The syntactic structure of pelirrojo compounds

This paper deals with Spanish adjective-headed compounds (pelirrojo), whose main properties are systematically compared with those of their English counterparts (red-haired). I provide a distinction based on the grammatical category of the non-head projections peli and red and on their relative complexity to capture the relevant differences. Additionally, I focus on the most noticed feature of Spanish pelirrojo compounds, which is the fact that they only codify inalienable possession relationships. I argue that the existence of an inalienable-only compound pattern shows that Spanish pelirrojo compounds contain a relational head which establishes an inalienable possession relationship between a possessee located inside the compound (pelo ‘hair’), and a possessor located outside it (e.g., Juan in Juan es pelirrojo. ‘John is red-haired.’). The same kind of relational structure is not identifiable inside English parallel structures or other Spanish compound patterns.


Introduction
Pelirrojo 'red-haired', like any other lexical compound, is usually characterized as a morphological construction, in contrast to phrasal compounds like oro negro, lit.'gold black', 'oil', which more frequently receive a syntactic treatment (Rainer & Varela 1992;Val Álvaro 1999).The tradition of Spanish morphological studies has favored exhaustive descriptions of compound patterns by themselves (Bustos 1986;Buenafuentes 2007) and, for that reason, their properties have been frequently considered independently from the properties of related phrasal constructions.In this paper, it is shown that even the behavior of lexical compounds like pelirrojo can be explained by appealing to syntactic properties, which apply similarly to both phrasal and word syntax.
An old and well-known controversy exists between those who defend the idea that word and phrase generation require distinct grammatical components or rules (Di Sciullo & Williams 1987;Lieber & Scalise 2006) (lexicalist approaches) and those who think that syntax is responsible for both word and phrase generation, without the need for either a specific morphological component or morphological rules (Harley 2009; Fábregas 2016) (neoconstructionist approaches).The data analyzed in this work supports the latter theoretical framework.
It would follow from the syntactic nature of compounding rules that the internal structures of red-haired and pelirrojo can be differentiated by appealing to the nature of their non-head projection, which is an Adjective Phrase in the first case and a Noun Phrase in the second case.Additionally, pelirrojo contains a relational category provided with semantic-functional content (i), which makes its projection more complex and its denotation more restricted than that of its English N+Adj correlates (color-blind).The more common morphological approaches assume that i is a linking vowel without structural significance.Our analysis will even include a potential violation of one lexicalist's prediction,1 as the kind of mandatorily inalienable relationship between the Noun Phrase outside the compound (Juan 'John') and the one located inside it (pelo 'hair') in constructions like Juan es pelirrojo is explained more straightforwardly with the help of syntactic rules.
The paper is organized as follows.In section two, I provide different syntactic structures for certain classes of constructions discussed by the literature on compounding.Those are related to pelirrojo compounds as they present both noun and adjective2 constituents, the latter introducing an attributive relationship in the sense of Bisetto & Scalise (2005).However, they differ significantly from pelirrojo in syntactic, semantic and/or phonological terms.
I will make three relevant assumptions about the structure of pelirrojo compounds (1), all of which can be found in the previous literature (we will comment on them later).The first assumption is that the head of the compound is the simple adjective (rojo 'red', alegre 'happy' or difuso 'vague ', etc., in the examples below).The second one is that the adjective establishes a subject-predicate relationship with a Noun Phrase located outside the compound (e.g., Juan 'John' in Juan es pelirrojo.;me 'I' in Me dejaste patidifusa.'You left me astounded.').On the other hand, I assume that the nominal projection inside the compound functions as a restrictive complement of the adjective itself -it is not predicated of Juan to be rojo, but only to be rojo regarding his hair.Finally, I assume that the nominal non-head establishes a mandatory inalienable relationship with the subject outside the compound.I will explain what inalienable relationships are later.In a nutshell, inalienability implies that the nouns inside pelirrojo compounds, such as pelo, can be understood only in relation to the entity that possesses them (Juan in Juan es pelirrojo's case).
(1) pel-i-rroj-o car-i-alegre pat-i-difus-o boqu-i-suci-o hair+I+red+m face+I+happy leg+I+vague+m mouth+I+dirty+m 'red-haired' 'smiling' 'astounded' 'foul-mouthed' piqu-i-tuert-o peak+I+crooked+m 'crossbill' Being endocentric and adjective-headed constructions, pelirrojo compounds must be distinguished from the agglutinated phrases shown in (2).Their categorial head is the noun element, which shows gender agreement with the adjective non-head, as well as number agreement in the less agglutinated cases.Nevertheless, the phrases in (2) are exocentric from the semantic point of view, as the body part nouns (cara 'face ', cabeza 'head' and piel 'skin' in the examples below) are used to refer metonymically to whole individuals. 32) cara-dura cabeza-s loca-s face-fem+hard-fem head+fem+pl+crazyfem+pl 'cheeky' 'dim wit' piele-s roj-a-s cara-pintad-a skin-fem+pl+red+fem+pl face-fem+painted-fem 'redskin' 'extreme right wing' Pelirrojo compounds can be further distinguished from constructions where it is the adjective which acts as the modifier of the noun (as opposed to (1)).Alto 'high' modifies Aragón 'Aragon', libre 'free' modifies cambio 'change' and bajo 'low' modifies medievo 'Middle Ages' in (3).The modifier+head constituents are turned into adjectives by the adjectivizing suffixes afterwards.
(3) altoaragonés librecambista bajomedieval high+Aragón+IAN free+change+IST low+Middle Ages+AL 'from north Aragón' 'free trader' 'from the late Middle Ages' (English examples: red-haired, quick-tempered, high-pitched) Finally, pelirrojo compounds differ from other adjective-headed compounds with nominal non-heads or complements − (4) − where inalienable relationships are not established − e.g., in a sentence such as Juan es videoaficionado 'John is video fan', it is not understood that John possesses all the videos he fancies.4(4) man-i-larg-o pisc-i-forme videoaficionado hand+I+long-m fish+I+form video+fan 'light-fingered' 'fish-shaped' 'video fan' (English examples: color-blind, stone-cold, home-made) The reason why the above compounds with modifiers and complements are exemplified in Spanish and English is that we will make use of the latter language throughout the paper as a reference for structural variation.I have observed that, whereas the modifierhead compounds are constructed similarly in both languages, the complement-head ones present some non-trivial semantic and phonological differences that will be related to the characteristics of their non-heads.
In section three, some previous morphological analyses of pelirrojo are summarized.We will focus on the proposals questioning the head status of the adjectives in the Spanish pelirrojo, such as Rainer and Varela's (1992) approach.It will be demonstrated that there is no convincing evidence to deprive the adjective rojo of its head status.On the contrary, it behaves as a head following the usual syntactic, semantic and morphological criteria, as several previous analyses of the compound have demonstrated (see Val Álvaro 1999 for a review).According to these criteria, pelirrojo is endocentric -headed by the adjective -because the compound is a singular, masculine adjective like rojo and denotes the property rojo, despite being applied to an individual.
Section four is concerned with the most noticed feature of Spanish (complementhead) pelirrojo compounds, which is the fact that they only codify inalienable possession relationships between possessees and possessors (5a).This is the reason why the Spanish example in (5b) is ill-formed.Inalienability is not obligatory in the English counterpart (5b).The term inalienable usually refers to the fact that certain kinds of nouns, such as pierna/leg, imply the existence of the entity to which they belong, whereas non-inalienable ones, such as alfombra 'carpet', do not. 55) a.In a sentence such as the one in (6a), some authors have identified a binding relationship between Juan 'John' and el pelo 'his hair' (Guéron 1985).Binding is the term used to describe how a Noun Phrase or pronoun, called the antecedents, are related to others of their kind, called anaphors, because of their reference.I suggest that a similar kind of relationship could be established in Juan es pelirrojo 'John is red-haired'.

