly expressed in the preceding context, and not only some of them, as in Pro[NP]). Pro[x-NP] in languages such as Finnish and Russian has or can have comitative readings (‘I/we with Olaf’), in addition to coordination readings (see Cable 2017; Homberg & Kurki 2019), but this is never the case in Icelandic. Icelandic Pro[NP] has no readings that plain plural pronouns cannot have. Overtly expressing the annex obviously disambiguates the reference set of the pronoun, but it does not add any reading that the pronoun cannot have by itself, given the right context. We will return to this important issue in Section 5.

The order of the pronoun and its annex cannot be reversed: við/þið/þær/þau María and við/þið/þeir/þau Ólafur but *María við/þið/þær/þau and *Ólafur við/þið/þeir/þau. As far as we know, this is also commonly true of Pro[x-NP] languages, but, in at least many such languages, the pronoun and its x-NP (or NP-x) annex can be disjoint (as seen for Finnish in (1a); see also Vassilieva & Larson 2005: 111–112 on Russian Pro[x-NP]). This is impossible in Icelandic; the pronoun and its NP annex are always adjacent, locked together, regardless of case and grammatical function. While orders like “we Mary have both been there” and “Olaf had seen us Mary there” are perfect, splitting the annex “Mary” from its pronoun leads to sharp ungrammaticality. This is illustrated in (9) and (10).

Also, the pronoun cannot be expanded: *við á Brekku Ólafur “we in Brekka [a farm name] Olaf”. In contrast, the annex may be expanded: við Ólafur á Brekku ‘we Olaf in Brekka” = ‘Olaf in Brekka and I’, an issue we will return to in Section 4.

All this might seem to suggest that Pro[NP]s are just regular DPs, with the pronoun as a head or a determiner and the NP annex as its direct complement. However, such an analysis cannot be upheld. For one thing, nouns in Icelandic can, in marked circumstances, precede determiners, as in mynddiskur þessi “DVD this” = ‘this DVD’, Ólafur þessi “Olaf this” = ‘this [person] Olaf’, kona sú “woman that” = ‘that woman’, or kona ein “woman one” = ‘one/a woman’ (see Sigurðardóttir 2019 on kona ein; see also http://timarit.is for more examples).¹⁰ In contrast, as already stated, the annex in Pro[NP] can never precede the pronominal head, *Ólafur við, suggesting that the structure of við Ólafur is not parallel to the structure of ordinary DPs, such as þessi Ólafur/Ólafur þessi, and so on. We assume that the type Ólafur þessi is derived from the more neutral þessi Ólafur by Ólafur moving out of [_DP þessi Ólafur] and adjoining to the DP, yielding [_DP Ólafur_i [_DP þessi ___i]. Moving Ólafur in this way out of við Ólafur is completely blocked, which is not surprising if the structure of við Ólafur is different from that of þessi Ólafur, an issue we will return to.

Further, as we will see, the connection between the pronoun and its annex in Pro[NP] is rather different from that seen between a head and its direct complement, Pro[NP] instead being parallel to plain plural pronouns. We will return to this. Before we can do so, we need to, first, describe the properties of the annex, and, second, analyze its relation to the pronoun. We will describe the annex in the remainder of this section, turning to the analysis of the pronoun-annex relation in Section 4. In between, in Section 3, we discuss constructions that are seemingly similar to but in fact different from Pro[NP].

The annex in Pro[NP] must normally be part of the common ground, presumably for the same reason as the silent referents of bare plural pronouns must usually be part of the common ground (i.e., in, e.g., we = ‘I and x’, x must normally be part of the common ground). It is typically a personal name, either a regular one, such as Ólafur, or a simplified/short form, such as Óli. Many other categories are possible as well, though, including short form kinship terms (mamma ‘mom(my)’, pabbi ‘dad(dy)’, etc.), domestic animal names (Snati, a dog, etc., with a personification effect), definite animate group noun NPs, often denoting occupations (kennarinn ‘the teacher’, etc.), and plural noun reciprocals, such as vinirnir ‘the friends’ and systurnar ‘the sisters’.¹¹ Some of this is illustrated in (11).

