Proper name compounds: a comparative perspective

The article discusses compound formation involving proper names from a comparative perspective. While proper names can appear within compounds in English, this is not possible in Greek. The article argues that this follows from a basic difference between English and Greek: English, but not Greek, allows phrases as non-heads of right-headed compounds. As proper names in English are referential in the absence of a determiner, due to the process of D-N merger, they can still be recognized as such within compounds. This is not possible in Greek, where proper names require the presence of a determiner to establish reference.

(1) Kerry supporter (2) Goethe-Verehrung 'Goethe adoration' (3) Palmemord-et 'the Palme murder' In these languages, proper names can appear in root compounds as well, illustrated here with English: (4) Rembrandt year Proper names in such compounds refer to persons well known within the speech community, as detailed in Koptjevskaja-Tamm (2013). According to Koptjevskaja-Tamm (2013), proper name compounds (henceforth PNCs) are intriguing for the following reasons: (b) etos Luthiru year Luther-GEN (c) tsipri-es idees Tsipras-ADJ ideas While (7a) is a deverbal nominalization, (7b) corresponds to a N-NGEN left-headed phrasal compound known from Romance, as will be discussed in section 2. Thus, we can reformulate the puzzle: while in English phrases can appear as non-head elements in right-headed compounds, this is not possible in Greek. In Greek, this is only possible in left-headed compounds, as is the case in Romance. As (7a) shows, it is impossible to leave out the determiner in the case of the deverbal nominalization, supporting the conclusion that there are no phrasal compounds in Greek with a deverbal head; see Alexiadou (2017a). In (7b) the genitive is non-referential and seems to provide the name for the year, similar to proper noun modifiers in English. Examples such as (8), found online, do not challenge this generalization, as (8) involves a coordination structure. Coordination structures have been shown to license bare singulars even in languages such as English that disallow them. Heycock & Zamparelli (2003) argue extensively that coordinated bare nouns have an interpretation that is very close to that of definite noun phrases, and the Greek data seem to support their overall conclusions. 4 (i) A black cat and a brown dog were fighting in the street.
(a) [Cat and dog] were equally filthy.
While Greek allows bare singulars in object positions (Alexopoulou, Folli & Tsoulas 2013), and as we will see later on, bare proper names with a non-referential interpretation in argument positions, it crucially does not allow proper names to occur bare in any argument position (subject or object) preserving a referential interpretation: (iii) (a) *Tsipras episkeftike ton Proedro Tsipras-NOM visited the President-ACC (b) *ida Tsipra saw.1SG Tsipra-ACC Interestingly, in headlinese, it is possible to use the proper name without an article in the nominal domain, as in (iv). This has been noted in passing in Lazaridou-Chatzigoga (2009: 189) for common nouns in Greek, and I think it is more salient for proper names: The second issue that will concern me here relates to the additional ways a proper name may combine with another noun in the two languages: in English, next to the PNC there are two genitival constructions, the so-called Saxon genitive as well as post-nominal genitive, e.g. Kerry's supporter, supporter of Kerry and Kerry supporter. In Greek only the post-nominal genitive is available, (7a-b), whereby, as just mentioned and as we will see in detail, the nominalization behaves differently from the phrasal compound; alternatively, the adjectival formation in (7c) can also be employed. There are of course differences between all these cases, which I will revisit in section 4.