Complement-head versus modifier-head compounds and attributive phrases
Pelirrojo is classified as an attributive compound in the more accepted typologies, since the relationship established between the adjective and the noun inside the compound is characterized as one of attribution (Bisetto & Scalise 2005).In (7) I exemplify the kind of compounds usually treated as attributive.
(7) palabra clave agridulce pelirrojo piel roja key word bitter-sweet red-haired redskin The aim of this section is to put pelirrojo inside a classification of the attributive compounds according to structurally relevant criteria.Although previous literature pays attention to the grammatical category of the constituents (Buenafuentes 2007;Moyna 2011) or to their internal relationships (Castronovo & Gariano 1997;Padrosa 2011), their observations are not intended to capture the existing contrasts between constructions regarding inalienable possession.I will distinguish three classes of constructions, complement-head compounds, modifier-head compounds and attributive phrases.

The complement-head structure
From a morphological point of view, it is possible to identify in Spanish different classes of compounds where a noun in the internal position behaves as the complement of an adjectival head.For instance, there are compounds headed by simple adjectives (8), and compounds headed by derived adjectives (9). 7In both groups it is possible to identify the native inalienable-only Spanish constructions of the pelirrojo class − (8a) and (9a) − which are the focus of this paper, and to distinguish them from the Classical ones of (8b) and the Latin/English-influenced ones of (9b-c). 8In (9) we could further differentiate the more numerous participial compounds from the denominal ones.
( As we can see, pelirrojo compounds share the use of i as a theme vowel with the other complement-head structures.However, the latter make use of alternative vowels, whereas the presence of i in Spanish pelirrojo compounds is mandatory.There are a few ancient examples among pelirrojo compounds that lack i (e.g., manvací-o lit.'hand empty-m', 'empty-handed'; cabizbaj-o/cabezcaído/cabizcaido lit.'head lowered-m', 'depressed'; bocabiert-o lit. 'mouth opened-m' 'open-mouthed').As Moyna (2011: 143) points out, these cases exist because pelirrojo compounds underwent a period of formal instability, but their current characteristics became established around the 16 th century.She suggests that the old examples were tied to different prosodic constraints on monosyllabic and three-syllabic nouns; monosyllabic nouns are now forbidden (I would create the neologism manivendado 'hand+vandaged', not *manvendado), whereas three-syllabic nouns (cabeci) are preferred over i's absence (cabiz) since the 15 th century.To account for their appearance 7 This distinction is in line with Moyna's (2011: 133), who differentiates between integral and deverbal compounds.The notion of integral nicely characterizes the pelirrojo type, as it contains both simple and deverbal adjectives (e.g., al-i-quebra-d-o lit.'wing+I+break+pp+m.' 'discouraged'). 8The Classical puntiforme 'point-shaped' and the English-influenced drogodependiente 'drug addict' are usually treated as subordinative from the point of view of the internal relationships (drug is an argument of addict), but in this paper they are also regarded as attributive.This decision is justified because the relationship between the adjective heads forme/shaped and dependiente/dependent and their required subjects, which are outside the compound, is one of attribution (external predication).In a phrase like John is tired of listening, tired of listening would be considered an attribute of John.The phrase would not be referred to as "subordinating" because it has a complement inside (of listening).There is no logical reason why the structure of a compound should be considered otherwise.It is chosen the general term attributive because the term possessive is often used to treat unrelated compounds such as craftsman or Turkish kadin doktoru 'gynecologist' (Yükseker 1987).
in synchronic Spanish, it is possible either contemplate these compounds as stored, monomorphemic historical compounds, or to explain the selection of stem allomorphs from a framework with competing constraints such as Stratal Optimality Theory (Bermudez Otero 2013), where a prosodically unmarked solution could be favored over a morphologically marked one, and vice versa.
From a semantic point of view, the noun complements of (9b) can be understood as adjuncts with different kinds of thematic roles (e.g., instruments 'made with hands', locations 'made at home'), whereas the ones in ( 8) can be understood only as themes 'red of hair', 'with the shape of a cancer').