We will return to plural annexes, such as systurnar ‘the sisters’ in (11d), in Section 3.2, focusing on singular annexes in this section. Before proceeding with singular annexes, though, we note that the annex cannot easily be another pronoun, plural or singular: *þið við “you.PL we”, *þið ég “you.PL I”, *þeir ég/þú “they.M I/you.SG”, and so on (in contrast to regular coordination, þið og við ‘you.PL and we’, þið og ég ‘you.PL and I’, þeir og ég/þú ‘they.M and I/you.SG’, etc.).¹²

The annex cannot be inanimate, apart from potential but normally quite far-fetched coercion under personification. In a fairy tale and other similar contexts, one could say things like við/þið/þær bókin “we/you.PL/they book-the” = ‘the book and I/you.SG/she’, but this is excluded in all regular contexts. Also, even in fairy tales, #við/þið/þær náttúran “we/you/they nature-the” is aberrant (unless Náttúran would be understood as a nickname for a person). This is parallel to the fact that plain við/þið/þær ‘we/you.PL/they.F’, without an overt annex, cannot normally refer to a human referent (the speaker, the addressee, a human non-speaker/non-addressee) and an inanimate or an abstract one, such as a book or nature.

The pronoun itself cannot normally have inanimate referents either. That is natural enough for the first and the second person pronouns, as these must include the speaker or the addressee, but this is also excluded for the third person pronouns, even though they commonly refer to inanimates. For example, masculine þeir ‘they’ may refer to bátarnir “boats-the.M”, bílarnir “cars-the.M”, or (albeit slightly more reluctantly) báturinn og bíllinn ‘the boat and the car’, but, apart from personification in fairy tales and the like, the referent of þeir (or þær or þau) in Pro[NP] cannot be inanimate: In regular contexts, þeir Ólafur means ‘Olaf and he/they’, where ‘he/they’ are male persons; it cannot mean anything like ‘Olaf and the boat(s)’; nor can þeir bíllinn “they car-the” be used to mean ‘the boat(s) and the car’.

The general rule, then, is that both the pronoun and its annex must have animate referents, normally human. Usually, the pronoun and its annex are companions, relatives or associates of some sort. However, even though this is generally the case, it is not a rule or a restriction, see the example in (12) (stated by a person who had never met Queen Elizabeth).

Furthermore, one can say things like (13).

Moravcsik (2003) and Daniel & Moravcsik (2013) assume that Icelandic Pro[NP] is an instantiation of the associative plural, as in Japanese Tanaka-tachi ‘Tanaka and his family or friends or associates’ and Hungarian Péter-ék ‘Peter and his family or friends or associates’, where the (silent) associate has a kind of a generic unspecific group reading.¹³ While this kind of usage, as in þeir Gunnar(r) “they Gunnar” = ‘Gunnar and his follower(s)’ is seen in Old Icelandic, it is obsolete in the modern language. Actually, the pronoun and its annex can easily be adversaries or “dissociates”, as in (14) (where the speaker is a female).

Singular annexes in Pro[NP] are always semantically definite in Modern Icelandic. Indefinite singular kennari is ill-formed in Pro[NP]: *Við/þið/þeir kennari “we/you/they teacher”. Similarly, indefinite *við móðir “we mother”, in contrast to the semantically definite við mamma ‘(my) mom and I’, við móðir mín “we mother my” = ‘my mother and I’, and við móðirin “we mother-the” = ‘the mother and I’. The reason why this is the case is presumably that the annex must be a referential expression and part of the common ground, hence either a contextually salient, that is, definite, common noun NP or a name (or a name-like expression like ‘mom’). In a somewhat old-fashioned style, though, some singular formally indefinite occupational NPs can be annexes in Pro[NP]: við prestur “we priest” = ‘the priest and I’, and so on. The nouns involved typically denote (some of the) traditional occupations in the old farming society: bóndi ‘farmer’, prestur ‘priest’, biskup ‘bishop’, and so on. In all such cases, however, the formally indefinite noun must be semantically definite. Við prestur means ‘the priest and I’, and not ‘a priest and I’.¹⁴ This is thus presumably a reflection of the fact that the suffixed definite article was an innovation in Old Icelandic and is still extending its domain in the modern language (see Nygaard 1906: 34 ff. on Old Icelandic; Þráinsson et al. 2015 on Modern Icelandic).

Many plural annexes are also well-formed in plural pronoun + NP constructions. We will discuss these in Section 3.2, in relation to the we linguists construction, found in English and many other languages. Here, we note that Pro[NP] annexes are easily expanded by coordination as in við Ólafur og María. This yields a systematic structural ambiguity: [[við Ólafur] og María] vs. [við [Ólafur og María]], the former meaning ‘[[Olaf and I], plus Mary]’, the latter meaning ‘[Olaf, Mary, and I (all together)]’.¹⁵

The semantics of við Ólafur is commonly “unified” (cf. Haspelmath 2007: 33): ‘I unified with Olaf’, ‘I in combination with Olaf’, or ‘I associated with Olaf’. However, even though this kind of semantics is common for Pro[NP], it is not mandatory. It seems to be derived from coordination semantics at the semantic/pragmatic interface (the C[conceptual]-I[ntentional] interface). Við Elísabet Englandsdrotting in (12) and við stelpan in (13) have only plain coordination semantics, and the pronouns and their annexes in (14) are adversaries and not associates.