The aim of this article is thus to offer an explanation for these two observations: (i) no proper name can appear as the non-head member of a right-headed SC or root compound in Greek, as opposed to English; (ii) no proper name can appear as the non-head element of phrasal left-headed deverbal compound, while it can appear in this position when the head is not deverbal. I will use the Greek vs English contrast as a tool to understand the structure of compounds. Specifically, I will address the following questions: (a) why is it impossible to form a compound with a proper name in Greek, while this is possible in English? (b) What does the above puzzle tell us about the structure of compounds and the nature of proper names in Greek and English and generally across languages? As I will show, the behavior of Greek vs English PNCs is a general reflex of the compounding parameter proposed in Ralli (2013) and further supported in Ioardachioaia et al. (2017), according to which languages fall into two groups: those that have non-phrasal left hand non-heads like Greek vs those that have phrasal left hand non-heads like English. Assuming that proper names are phrases, the ban of PNCs in Greek is a reflex of this property: while in English compounding involves phrasal elements as non-heads, in Greek it must involve roots as non-heads. I will also argue that there is a difference with respect to the structure of proper names in the two languages. As proper names are obligatorily DPs in Greek in order to yield rigid designation, this leads to a general ban on compound formation involving proper names in the language. The only possible way to express a meaning close to that of compound is thus via the genitival or the adjectival construction, as in (7b-c). Proper names in English are phrasal and can refer to individuals, because D and N combine. As English compounding involves phrases in general, there is no blocking of PNCs. As such a merger is not an option in Greek, PNCs are out. Turning to the second observation, i.e. the type of head in left-headed compounds, deverbal vs root, the ban on PNCs with deverbal heads relates to the fact that Greek simply does not form phrasal compounds of this sort, as will be discussed in section 2.
The article is structured as follows: in section 2, I present a detailed description of Modern Greek compounding. In section 3, I turn to the status of proper names in As is well known, articles can be dropped from common nouns in English headlinese as well, but not much is known about the features of this particular register in Greek. The fact that proper names can appear without an article in the nominal domain but not with verbal predicates suggests that headlinese generalizes the N-N GEN pattern, when specific reference can be recovered from the context.

ARTEMIS ALEXIADOU
Greek as opposed to English and offer an account that explains their properties. I also discuss in detail the restrictions on compounding. In section 4, I turn to an analysis of the two compound structures and then to a comparison between Greek and English genitival constructions. In section 5, I conclude and discuss some interesting consequences of the analysis puts forth in this article. I will point out that the explanation offered here might help us understand another cross-linguistic puzzle on compounding, namely the lack of quotation compounds in Greek as opposed to English.
(9) (a) níhta lulúdi niht-o-lúludo stem-stem night flower night-LE-flower (b) thrirío damastís thiri-o-damastís stem-word beast tamer beast-LE-tamer (c) antropos arahni spider-man N-N man spider (d) praktorio idiseon news agency N-NGEN agency news-GEN The first two types are right-headed structures while the third one is a left-headed structure of the type familiar from Romance languages. The first type involves inflectional and phonological variation, as the compound element bears a different stress from the pattern the two members would bear in isolation. According to Ralli, both the head and the non-head are stems. In the second type, the non-head is obligatorily a stem, while the head of the compound bears the same set of inflectional affixes that it would have in isolation, and thus qualifies as a word. These two types are the ones that contain a linking element. Ralli (2013: 59) convincingly distinguishes the linking element from the inflectional endings of the non-head and I will adopt her conclusions here; see footnote 2. These two types also have in common that they form a single stress domain.
The third type comes in at least two variants, N-N and N-NGEN, i.e. either the two nouns agree in case, as in (9c), or the non-head which is on the right surfaces with genitive case, as in (9d). In both instantiations, we have two stress domains. As Ralli (2013: 248) details, the constituents that build the third type lack syntactic autonomy, as the non-head cannot be modified, their order cannot be reversed, and nothing can intervene between the two constituents. Importantly, however, they involve phrasal elements as non-heads, as can be seen by the fact that they bear inflection class markers.
Greek right-headed compounds are subject to the bare stem constraint, according to Ralli (2013), i.e. the non-head must appear as a bare stem. Ralli uses the label stem in her work to emphasize the fact that in principle this form could not stand in isolation and cannot carry any type of inflectional or derivational morphology. This means that 859 PROPER NAME COMPOUNDS non-heads are morphosyntactically dependent: as Ralli (2013: 133-4) notes, non-heads disallow both derivational and inflectional suffixes. A derived noun like player is not possible in a compound, as in (10a). Such nouns are categorized as words and have to be part of analytic compounds, as in (10b) (NZ = nominalizer).