Even if the non-heads in Classical compounds and native ones in (8) are provided with the same thematic role, a structurally relevant semantic contrast between them is still at work.Nouns like color and air appearing in the Classical constructions (10a) are absolutely prohibited in the pelirrojo structure (10b).There are in English some examples of compounds constructed with simple adjectival heads plus noun complements (11), and they allow that kind of nouns as well.Pelirrojo compounds are tied to special restrictions regarding the selection of complements.Authors such as García Lozano (1993) or Sánchez López (2003) have claimed that the nouns appearing in the compound are always among those classifiable as inalienable (12), which include integral parts of humans (barba 'beard'), animals (pico 'peak'), objects (punta 'point') and some clothes (capa 'cape').These authors consider that nouns such as color, air, tree or stone cannot be interpreted as inalienable possessions and that that fact alone explains why these nouns do not appear in the pelirrojo pattern.
But we would leave the truly crucial issue unanswered, namely how and why the Spanish construction, and not the others, is tied to the inalienability restriction.This question will be addressed in sections three and four.

The modifier-head structure
Apart from the complement-head compounds, it is possible to find in Spanish some examples of compound parasynthetic constructions, that is, structures formed merging two lexical units plus a derivative morpheme.Overall, these structures are much more productive in Spanish with prefix modifiers and quantifiers (13a), and there is also a considerable number of Classical examples (13b), but there are very few examples that contain "proper" adjectives, such as the ones in ( 14).The resemblance between the Spanish modifier-head constructions and the English ones is apparent, as we can deduct from the examples and their corresponding translations in ( 13).Nevertheless, the structure with proper adjectives is much more productive in English ( 15) than in Spanish ( 14).
(15) left-handed, kind-hearted, good-natured, wide-eyed, short-sleeved The complement-head structure of pelirrojo seems to be the default one in Spanish to create attributive-descriptive compounds with nouns and adjectives, whereas it is the modifier-head one in English.In fact, the semantic counterparts of a considerable number of pelirrojo compounds can be found in the class represented in ( 15), but they are not attested in the complement-head structure of airsick and stone-cold.
Nevertheless, the modifier-head structure does not seem to force an inalienable relationship in either Spanish or English.We can attribute the property of being ill-fated to a family and we can also attribute the property of being altoaragonesas to costumbres 'traditions'.But neither families possess fate nor costumbres possess Aragon.
We can conclude that inalienability is not mandatorily codified in modifier-head structures.Therefore, there are empirical grounds to propose that red-haired and pelirrojo must be treated as different structures, which will be discussed in section three.

Attributive phrases
We would like to comment on a last kind of attributive constructions, those exemplified in ( 16), although they would not be considered compounds, but agglutinated phrases (see also Moyna 2011: 118).( 16) cara-dura cara-pintada cabezas locas face-fem+hard-fem face-fem+painted-fem head-fem.pl+crazy-fem.pl'cheeky' 'extreme right wing' 'dim wit' piele-s roja-s skin-fem.pl+red-fem.pl'redskin' The main reason to exclude them from compounding is the fact that they show agreement between members, or traces of it, even when the agglutination process has resulted in monographic and/or single-stressed words.I follow Rainer & Varela (1992), among many others, in considering that, whenever we find internal agreement, we are dealing with phrases and not with compounds.Agreement relationships indicate that the constituents in ( 16) establish a different syntactic relationship between them (an internal subjectpredicate relationship, against the complement-head one of pelirrojo).
But there are other reasons to exclude these constructions from the pelirrojo class, even if the noun heads in these constructions are certainly inalienable possessions of their referents.
First, the morphophonological instability in N+Adj agglutinated phrases continues in synchronic Spanish (Buenafuentes 2007), but, as we indicated previously, it finished long ago for pelirrojo compounds.Second, there is an obvious relation between the constructions in ( 16) and others such as el Pirata Barbarroja 'Barbarrosa the Pirate', Margarita seis dedos 'Marguerite six-fingers', (ser un) cuatro ojos 'to be a four-eyes', or (ser un) manos de árbol lit.'to be hands of tree', 'butter-fingered'.All these constructions reveal the two main functions of phrasal metonymies: to form epithets and/or to insult people.The function of pelirrojo construction is, on the contrary, mainly descriptive (García Lozano 1993).