Pro[NP] is reminiscent of regular coordination. Ólafur og ég ‘Olaf and I’ and ég og Ólafur ‘I and Olaf’ mean more or less the same as við Ólafur in the singular reading of the pronoun; similarly, við kennarinn and við mamma (in the singular reading of the pronoun) mean more or less the same as kennarinn og ég ‘the teacher and I’ and mamma og ég ‘(my) mom and I’, respectively. Nevertheless, coordination must be kept apart from Pro[NP]. First, while the unmarked reading of við Ólafur is ‘I and Olaf’, við og Ólafur can only mean ‘we and Olaf’. Second, as we have seen, the order in Pro[NP] is irreversible, which makes it syntactically different from coordination, as conjuncts are generally reversible: við og Ólafur = Ólafur og við ‘we and Olaf’ = ‘Olaf and we’. Third, conjuncts do not have any head properties in relation to each other, while the pronoun is the head of Pro[NP]. Fourth, of course, coordination is cross-linguistically widespread, in contrast to Pro[NP].

We will return to the comparison of Pro[NP] and coordination in Sections 4 and 5, where we add further evidence that the relation between the pronoun and the annex cannot be represented syntactically as coordination (whereas there is a silent coordination internal to the annex).

3 Comparison with other adnominal pronoun constructions

In this section, we discuss adnominal pronoun constructions that are seemingly similar to Pro[NP] but are in fact distinct from it. In Section 3.1, we compare Pro[NP] with a singular pronoun + name construction, type “he Olaf”. In Section 3.2, we discuss plural pronouns + plural NPs (types we linguists, we brothers).

4 The relation between the pronoun and the annex: the lock hypothesis

As previously mentioned, the annex in Pro[NP] can be expanded: við Ólafur á Brekku “we Olaf in Brekka” = ‘Olaf in Brekka and I’. There are some restrictions on this, though. While þessi Ólafur ‘this [person] Olaf’ is well-formed in most contexts, ?við þessi Ólafur is unnatural; we find no examples of this sort in corpora. As þessi ‘this’ is a determiner, a high D-domain element, this might be taken to indicate that the annex is not a full DP, but, rather, a smaller category, perhaps an nP, a NumP or an Art(icle)P; see the extensive discussion of DP structure in Julien (2005); Roehrs (2006); Pfaff (2015); Ingason (2016); Harðarson (2017). However, as the type ?við þessi Ólafur is not clearly ungrammatical, we assume that its markedness is not a matter of grammaticality, but rather a matter of style (and perhaps prosody), and that the annex may be either a full DP or a smaller category. In corpora (http://timarit.is), we find examples like við Ólafur þessi “we Olaf this” = ‘this [person] Olaf and I’ and við Jón þessi “we John this” = ‘this [person] John and I’. Recall, from Section 2, that the order of determiners and bare nouns can be switched in full DPs: þessi mynddiskur ‘this DVD’, þessi Ólafur ‘this [person] Olaf’, vs. mynddiskur þessi, Ólafur þessi, and so on. The switched order, Noun-Det, as in mynddiskur þessi, is usually more marked than the unswitched Det-Noun order, but við Ólafur þessi, ‘this [person] Olaf and I’, is nevertheless more natural than ?við þessi Ólafur; evidently, the name is preferably adjacent to the pronoun.

As we have seen, there is a silent or a concealed coordination semantics between the plural pronoun and its annex in Pro[NP]. However, as mentioned in Section 2, there are differences between overt coordination and Pro[NP]. We show some of these in (19).

In addition, as previously mentioned, the pronoun in Pro[NP] cannot be expanded: *við á Brekku Ólafur “we in Brekka Olaf”. In contrast, pronouns in coordination can be expanded, much as in for example English. See (20) for just two examples.

Thus, the relation between the pronoun and the annex in Pro[NP] is much tighter than the one between the conjuncts in regular coordination. It is not obvious what it is, though.