Ralli argues that this relates to the characteristic property of Greek as a stem-based language. I will use the term root here to refer to what Ralli calls stem, as I take it that these forms have not been categorized as bearing nominal features What this crucially amounts to is that the language does not allow roots to surface as words without carrying any inflection, where it essentially differs from English. In other words, a noun in Greek always bears inflection and gender features; see Alexiadou (2017b) for a recent discussion. Importantly, it involves the combination of a root together with a particular functional layer where this inflection is introduced. The obligatory presence of the linking element /o/ and the absence of any inflectional marking for declension class (see Ralli 2013) enforces the idea that non-heads in Greek compounding are roots, which are morphosyntactically incorporated. This will be an important ingredient in approaching the ban on PNCs in Greek. When nouns appear with inflectional endings, they must appear in a phrasal compound, as in (9c-d). In this case, Ralli's constraint does not hold and the linking element does not appear, as mentioned in footnote 3.
While formations corresponding to English root compounds can belong to any of these three types in Greek, SCs qualify as stem-word compounds in Ralli's system, as the stress falls on the same syllable of the nominal head as it would if it appeared in isolation; see (9b) and (11): 5 (11) kapn-o-kaliérgia kapn-o-kaliergitís (cf. kapnós; kaliérgia; kaliergitís) tobacco-LE-cultivation tobacco-LE-cultivator tobacco; cultivation; cultivator 5 Kechagias (2005) shows that this does not hold for all SCs. There are also SCs for which no corresponding verb exists. As in some cases, the N-V verbal form is obsolete, it could be argued that these forms are relics of older N-V verbal compounds (see Alexiadou 2017a for discussion). It is also not clear where in these cases the head functions more like an affix; see footnote 2.
According to Ralli (2013), there is also a further case of SCs, where the N-V verbal element seems to be a back formation, as illustrated in (iii), where the verbal form in the SC is different from that of the verb in isolation.
Proper names cannot appear in any of the above formations, which would in any case also involve a root as the non-head of the compound.

ARTEMIS ALEXIADOU
Such compounds are not very productive, i.e. some non-heads are better than others: Unlike root compounds, the non-head functions as the internal argument of both the SC and the N-V verbal compound. Evidence for this comes from the fact that neither SCs, nor their N-V corresponding verbal compounds allow the realization of a phrasal argument next to the element that appears as the non-head of both the SC and the verbal compound, as discussed in Iordachioaia et al. (2017).
Both the examples in (16) have been argued by Horrocks & Stavrou (1987) to involve fronting of the genitive to a position higher than the determiner, identified as Spec,DP: According to one influential approach in the literature (Longobardi 1994), proper names across languages are referring expressions, and as a result a D layer is always present in their morphosyntactic representation. This explains the fact that in English proper names are not introduced by articles in argument positions. According to Longobardi (1994), in English proper names involve N-to-D movement that takes place at LF, i.e. covertly. To the extent that there are languages such as Greek in which proper names co-occur with articles, these articles must be void of semantic import. However, there is evidence for the fact that in Greek, articles are not expletive, thus providing support for an analysis according to which proper names have the syntax of common nouns, e.g. Borer (2005) The proper name Judas appears without the determiner. In this case, the sentence means that someone with the properties of Judas (e.g. a traitor) advised them. Thus, they conclude that the Greek definite determiner is not expletive and, when present, it has a semantic effect, namely, to introduce unique reference. Assuming then a structure of the noun phrase that separates the functional layer of definiteness from that of the lexical content, we can assume (22) for the representation of proper names in Greek: 6 (22) DP (reference, definiteness) \ NP/nP (lexical content) In English, where proper names occur in compounds, the proper name is recognized as such. 7 Ghomeshi & Massam (2009), discussing examples such as Nixon hater, point out that since this is so, the features that are responsible for this must be located at the level of the NP/nP. To explain this, they distinguish between different flavors of n, namely common n vs proper n. However, there is an alternative way to capture this without having to assume flavors of n. Matushansky (2006) argues that the proper name and the definite article undergo morphological (m-)merger in English, when nothing intervenes. Specifically, Matushanksy (2006: 296) states that 'm-merger is a strictly cyclic morphological operation that takes two syntactic heads in a certain configuration and returns one syntactic head. It is subject to strict locality: nothing may intervene between the two heads: neither a Spec nor a modifier.' This in turn means that in (22), m-merger takes place in English, but not in Greek, assuming as in Longobardi (1994) that proper names always contain D.