Variation in complement-head structures
In section 2.2, we saw that pelirrojo compounds differ from Classical, English/Latin-influenced and English complement-head compounds because they are the only ones that always codify inalienable relationships and select a i theme vowel.The aim of this section is to propose differentiated structures for them in order to capture those differences.
Pelirrojo is the most complex of the structures because it contains a two-layered complement, peli, the result of merging the noun pelo with a relational category i called inalienable Possession (henceforth iPoss) (17). 9The starting point of my analysis is Myler's (2016) approach to possession constructions.The author proposes that the codifying of possession relationships varies depending on the functional categories that introduce them (their category, the size they allow for their complements, whether they take possessees or modified possessees, and so on).It departs from it slightly, however, as I assume that the possibility of codifying inalienable possession is dependent on the presence of the iPoss head, that is, it does not originate in the lexical information of nouns like pelo (Adger 2013).
The head of the compound, rojo, is a predicate that must take a subject (e.g., Juan), but not a complement.In that sense, it differs from other adjectives such as próximo 'close', which requires both a subject and a complement.So, from the point of view of rojo, the complement peli is an adjunct and not an argument − John is red regarding his hair.Be that as it may, both pelo and rojo must be present for the compound to be possible: that fact distinguishes pelirrojo from other possession structures such as the derived with -ado and -udo (barbado/barbudo,'bearded', deslenguado 'foul-mouthed'); -ado constructions may or may not have a modifier, -udo constructions cannot (see Myler 2016: 370-ss.for a more detailed analysis of this contrast).
To solve this puzzle (why the adjunct of rojo, peli, is obligatory), I propose that the inalienable possession head merges with rojo forming a second predicate or co-predicate with it, which is the same analysis that Myler proposes for obligatorily modified possessees.
The iPossP introduces a relation in need of a possessor, but this possessor is not provided directly in the specifier position of iPoss., but later in the specifier position of another relational projection, PredP, introduced by rojo itself.10This phenomenon is called Delayed Gratification (Myler 2016: 444): It happens whenever an element is the semantic argument of a low projection (iPossP), but it is the syntactic argument of a higher one, which saturates its thematic role.When the projections coincide, we speak of instant gratification instead.
I assume that iPoss is spelled-out by i in the Spanish compounds of the pelirrojo type.In other languages, such as Catalan, there exist pelirrojo compounds, but they have no i (except for Spanish borrowings) (Padrosa 2011).I suggest that the structure of Catalan compounds is the same, but it lacks a specific exponent for lexicalizing it.IPoss may be lexicalized by noun vocabulary entries if we accept the nanosyntactic lexicalization principle of Phrasal Spell-out (lexical entries can lexicalize non-terminal nodes) (Fábregas 2016), or we can simply propose a null iPoss head.The specifics about iPoss will be developed in section four.The focus will be on how i introduces inalienable relationship and determines why certain classes of nouns, and not others, are treated as inalienable possessees. (17) Structure for inalienable-only complement-head compound pelirrojo The complement color in English projects an NP that can be merged with the adjective head as any non-head does in a typical root compound, that is, without explicit relational material or significant meaning restrictions (there are alternative readings to the theme interpretation of color in color blind, for example, air can be interpreted as the cause of sickness in airsick) (18).
(18) Structure for English complement-head compound color blind The complement aeri in Latin projected a PP/KaseP (Svenonius 2010) but it does not now, as case-marking of nouns in Spanish is no longer active.11This fact would make the structure of aeriforme as simple as color-blind's, although forme is not an existing adjective in Spanish − we capture that fact considering forme a predicative category-less head that introduces the external subject − and the theme vowel i has remained, together with the theme-only interpretation of the noun complement (19).12 (19) Structure for Classical compound-head compound aeriforme Summing up, throughout this section it has been shown that the morphological traditions of languages do not by themselves justify the variation inside compounding, as the modifier-head structure is attested both in Spanish and English sharing their main characteristics, whereas the complement-head structures vary regarding their semantic interpretation and their morphophonological shape.
3 Red-haired and pelirrojo's structure cannot be the same

Similar structures, different morphophonologies?
Red-haired and pelirrojo are prototypical examples of productive patterns of compounding with adjectives in their respective languages.The fact that they denote the same kind of individual with the same lexemes may reasonably lead us to conclude that they are formed equally.Serrano (1996) states that, despite their common denotation, compounds formed in the Germanic pattern of red-haired are quite distinct from pelirrojo.The key of his differential analysis is the noticeable absence of true parasynthesis in Spanish compounding, which leads the author to conclude that pelirrojo cannot be endowed with a parasynthetic structure such as the one of red-haired.
More recently, however, Gil Laforga (2014: 269) maintains that red-haired and pelirrojo only differ because of the position of the possessive morpheme: The English -ed is a suffix, whereas the Spanish i is a medial affix.She suggests that the analysis proposed for pelirrojo compounds in her work can be successfully implemented in the red-haired type.Both compounds would be paraphrased as in (20). (20) Juan es pelirrojo.→ Juan es de pelo rojo.
John is red-haired.→ John has red hair.