A plain syntactic solution would be to say that the pronoun is a determiner and that the annex is its direct complement, say, [_DP we [_nP Olaf]], in line with a popular approach to English APCs, such as we/you linguists, initiated by Postal (1966) and reignited by Elbourne (2005) and Collins & Postal (2012), see also for example Vassilieva & Larson (2005), and Höhn (2016; 2017). One could then perhaps assume that við Ólafur is syntactically parallel to [_DP the [_nP man]] and that Pro[NP] readings merely arise at the C-I interface, due to the semantics of plural pronouns.

While the direct head-complement approach to APCs has been argued for in many works, including the references above, such an analysis is untenable for Icelandic Pro[NP]. First, as we have seen, the annex can be a larger category than nP, for example an Art(icle)P (see Pfaff 2015), as in við málfræðingurinn “we linguist-the” = ‘the linguist and I’, or even a full DP, as in við Ólafur þessi “we Olaf this” = ‘this [person] Olaf and I’. Second, there is a sharp contrast between þessi Ólafur/Ólafur þessi and við Ólafur/*Ólafur við. While Ólafur þessi arguably involves movement of Ólafur out of [_DP þessi Ólafur], yielding [_DP Ólafur_i [_DP þessi ___i], such movement out of við Ólafur, yielding *Ólafur við, is ungrammatical, indicating that the structure of þessi Ólafur is different from the structure of við Ólafur. Third, Icelandic plural pronouns do not seem to have any special properties as compared to plural pronouns in related languages, apart from the gender distinctions in the third person (which are, however, also found in many other languages). So, if things were as simple as this, we would expect, first, that Pro[NP] expressions like “we Olaf” and “you.PL mom” should be cross-linguistically widespread, contrary to fact, and, second, that *Ólafur við should be available, also contrary to fact.

There are constructions that do involve a determiner and a name complement, such as English this Mary, not that Mary, or the Mary I know is fun (Matushansky 2006). This is also found in Icelandic, see (21) and (22).

These constructions do not have any silent or a concealed coordination semantics and are thus different from Pro[NP] examples such as “we Olaf” and “you.PL mom”. There is no question, though, that the pronoun is the head (or the label) in Pro[NP], but its relation to the annex is not a direct head-complement relation. There seems to be some hidden element or elements in the construction.

As Pro[NP] is clearly different from regular direct head-complement relations, we need to develop an alternative understanding of the relation between the pronoun and its annex. We propose that there is instead an indirect head-complement relation between the two, mediated by a silent element that heads the annex. We could refer to this silent head as simply h (for head), F (for functional category), or P or p (for preposition) but we opt for the notion lock instead, denoted as , to express the fact that the pronoun and the annex are always interlocked, cannot be broken apart. Given this, the structure of Pro[NP] expressions is as sketched in (23); the pronoun is the D head of the whole DP, and the annex is a maximal category P, headed by the silent lock.

(23)	[DP D [P [NP]]]

The lock may possibly be conceived of as a deficient or a “weak” head in the sense of Chomsky (2015) (or as a silent relator in the sense of den Dikken 2006).

Given this analysis, Icelandic Pro[NP] is structurally parallel to the Russian/Finnish Pro[x-NP] construction type, the difference being that the connector x is overt in Russian and Finnish, thereby making it possible for the annex to move from the pronoun (as discussed for Finnish by Holmberg & Kurki 2019).

As mentioned before, there is also another difference between Icelandic Pro[NP] and Russian/Finnish Pro[x-NP], namely that the Russian/Finnish constructions have or can have comitative semantics, in addition to coordination semantics, while this is excluded for Icelandic Pro[NP]. Evidently, the silent connector, the lock, does not add any semantics, in contrast to the overt ‘with’ connector in Russian and Finnish.²³

The lock seems to be related to role linkers, such as Icelandic sem, English as, German als, etc., in constructions like She was efficient as a priest/Hún var dugleg sem prestur. This Icelandic sem (as German als, etc.), is a nonassigner of case, which would fit well with the við Ólafur construction. However, the lock is not the same element as the overt role linker or just a silent version of it. First, the overt linker has clear semantic import, while, as we have seen, við Ólafur has no detectable semantics that differ from plain we (in contrast to “we with Peter” in Russian, etc.), apart from the fact that Ólafur specifies or limits the reference set of the pronoun. Second, Icelandic NP₁ sem NP₂ is usually disjoint, much as for example English NP₁ as NP₂. Third, as related languages have parallel elements, as, als, etc., claiming that the lock is a silent sem would seem to wrongly predict that Pro[NP] should be cross-linguistically common. Fourth, under this sem hypothesis, it would be unclear why the sem element must be spelled out in hún sem prestur ‘she as a priest’ (and in hún … sem prestur), while it cannot at all be spelled out in Pro[NP]: *við sem Ólafur, and so on. In NP₁ (…) sem NP₂, the sem-NP₂ phrase exhausts the comparison or the equalization with NP₁ (informally: ‘she as a priest [and not (necessarily) as anything else]’). This is excluded for Pro[NP]; it does not compare or equalize the pronoun and the overt annex. The reason for that, in turn, is that the annex also contains a silent conjunct, which means that comparison with the overt annex cannot be exhaustive, see Section 5. So, even though the lock shares some properties with overt sem, the two are distinct.