Evidence for m-merger in English comes from the following facts: the determiner is obligatory when the proper name is modified by an adjective or the proper name is plural. Burge (1973: 429) offers the following examples, where a proper name can carry plural morphology and appear after quantifiers, just like common nouns. As is well known, proper names obligatorily co-occur with a determiner if they are plural: (23) (a) There are relatively few Alfreds in Princeton.
(b) Some Alfreds are crazy; some are sane.

(c) The Clintons
Modification and plurality block m-merger because they intervene between D and N: it has been argued by Ritter (1991) that Number is a functional projection between D and N (24); more recent approaches to the functional architecture of the noun phrase make similar claims, although they use a different label for the projection that hosts plurality, e.g. Borer (2005) and Heycock & Zamparelli (2005). 6 For the purposes of the argument it does not matter if lexical content is associated with the NP layer, or, as is standard in Distributed Morphology, with the nP layer. I will come back to that in section 4. 7 In fact, this is the case even with proper names that necessarily require the definite article even in English, e.g. Bronx lover vs The Bronx (Huddleston & Pullum 2002: 517;Borer 2005 Abney (1987) has argued that adjectives in English are heads in the extended projection of the noun, thus this would also block m-merger of D-N in English. Fara (2015) shows that there are two type of modifiers, those that she labels permissive, which allow the definite determiner with the proper name, as shown in (25), and those that she labels strict, as shown in (26) We can thus conclude that in English, but not in Greek, D undergoes merger with N when nothing intervenes, the result being that proper names are recognized as such and are referential even in the absence of an overt determiner. In other words, the contrast between English and Greek is that rigid and unique reference is established via overt D in Greek, while D and N merger is possible in English. Matushansky (2006) does not offer an explanation as to why merger does not take place in some languages, but it is conceivable to think that Greek disallows morphological merger of D and N as in this language merger always applies between Number and NP/nP, as the language has syncretic nominal morphology; see Alexiadou & Stavrou (1997). Since Number fuses with N/n, D, when present, is realized by an overt determiner. In other words, Number-N/n merger blocks D-N merger; in turn, D receives a realization agreeing with the head noun. This necessarily means that in Greek no noun can appear without inflection class marking; when this happens, then there is simply no NP/nP in Greek, we have units that are smaller than NPs/nPs, i.e. basically roots. Although I consider the evidence in favor of m-merger substantial, I'd like to point out that for the purposes of this article what is important is that D and N always combine under one head in English, while they never do in Greek. M-merger might be one way in which this is done, but it might not be the only alternative.