Italics: Possessive predication Bold: Attributive predication
As in (20), these compounds would establish two predications.About pelo/hair, the property of being rojo/red is predicated.About Juan/John, the possession of pelo/hair that is also rojo/red is predicated.This interpretation of the structure of pelirrojo is not shared by other authors.Moyna (2011: 137-140), among many others (Alonso Cortés 1987;Sánchez Lopez 2003), maintains that the structure is headed by the adjective and that the noun is a complement of it.This structure allows the paraphrasing in ( 21).
(21) Juan es rojo de pelo → John is red of hair In the previous section, a structure was proposed that favors the paraphrase in ( 21), since it represents a complement-head structure. 13 Even if Gil Laforga and Moyna do not (explicitly) adopt syntacticist approaches to compound formation, both accept the presence of small clauses (subject-predicate structures) in the internal structure of pelirrojo, a necessary apparatus to support their predication hypotheses, but hard to conciliate with the idea that compounds are constructed differently than phrases (an idea they support).
According to the former author, an adjective such as rojo/red needs relational structure to predicate from its subject pelo/hair (Pred1), and more relational structure is needed to predicate the possession of the complex rojo+pelo from a subject like Juan/John (Pred2), ( 22). ( 22) Structure for red-haired (and pelirrojo), adapted from Gil Laforga (2014: 270) 13 Nevertheless, there are some non-trivial differences between pelirrojo and red of hair.For that reason, the parallelism should be kept to a minimum.For example, *ser azul de ojos 'to be blue of eyes' is said to be ill-formed, whereas the compound ojiazul is not.Español Echevarria (1997) states that the structure 'to be blue of eyes' requires using gradable adjectives that, when applied to the possessee, affect to the possessor extensively: ancha de cadera 'wide of hip' would be well-formed because the width of the hip contributes to the overall width of the possessor.The compound does not have to fulfill this condition because any adjective can appear and i is responsible for the choice of the possessee.In contrast, de has no responsibility at all over the possessee: corto de entendederas 'short of understanding' has an abstract noun that is inconceivable as possessee inside the compound (*entendedericorto).Despite these facts, we can safely maintain that pelirrojo most resembles to the one-predicate structure ser ancha de cadera 'to be wide of hip' than to the double-predicate structure tener los ojos azules 'to have the eyes blue': the first one only codifies inalienable possession, the second one does not (Juan tiene la copa vacía 'John has the cup empty').The first one forbids the article in the complement projection (*ser ancha de la cadera '*be wide of the hip'), the second one does not (have the cup empty); In the first one, the complement is introduced by a relational category (de/of), in the second one is not.Finally, the adjective agrees with the possessor in the first one (María es ancha… 'Mary is wide.fem…'),but with the possessee in the second one (Juan tiene la copa vacía 'John has the.fem cup empty.fem').Pelirrojo, then, share four characteristics with 'to be wide of hip': It is inalienable, it does not introduce the possessee with an article, but does introduce it with a relational category and, finally, it displays possessor-head agreement.
But if we expect ( 22) to be the structure of both red-haired and pelirrojo, we lose the possibility of explaining the differences between them appealing to the medial affix (i) as opposed to the suffix (-ed) condition of the possessive morphemes, which was Gil Laforga's original intention.
The structure in ( 22) predicts an asymmetry between semantic and phonological representations in the Spanish case, because if the possession relationship dominates the syntactic constituent [pelo rojo], the isomorphic phonetic spell-out of the possessive morpheme i should occur over the whole, delivering a non-existent *pelorroji.The term spellout describes the translation of the syntactic construction into phonological substance.
Anyway, the proposal in ( 22) seems to be less troublesome for red-haired than for pelirrojo, as it clearly states that the noun behaves as the subject/external argument of the adjective.We cannot find a single unambiguous case of a lexical compound with a noun behaving like a subject/external argument in Spanish ( 23). ( 23) robamaridos 'Woman who steals other woman's husbands.'(husbands is the stolen thing, the theme) *'Husbands who steal women.' (husbands would be the external argument/the burglars) castellanohablante 'speaking the language of Castilians' (Castilian is the spoken thing, the theme) *'Castilian person who speaks.' (Castilian would be the external argument/the speaker) Gil Laforga already notices some of the paradoxical aspects of her analysis and suggests that the non-availability of structurally-isomorphic forms like *pelorrojado/pelorroji lit.'hair-redded' is an ad hoc prohibition against Spanish possessive suffixes merging with complex stems (Gil Laforga 2014: 117).14But this is not true, because the Spanish -ado can merge with complex stems,15 such as the ones in (24): (24) malhumorado deslenguado bad+temper+pp without+tongue+pp 'bad-tempered' 'foul-mouthed' An alternative answer, which was provided in the previous section, is that a word such as deslenguado exists, and not *pelirrojado, precisely because the examples in (24) share the modifier-head structure of red-haired (not pelirrojo).In fact, Nevins & Myler (2014) and Myler (2016) propose an analysis of red-haired which is more akin to the modifier-head analysis put forward in the previous section.According to these authors, -ed is an adjectival head that takes a complement of compound size (red-hair) and not bigger (*John is lovely big blue-eyed). 16e should remember that the semantic-phonological asymmetry simply arises if the structure proposed for pelirrojo is the same as the one proposed for red-haired.Therefore, we think it is advantageous to assume that only pelo (and not pelo+rojo) is within the scope of the possession relationship introduced by i.In my proposal, this is predicted by the fact that i only takes pelo as its complement.