Having postulated the lock hypothesis, we turn to the analysis of the annex in the next section. The proposed analysis will eventually lead us to hypothesize that bare plural pronouns are Pro[NP]s in a disguise, with a silent lock and a silent annex.

5 The gapped annex analysis

Cable (2017) argues that Pro[x-NP] in Russian, Polish, and Tlingit is appositional. Cable’s analysis of Russian my s Petej “we with Peter” = ‘Peter and I’ is sketched in (24), where the angled brackets indicate silent material.

On this account, the construction involves ellipsis or gapping of the first person singular pronoun (ja ‘I’), and the overt connector s ‘with’ provides the conjunct or coordination relation between the silent singular pronoun and “Peter”, plus potential comitative semantics. As Icelandic Pro[NP] does not contain any overt connector, and does not have comitative semantics, this analysis does not carry straightforwardly over to Icelandic, but two similar analyses come to mind. They are sketched in (25).

The analysis in (25a) was suggested by Sigurðsson (2006); see also Thráinsson (2007: 91–92); Wood (2009). A virtue of the analysis in (25b) is that it conforms directly with the overt order of the pronoun and its annex, and also that the gap is a phrase, [_&P & NP] (on the widely adopted analysis in Johannessen 1998). We believe that (25b) is on the right track and refer to it as the gapped annex analysis.

Even if the gapped annex analysis in (25b) is on the right track, it is not complete. Consider the apposition structure in (26).

The most natural reading of constructions like við(,) María og ég in (26) is an appositional reading, yielding intonation breaks, indicated by commas in writing. Such a reading is impossible for Pro[NP] constructions like við María; an appositional reading of við María, ‘we, that is, Mary’ is nonsensical and completely excluded.

Apposition structures also differ syntactically from Pro[NP]; as illustrated in (27a), they can be discontinuous (as can Pro[x-NP] in Russian, Finnish, etc.), and as shown in (27b) their internal order can be reversed in left dislocation or hanging topic constructions.

In contrast, as we have seen, Icelandic Pro[NP] can never be reversed and the pronoun and its annex are always glued or locked together, cannot be disjoint or split. Thus, Pro[NP] and apposition structures such as Við(,) María og ég ‘We(,) Mary and I’ are syntactically different. Nevertheless, they are semantically almost equivalent (as the annex in við María contains a silent conjunct, ‘and I’, etc., see below). They are not entirely equivalent, though, at least not in information structure, as appositions, such as María og ég ‘Mary and I’ in (26) and (27), are not at issue, in contrast to Pro[NP] expressions (see Potts 2005 on at-issue and not-at-issue expressions). However, it is difficult to tease the two apart, and we will not try to do so here.

The fact that the pronoun and its annex are always interlocked is a central fact of the Icelandic Pro[NP] construction. Any approach to this unusual DP type must come up with a story on this lock effect. We account for it by postulating that Pro[NP] has a silent head between the pronoun and its annex, which we refer to as the lock. At first sight, the lock would seem to be sui generis. However, we propose that the lock is a general property of plural pronouns rather than a special property of Pro[NP]. As mentioned in Section 2, plain plural pronouns can be truth-functionally non-distinct from Pro[NP]: Icelandic við, English we, Russian my, and so on, can refer to ‘Mary and I’.²⁴ Consider (28).

The pronoun we does not have to have only the reading ‘Mary and I’ in all these cases; the relevant point is merely that it can have this reading, raising the question of how this reading is represented in syntax. That it is syntactic (and not merely semantic/pragmatic) is suggested by a number of facts, for example by dual/plural distinctions in many languages (see Corbett 2000), in particular dual/plural agreement distinctions in languages like Gothic and Slovenian, for example Gothic wit nimōs ‘we.DU [both] take.DU’ vs. weis nimam ‘we.PL [all] take.PL’. English does not have such distinctions in its pronominal and verbal systems, but it does in quantifiers: we both went vs. we all went. While we both can refer to ‘Mary and I’, we all cannot (in normal language use). That is, when deciding between we both and we all, syntax must “know” whether we refers to “individual x and I”, for example ‘Mary and I’, or “plural x and I”, for example ‘Peter, Mary and I’ or ‘they and I’.