Restrictions on compounding in Greek versus English
Let us recall the basic restrictions once more. As mentioned in the introduction, although Greek allows (6), it disallows (5), repeated below. The only possibility is to make use of a genitive construction, as in (7a-b), or to form an adjective out of the proper name and use it as a modifier, as in (7c), repeated below: (7) (a) thavmastis tu Tsipra admirer the Tsipras-GEN (b) etos Luthiru year Luther-GEN (c) tsipri-es idees Tsipras-ADJ ideas As (7b) shows proper names can appear in left-headed compounds, which are phrasal. However, they cannot appear in a particular type of left-headed exocentric compounds described in Ralli (2013: 113-15). Ralli points out that Greek has some relics of V + N compounds, as in (27), which violate the requirement that compounds in Greek are right-headed. The left member corresponds to the Ancient Greek verb 'filo' 'to love': 8 (27) (a) filo-zo-os love-animal-MASC.SING 'the one who loves animals' (b) filo-amerikan-os love-American-MASC.SING 'the one who loves American people' Proper names cannot appear in such compounds either, unless they become adjectives: (28) (a) *filotsipras (b) filo-papandre-ik-os love-tsipras love-papandreu-adj-MASC.SING 'the one who loves Papandreou' Where a proper name is, however, possible in Greek without a determiner is in N-NGEN phrasal compounds of the type zoni asfalias 'belt safety.GEN'; these do not have a deverbal head: 8 Thanks to Elena Anagnostopoulou for pointing this out to me. However, the genitive in these strings does not have an argument status and is not referential. This can be shown by applying Kolliakou's (2003) pronominalization diagnostic. As Kolliakou argues, a pronominal clitic can replace a noun phrase that has independent reference, while it cannot replace a modifier genitive: (30) (a) o dolofonos tu/tu Jani the murder his/the John-GEN (b) to vivlio tis istorias/*to vivlio tis the book the history-GEN/the book hers Applying this test to the data in (29), we note that a genitive clitic is possible, but it cannot refer to the proper name if this is not introduced by a determiner: (31) to ergo tu/tu Tritsi the work his/the Tritsi-GEN 'his/Tritsi's work' N-NGEN phrasal compounds contain thus non-referential genitives; see Alexiadou, Haegeman & Stavrou (2007) for discussion. In the literature, such genitives are referred to as classifying or property genitives. As we will see in section 4, they behave in a very similar way to proper noun modifiers in English. A proper name lacking a determiner is very degraded with derived nouns unless these are coordinated, as we saw above in (8). (32) might be found in headlinese, as pointed out in footnote 4, where the referent both of the deverbal noun and the proper name can be retrieved from the context: (32) ??thavmastis Tsipra admirer Tsipra-GEN Recall that in section 2, I mentioned that Greek does not have phrasal compounds with derived nominal heads and that strings such as (14), repeated below, are best analyzed as dispositional nominalizations: (14) damastis thirion tamer beasts-GEN 'animal tamer/tamer of animals' The analysis of those as dispositional nominalizations adopted the view in Alexopoulou, Folli & Tsoulas (2013)  (2017a) argued is that when such nouns appear with deverbal nominals they lead to a dispositional reading of the overall nominal, which is the case in (14). But, importantly, proper names in argument positions in the absence of a determiner have a simple predicate meaning, i.e. they lack a referential interpretation; see also footnote 4 and (21) above. Thus, to the extent that (32) is possible at all it would force the proper name to be interpreted as a modifier of the head noun similar to the cases in (29). Basically, it would imply that a particular individual would have the profession of being an admirer of someone with the properties of Tsipras, and this is why such structures feel very degraded.
With these remarks then, I will turn to an analysis in the next section. I will argue that the difference seems to be related to size of units that enter compounding, uncategorized stems vs categorized words (Iordachioaia et al. 2017). We saw that Greek forms SCs using stems as non-heads. English, on the other hand, can build SCs on the basis of phrasal elements, i.e. NP/nPs.