Exocentric pelirrojo versus endocentric red-haired?
At least since Bloomfield (1933) the lack of head in constructions from the semantic point of view is called semantic exocentricity (e.g., the fact that a petirrojo lit.'breast-I-red' is not a kind of 'red' but a 'robin').Other authors also take into account categorial exocentricity, for instance, the absence of an affix head -ed in cases such as grey beard (Olsen 2001).Rainer & Varela (1992) propose that pelirrojo compounds are exocentric in both senses.The idea is controversial because in pelirrojo there is a plausible candidate to be the head of the structure: The adjective agreeing in number and in gender with the external argument of the constructions in ( 25). ( 25) Los chico-s so-n pelirroj-o-s.det.m.pl boy.m.pl be.pl hair-red.m.pl 'The boys are red-haired' La chic-a es pelirroj-a Det.F.S girl.f.s be.s hair-red.f.s 'The girl is red-haired' The authors find one interesting argument to support the categorial exocentricity, which is the fact that the structure apparently cannot be affixed (26).17But the affixations in ( 27) are perfectly fine for Spanish speakers, so the weirdness of cuellilargura should be related to stylistic reasons.
(27) pelirrojaza 'gorgeous red-haired woman' pelirrojillo 'little red-haired' pelirrojez 'redhairedness' pelirrojitis 'obsession that some men have with red-haired women (or vice versa)' Summing up, Rainer and Varela (1992: 133) and Gil Laforga (2014) share the assumptions that the noun in the compound satisfies the external argument of the adjective and that the adjective is not the head in the compound.They differ because the former consider that the compound has no head and the latter thinks that the affix i is the head.I have argued against these approaches.Additionally, my proposal of the complementhead structure presented in section two is reinforced by the fact that the more convincing pieces of evidence used to argue the non-head status of rojo or the subject condition of pelo do not actually justify the intended structure, because they also work nicely with the complement-head one.
The first piece of evidence offered by Gil Laforga (2014: 74) is that, in a compound like ojizarco (30), the selection of the adjective zarco depends on the internal noun ojo (28a) and is independent from the external one people (28b).That would justify the subject-predicate relationship between ojo and zarco.

b. People can be ojizarcos
This evidence is weak because the selection of the adjective when it is complemented by a noun can also be influenced by the complement, as in ( 29).That phrase is headed by the adjective zurdo and the internal noun pie acts as its complement (29a).When we change the noun pie with the noun nariz in (29b) the choice of the adjective zurdo becomes strange.
'There are left-nosed people (they only breathe through the left nostril).' The piece of evidence that could demonstrate that the adjective is not the head of the compound is that its distribution changes in the compound.Roto 'broken' functions with the copula estar 'be' in phrases (30a), but the same adjective can function with ser 'be' in compounding (30b) (Gil Laforga 2014: 66-67).If the adjective was the head, the copula would be the same it normally selects.
(30) a.El juguete está roto.The toy was.estar broken 'The toy is broken.' b.Juan es un man-i-rrot-o.John was.ser a hand+I+break+pp+m 'John is a spendthrift.'But in (31) we can check that the change of copula is not related to headness.Cerrado 'closed' typically functions with estar in the syntax, like in (31a), but it can function with ser when the adjective is complemented (31b).If the change of copula can be related to a complement-head structure, as in (31b), then the complement-head structure for pelirrojo could be supported instead.
John was.estar closed to love 'John was not willing to fall in love ever again.' b.Juan *estaba/era cerrado de mollera.John was.ser closed of brain 'John was dim-witted.'