It seems, then, that the various readings of plural pronouns, for instance the reading ‘Mary and I’ of we, are read or reconstructed into syntax, before syntax can proceed in making lexical distinctions that depend on the pronominal reading. If so, Icelandic við, English we, Russian my, and so on, can have the same syntactic representation as Icelandic við María ‘Mary and I’. We thus propose that plain plural pronouns are syntactic Pro[NP]s, with a silent lock and a silent annex.²⁵ If so, the lock between the pronoun and its annex in Icelandic Pro[NP] is an inherent property of plural pronouns rather than a specific property of the Pro[NP] construction. The lock can be broken by syntactic and/or PF means. One such means is the introduction of an overt connector in the Russian/Finnish Pro[x-NP] type, and another one is the introduction of intonation breaks in appositional constructions. Accordingly, appositional constructions and Russian/Finnish Pro[x-NP] constructions can be split or disjoint. This suggests that the Russian/Finnish constructions start out with the silent lock, the lock subsequently being broken by with-insertion in syntax or PF.

We hypothesize that spelling out the annex generally breaks the lock, and that Icelandic Pro[NP] is special in being exempted from this lock breaking (while the lock is broken in other Icelandic constructions, including overt comitative constructions, see the next paragraph). If so, the lock is broken in the we linguists type as opposed to pure Pro[NP] constructions, such as við Ólafur. We(,) linguists, as also Icelandic við(,) málfræðingar(nir) “we(,) linguists(-the)”, can have intonation breaks. This is excluded for við Ólafur, við María. Notice that this implies that our indirect head-complement analysis of Pro[NP] does not carry over to APCs of the we linguists type. As the lock is broken in we linguists, it cannot function as a mediating connector; accordingly, we linguists may be analyzed as a direct head-complement construction, as argued by Postal (1966) and many others since.

Much like for example English, Icelandic has a comitative construction with an overt preposition, við ásamt Ólafi ‘we.NOM along with Olaf.DAT’ or við með Ólafi ‘we.NOM with Olaf.DAT’, but here there is no (effective, unbroken) lock between the pronoun and the PP complement. The pronoun and the PP can be (and most commonly are) disjoint, and the pronoun cannot get a singular reading; the PP does not contain a silent coordination, so it is excluded for við to single out a conjunct within the PP.

Adopting a central idea also pursued in Vassilieva & Larson (2005) (without adopting all their technical details), we conceive of plural pronouns as sets of variables, {X, Y}, as mentioned in Section 1. We describe this in (29), where SP = speaker and AD = addressee (and where the lock addition is ours).

(29)	a.	we	= [DP {XSP, Y} [P [NP]]
	b.	you.PL	= [DP {XAD, Y} [P [NP]]
	c.	they	= [DP {X–SP/–AD, Y} [P [NP]]

Informally, thus, we usually denotes I & Y, plural you usually denotes singular you & Y, and they usually denotes he/she/it & Y (see Vassilieva & Larsson 2005: 115). The primary X variable is fixed, as the speaker, the addressee, or a non-speaker/non-addressee, but the secondary Y variable is open to interpretation/reconstruction, depending on the context.

Before proceeding, we note that this raises the question (not addressed by Vassilieva & Larson 2005) of whether the X and Y variables are syntactic features or “merely” semantic. In the general Minimalist model adopted here (Chomsky 2001 and subsequent), the variables must be syntactic features. In this model, semantics has two sources: syntax, that provides “basic semantics”, and pragmatics, that provides “additional semantics”, arising at the C-I interface from the interaction of clausal syntax with the context. Plural pronouns are partly context-sensitive, so they can have some pragmatic semantics or interpretations, but their basic semantics is clearly syntactic.

The Y variable can be constrained or specified in a number of ways. Clusivity is a widely discussed specification of this sort (Cysouw 2003; Siewierska 2004; Filiminova 2005): The first person plural pronoun may include or exclude the addressee, as sketched in (30).

The central reading of inclusive we, ‘you.SG and I’, can be analyzed as in (31); the angled brackets indicate silent material.

(31)

[we=DP {XSP, Y} <[P [NP Y=AD & XSP]]>]

inclusive we = ‘you.SG and I’

On this account, the X variable in plural pronouns is singular, thereby licensing singular readings of the silent X conjunct in the NP annex. Inclusive we can have more readings than just the central one in (31), ‘you.SG and we’, ‘you.PL and I’, ‘you.PL and we’ (see Cysouw 2003; Siewierska 2004), but it is difficult to distinguish between these in languages like Icelandic and English, which make no morphological distinctions between these readings; they are perhaps pragmatically derived at the C-I interface. However, if, for example, we = ‘you.SG and we’, has a special syntactic representation, it can be analyzed as in (32).