Differences in compound structures between Greek and English
In section 3.1, I adopted an analysis, according to which proper names in English D-N combine. My structural analysis is cast within the framework of Distributed Morphology, a basic assumption in which words are complex as they are built in the syntax via combining an acategorial root with categorizers (n, v); see Arad (2005), Embick (2010), Marantz (2013) and others. Crucially then, NPs are viewed as being complex, containing an n layer that combines with an acategorial root. This is an important assumption that enables us to draw a distinction between compound structures that truly involve roots as opposed to compound structures that involve non-heads which have a phrasal status. As in Greek all inflectional information is on n, the fact that compounds have non-heads which lack such information suggests that these are smaller than nPs. They are basically roots. nPs appear in phrasal compounds, which are right-headed, and in which case they lack determiners, while they carry all other inflectional information. Iordachioaia et al. (2017) offer a structural analysis of SCs of the type in (33)  These are examples in which the SC appears to be a nominalization of the N-V verbal compound. According to these authors, (33) involves incorporated roots. Specifically, 867 PROPER NAME COMPOUNDS SCs in Greek involve morphosyntactic incorporation, where the non-heads of the SC are roots, which are morphosyntactically dependent, as we have seen above, i.e. stems. The derivation of thiriodamastís ('animal tamer') is as in (34): (34) thiriodamastís ('beast tamer'); thiriodamazo ('to beast tame') In (34), according to Iordachioaia et al. (2017), the verbal root is categorized as a verb, forming the stem damas for the lexical verb damázo 'to tame'and then root thiri 'beast' moves to v and incorporates into it, forming the N-V compound thiriodamázo, a vP. 9 This structure basically involves a root incorporation analysis as in Harley (2009) for English (cf. Michelioudakis & Angelopoulos 2013). Following Ralli (2013), the linking morpheme -o-has no syntactic status, it is simply a phonological reflex, as mentioned in the introduction. This analysis explains, according to Iordachioaia et al. (2017), the limited productivity of Greek SCs as a result of the idiosyncrasy of the non-head roots: some can undergo incorporation, but others cannot, depending on our world 9 The difference between (34) and the cases mentioned in footnote 5, e.g. vivli-o-detó 'book-bind', has to do with the fact that in vivli-o-detó, the two roots combine with the categorizer as a unit. Crucially, in both the SC is a nominalization of a verbal compound.

ARTEMIS ALEXIADOU
knowledge. The opposite holds for English synthetic compounds, which are not constrained along these lines. These steps enable us now to approach the PNC issue. In Greek, Borer (2005) argues, the formation of a set with a single member necessarily involves the determiner, e.g. o Tsipras. If in Greek, however, right-headed compounds cannot have phrasal non-heads, it is not possible to form a compound that contains a proper name since proper name reference necessarily requires an nP structure combining with D, which is banned from Greek compounding. Now, in languages with noun incorporation, proper names never incorporate (Baker 1988). This has been argued by some researchers to be related to the fact that proper names cannot be interpreted as general enough to make good candidates for incorporation. While proper names can enter other derivational processes, they lack the unique reference property, as expected, and which we also observe in true phrasal compounds.
In English, according to Borer (2005), proper names involve, in some cases, phonological representations that make reference to a set with a single member (Borer 2005: 85). Following the analysis in section 3.1, this corresponds to a nP and, importantly, it does not require the presence of an article, since we have a combination of D and n. This in turn suggests that SCs in English do not involve root incorporation, as Iordachioaia et al. have argued. In English, the non-head is at least an nP, and the SC is derived as in (35) The difference between English and Greek is that the non-head disposition is a noun and not a root. English differs from Greek in that it lacks root incorporation. Rather, what happens is that the verb gets nominalized by the suffix -er and the noun disposition moves to the Spec nP of the nominalized structure. This is because it cannot be marked for Case and is illicit in this argumental position (cf. Longobardi 1994); thus, it has to move. Iordachioaia et al. (2017) argue against incorporation in SCs in English. Among other things, as they point out, see also Wiese (1996), non-heads can be morphologically complex. Such strings are impossible in Greek: (36) season ticket holder, air traffic controller, flight data recorder, child care provider, science fiction writer, ice cream maker, documentary film maker, sports car maker A similar type of analysis can be given to root compounds of the form Luther year, where the proper name functions as a modifier of the head noun, again situated in Spec,nP. I will come back to this in the next section. As Rosenbach (2007) observes, such strings deviate from the typical compound stress of English and show phrasal stress; see Plag (2006). This already suggests that compounds involving proper names cannot simply involve the combination of two nouns into a single unit. This is reflected in the analysis in (35). 869 PROPER NAME COMPOUNDS