Interim conclusions and prospects
Throughout this section, I have argued against unifying red-haired and pelirrojo structures.I have proposed instead that peli is a complement of rojo in Spanish, whereas red is a modifier of hair in English.From this perspective, several analytical problems were avoided.
The relationship between red and hair in red-haired was conceived as a modifier-head structure and not as a subject-predicate one (small clause approaches).Among other reasons, we should expect agreement relationships inside compounding if we choose the subject-predicate approach, even if English does not exhibit them.Agreement is hardly a property to expect from lexical compounds.Besides, red-haired displays the default modifier-head order of English syntax.The subject-predicate proposal will predict hair-redded, since subjects precede predicates in English (Cinque 2014). 18elirrojo's order, being a complement-head structure, remains problematic, as it is not attested in Spanish syntax (32).
(32) English head+complement order: the book of Mary English complement+head order: Mary's book Spanish head+complement order: el libro de Maria Spanish complement+head order: *de Maria libro As a matter of fact, every noun+adjective combination of (33a) is a right-headed complement-head structure.These constructions are in no reasonable way linkable to the left-headed noun+adjective constructions of (33b).Constructions in (33a) have only one main stress, are semantically compositional and their constituents do not display noun+adjective agreement.Constructions in (33b) have two main stresses, are semantically idiosyncratic and their constituents agree.
(33) a. videoaficionado pelirrojo canceriforme video+fan+m red+I+hair+m cancer+I+shaped 'video fan' 'red-haired' 'canceriform' b. prensa amarilla ensaladilla rusa press+fem yellow+fem salad+small+fem russian+fem 'gutter press' 'yellow' 'Russian salad' The more plausible explanation for pelirojo's order is found in the Latin-imitating Humanistic origin of the compound.Although the right-headed deverbal compounds such as videoaficionado can be related to the contemporary cultural influence of English-speaking countries, that is not the case of pelirrojo constructions.Pelirrojo compounds are not borrowed from any language, but began to be productive in the Humanist period, when Classical languages had a powerful and even increasing influence. 19In (34) we provide examples of the kind of Classical compounds that would have served as models for pelirrojo.20 (34) Meyer Lübke (1895: 72) barb-i-rras-us al-i-pes beard+I+shaved wing+I+footed 'without a beard' 'wing-footed' We cannot discard the possibility of the default noun+adjective order of Spanish attributive structures influencing the noun+adjective compounding.That idea might be supported by the fact that the first attested (unproductive) compounds, between the twelfth and fourteenth centuries, have no i (35a), although i was absent, too, in the Greek Accusative constructions of (35b), which has been said to influence pelirrojo.

Inalienable possession
Throughout the previous sections, it has been shown that the only compound that forces one of its constituents to behave as an inalienable possessee is the pelirrojo type.It was anticipated that the reason why this happens is that pelirrojo contains a non-head provided with more complexity, more concretely, a Noun Phrase plus the relational category spelled out as i.This point will be developed further along the section.Additionally, we will link the inalienable specificity in compounding to similar phrasal phenomena.

Inalienable vs non-inalienable possession
A typical or non-inalienable possession relationship is never entailed by the possessee (bolded).We understand that there is a possession relationship in (36a), but a location one in (36b).
(36) a. el coche de mi primo (possession) 'my cousin's car' b. el coche de la esquina (location) 'the car on the corner' If we search for similar non-inalienable relationships on compound patterns, we find that head would be in a possession relationship with woman in a wrongheaded woman (as women possess heads), but not with policies in wrongheaded policies (as policies do not possess heads).
Regarding inalienable possession, it is usually said that the relationship is entailed by the selection of the possessee.This fact can be syntactically represented making the possessee select the possessor as its argument (Vergnaud & Zubizarreta 1992;Alexiadou 2003) (37a).Thus, inalienable relationships are on a par with the ones between deverbal nouns and their complements (Barker 2008) (37b).
(37) a. la boca del paciente (patient is the argument of mouth) the mouth of.thepatient 'the patient's mouth' b. la decisión del presidente (president is the argument of decide/decision) the decision of.thepresident 'the president's decision' These "possessee-selecting-the-possessor-from-the-lexicon" approaches are supported by the seemingly universal preference to perceive specific classes of nouns, such as body parts of humans, animals, or kinship terms (38), as inalienable possessees.If, as (39) shows, even the most typical alienable nouns allow non-inalienable uses, we cannot codify them in the lexicon as always taking possessors as arguments, as it would overgenerate inalienability (see also Adger 2013).
An alternative is to maintain that there are indeed conceptual classes of nouns that favor inalienable uses, but that these uses are only compulsory in specific inalienable-only constructions.
An argument supporting this idea is that the conceptual noun classes treated as inalienable are variable among constructions, so we cannot predict the conceptual classes of the possessees that will appear inside them.
For example, the conceptual restrictions imposed by Spanish pelirrojo on possessees are quite severe, since a much wider amount of noun classes can be treated as inalienable possessees (Moyna 2011: 88).In the Italian dialects of Sardinian, for instance, we can find compounds translatable as tronqu-i-blanc-o lit.'trunk+I+white+m' or manill-i-larg-o lit.'handle+I+long+m' (Pinto, Paulis & Putzu 2011), whereas the conceptual classes to which the possessees trunk and handle belong are absent in the Spanish pelirrojo.
( Summing up, it does not suffice to say that pelirrojo codifies inalienable possession relationships because the nouns that appear in its internal position, such as pelo 'hair', are inalienable nouns.Although that constitutes a widely-held view, we need to supplement it with structural conditions.I argue that the access to the inalienable-only interpretation of nouns is structurally, not lexically, provided. of life'.The problem is that they do not constitute homogenous conceptual classes.Not even in the clearly polysemic pairs of (39) do we consider a good option to characterize the difference between alienable and inalienable uses through different lexical entries (see Vergnaud & Zubizarreta 1992: 596).Some authors have proposed that affixes such as -(c)ión determine the argument-taking condition of derived nouns like decision in phrases like the president's decision (Borer 2013).The reason is that event nouns without suffixes like war never take arguments.If we generalized that analysis to our data, we would say that it is the relational head i, which we named inalienable possession, that forces its complement (the possessee) to take an argument (the possessor).That would result in the obligatorily inalienable interpretation of the relationship.It would be the relational category spelled out as i the responsible for making its complement pelo be an inalienable possessee, not its noun class. 22 A consequence of that kind of approach is that it allows us to explain why the complement-head structures of videoaficionado and color-blind are not inalienable, since there is no relational category i.In addition to this, the I in canceriforme or maniobrado compounds, which it is not an iPoss, but a likely old case marker, determines the interpretation of the thematic role of the complement, but does not impose inalienability.
In this section, I have focused on why some compounds codify inalienable-only relationships and why the others do not.In the following section, I try to answer the question of how the relationship between possessor and possessee is established.