(32)	[we=DP {XSP, Y} <[P [NP Y=AD & XSP & W]]>]	inclusive we = ‘you.SG and we’
	‘you.SG’    ‘and we’

Also in this case, the silent X within the NP annex gets a singular reading (whereas its conjunction with W yields the plural reading ‘we’). Pronouns that translate as ‘we’ in English can have more distinctions, commonly dual/plural distinctions, such as the distinction between Old Norse dual við ‘we two’ and plural vér ‘we all’ (without a clear addressee inclusion/exclusion), but we will not probe further into this. Importantly, the annex approach yields an effective tool to exactly analyze the various readings of plural pronouns.

In Pro[NP], the Y variable is lexically specified, as illustrated for the singular reading of the pronoun in við Ólafur in (33), where, again, the angled brackets indicate silent material.

(33)

[við=DP {XSP, Y} [P [NP Y = Olaf <& XSP>]]]

við Ólafur = ‘Olaf and I’

The alternative reading, with the less central plural interpretation of the pronoun, is illustrated in (34).

(34)

[við=DP {XSP, Y} [P [NP Y = Olaf <& XSP & W>]]]

við Ólafur = ‘Olaf and we’

We find it worth pointing out here that Icelandic plural pronouns seem to be bimorphemic, v-ið, þ-ið, and þ-eir, þ-ær, þ-au. An intriguing thought is that the first element or “morpheme” actually represents the singular X variable in the pronoun, enabling the pronoun to identify and license the singular reading of X within the NP annex, across a lexicalized Y variable. This is clearly not a sufficient condition for Pro[NP] (cf. French n-ous, v-ous, Italian n-oi, v-oi, etc.), but it might be a necessary one. We leave this for future research.

The X variable is already in the pronoun variable set, so a silent copy of it within the annex might seem to be unnecessary. However, there must be a hidden conjunction in the structures, or else the coordination reading could not be derived (recall that there is no such reading in overtly comitative constructions in Icelandic, “we (along) with Olaf”, etc.). In our analysis, the hidden conjunction or conjunctions are within the annex. Having the conjunction(s) within the pronominal set instead, between X and Y, is not an option, as this would wrongly eliminate the difference between við and coordinates such as ég og þú: both would be [_DP {X_SP & Y}] on this approach. In our analysis, ég og þú has the structure [_DP X_SP [_&P & Y = AD]] (or simply [_DP ég [_&P og þú]]), whereas við = [_við=DP {X_SP, Y} <[_P [_NP Y & X_SP]>]]. In addition, this general [_DP {X_SP & Y}] analysis would raise the question of why [_DP {X_SP & Y}] would not always be spelled out as coordination; economy considerations would seem to predict that plural pronouns should either be marked or nonexistent under this approach.²⁶ Notice further that while plain við/we denotes a partly open set (for example ‘I + Olaf, Mary, Peter’), ég og þú (or ég og Ólafur) does not.

At first sight, it might seem to be a conceivable alternative to have a conjunction between the pronoun and the annex (instead of the lock), but that would make við Ólafur non-distinct from við og Ólafur, which would exclude the central ‘Olaf and I’ reading. This would also be problematic for the plural reading of the pronoun. Even though this reading is best rendered by the English translation ‘Olaf and we’ or ‘we and Olaf’, við Ólafur (við = plural) and við og Ólafur are not equivalent. In við og Ólafur, Olaf is always excluded from the reference set of the pronoun; the pronoun refers to ‘I + Z’ (=we), where Z explicitly does not include Olaf (just as in English we and Olaf). In við Ólafur (við = plural), in contrast, Olaf is included in the reference set of the pronoun, ‘I + Z + Olaf’; the only difference between plain við and við Ólafur is that the annex specifies that Olaf does belong to the reference set of the pronoun.

In sum: Plural pronouns contain a feature variable set, {X, Y}, and coordination semantics. The coordination cannot be within the variable set, as that would conflate plural pronouns and regular overt coordination, you and I, and so on. And in Pro[NP], the coordination cannot be between Pro and NP, as that would exclude the central singular reading of the pronoun. Thus, we conclude, the silent coordinator must be within the NP annex, in both “we Olaf” and in the silent NP annex in plain we/við. Spelling out the whole NP annex, including the coordinator, yields adjunction structures, as in we, Olaf and I and we, you and I, different from both “we Olaf” and plain we/við, and spelling out the coordinator plus or minus Olaf yields ungrammatical *we Olaf and, and *we and.

There is a general restriction on the relation between the X and the Y variables, such that the Y variable cannot include or overlap with the singular referent of the X variable.²⁷ Thus, we cannot mean “plural speaker”, ‘I & I’ (see Boas 1911; Benveniste 1966; Bobaljik 2008).²⁸ They can mean ‘he/she & he/she’, but the two “hes/shes” are not equivalent; the ‘he/she’ referred to by the Y variable is necessarily distinct from the ‘he/she’ referred to be the X variable (as sometimes made explicit by gender, as in Icelandic neuter þau, referring to males + non-males or females + non-females (or neuter(s) + neuter(s)). Similarly, even though plural you can mean ‘you.SG & you.SG’ the two singular “yous” must (obviously) be distinct; the Y variable ‘you’ cannot include or overlap with the X variable ‘you’. As a matter of fact, no language is known to indisputably have a plural pronoun that specifically refers to addressees only, excluding everybody else (Simon 2005; Bobaljik 2008).

Our analysis minimizes the difference between Pro[NP]s and plain plural pronouns, but it does not eliminate it. Without lexical specification, the NP annex is unsaturated, and við can include unlimited salient (primarily human) referents, in addition to the speaker: Olaf, Mary, Peter, and so on. With lexical specification (plus the representation of the X variable in the silent conjunct), the annex is either fully saturated or partly saturated. When fully saturated, any other referents than the lexicalized one (and the silent X variable) are excluded, and við Ólafur then means only ‘Olaf and I’. When partly saturated, the lexicalized referent is included, without excluding other potential referents, and við Ólafur then means ‘Olaf and we’.

6 Concluding remarks

We have argued that Pro[NP] constructions are constructed in a way that reflects a more general, deeper property of plural pronouns: that they are constructed from a set of variables and an annex, which is headed by a silent head, the lock, . The set of variables includes a primary variable, X, identifying the speaker, addressee, or neither, and a secondary variable, Y, which is context dependent and can be constrained or specified by the annex in a number of ways. What is unusual about the Icelandic Pro[NP] construction is that the annex can be partly lexicalized without breaking the lock. In this way, plural pronouns in general have a singular reading of the X variable, one that is usually not made overtly obvious. The overall picture strongly suggests that pronouns do not have meanings that are given in a presyntactic lexicon, but that plural pronouns are built out of features whose interpretation is syntax-/context-dependent. Our analysis also supports the view that silent elements (“dark matter”) are pervasive in syntax (as argued in Sigurðsson 2004 and much subsequent work).

Exceptional phenomena often throw new light on the nature of more general phenomena. Icelandic Pro[NP] is highly exceptional, and it does indeed throw new light on the nature of plural pronouns. Our analysis involves some unorthodox ingredients, not surprisingly, in view of these circumstances. At first sight, the lock might seem to be sui generis, but we have argued that it is actually a general property of plural pronouns. In addition, we have argued that plural pronouns, bare as well as in Pro[NP], have a silent copy of the X variable in the NP annex, and that the reading of the X variable is consistently singular. This is a universal property of plural pronouns, we claim.

We have shown that the widely adopted direct head-complement analysis of APCs like English we linguists does not work for Icelandic Pro[NP]. Instead, we have argued that there is an indirect head-complement relation between the pronoun and its annex in the Icelandic Pro[NP] type, mediated by the silent lock, which brings the Icelandic construction in line with the Russian and Finnish type of Pro[x-NP]s, “we with Peter” = ‘Peter and I’, and so on. The difference is that the overt connector in the Russian/Finnish type breaks the lock between the pronoun and its annex, thereby making it possible for the annex to move from the pronoun, or vice versa.

Abbreviations

In glosses:

ACC = accusative, DAT = dative, DU = dual, F = feminine, GEN = genitive, INSTR = instrumental, M = masculine, N = neuter, NOM = nominative, PL = plural, SG = singular, 1 = first person, 2 = second person, 3 = third person

Others:

AD = addressee, APC = adnominal pronoun constructions, ECM = exceptional case marking constructions, C-I = conceptual-intentional interface, NP = noun phrases of various sizes, PrC = preproprial constructions, SP = speaker, Pro[NP] = Plural Pronoun[NP] Constructions, Pro[x-NP] = Plural Pronoun[x-NP/NP-x] Constructions (where x is a connector/preposition)

Competing Interests

The authors have no competing interests to declare.