The syntactic distinctiveness of inalienability
Let us consider the behavior of pelirrojo in the context of inalienable phrasal phenomena.
We can find inalienable-only constructions in different kinds of languages, such as the double accusative constructions in Korean (41), in which alienable relationships are not codified (42) (Tomioka & Sim 2007: 2 The literature has paid special attention to Romance external possession structures (43).The possessor in these constructions tends to be the subject or the indirect object and it can be both nominal and pronominal.The possessee is a DP direct object.The contrast between ( 43) and ( 44) illustrate the different interpretations available for the inalienableonly constructions in (43) and the ones in (44). 22I would like to clarify that I differentiate the syntactically relevant aspects of meaning (e.g., alienable vs. inalienable contrast) from those that are not syntactically relevant (e.g., if the inalienable possessee is part of a human, an animal or an object, if it is material or immaterial, etc.).For that reason, I am not concerned with the variety of concepts appearing as internal nouns in pelirrojo, which can be treated as a purely conceptual phenomenon. 23See Gebregziabher (2012), who offers a review of Tigrinya, an Afroasiatic language of the Ethiopian Semitic branch, mainly spoken in Eritrea and northern Ethiopia.In that language, the syntactic differential behavior between inalienable and alienable constructions is rather striking.
The more remarkable fact about the inalienable constructions in ( 43) is, in fact, that their English equals with the intended meaning are constructed as internal possession structures: the possessor (determiner) is in the same phrase as the possessee noun, not outside like in the Spanish examples (see Guéron 1992: 135-138 for more details).
It is opportune to ask then if there is any relationship between this contrast and the different compounding patterns displayed by both languages.Guéron (1992: 141) stipulates that the nature of the external possession constructions in Romance constructions is different from the same relationship in English.In the former language, it is an argumental binding (A-binding) relationship established between subjects/indirect objects and direct objects (45a).In the latter, it is not, so only binding with adjuncts (A'-Binding) is possible (45b).Both Guéron and Vergnaud & Zubizarreta (1992) share the assumption that the differences involve the definite article.In rough outlines, for Guéron only the Romance definite article la can be non-referential, which makes a feature agreement relationship between le and cabeza in (45a) possible by not interrupting it.The relationship cannot cross the English (referential) the in the head, which would explain its ill-formation.For Vergnaud & Zubizarreta (1992: 635) only the Romance definite DP's can denote types (and not tokens), and being a type depends on the Romance article being an expletive in (45a), a determiner with no meaning or referential features.However, if we accept any of these proposals, we cannot link these differences between English and Spanish with the compounding facts.The reason is that compounds do not contain a DP-possessee either in English or in Spanish (*elpelorojo/*red-thehaired).
Le Bruyn & Alexandropoulou (2013), on the other hand, argue that the differences should not be linked to the determiners, but to the resources available to license the relational/inalienable 24 interpretation of nouns.For instance, in Spanish, a word like rebanada 'slice' can be interpreted relationally in internal possession structures, when possessor su 'his' and possessee 'slice' are in the same phrase (46a), but also in the external ones, when the possessor (me) is outside the Noun Phrase and behaves as the argument of a pronominal verb (46b).As English verbs are non-pronominal, hand cannot license its relational interpretation as the Spanish word mano in (46b).
For my proposal, the crucial point is that the external possession constructions with PP possessees is acceptable in both languages (46c).The authors blame the preposition on/en for guaranteeing the linkage between John/Juan and head/cabeza (Le Bruyn & Alexandropoulou 2013: 326).In section two, we analyzed peli as a noun projection consisting of a Noun Phrase plus a relational possessive head.Thus, i in peli was provided with relational content and it is responsible for the adjunct role of peli in relation to rojo, the same as en/on in the PP en la cabeza/on the head (46c).Following Den Dikken's (2006) approach to relators, we expect the relationship between possessor and possessee to be always mediated by a relational unit, no matter how it is spelled-out (as a P like on or as a suffix like i).
The binding approach assumed by Guéron could be, nevertheless, an alternative account for the fact that i makes pelo be an inalienable possessee of Juan in a sentence like Juan es pelirrojo.Forthcoming studies are welcomed to conciliate the existence of binding with the alleged absence of determiners inside compounds.
Summing up, the data analyzed through section four is intended to complement the proposals made in sections two and three in several ways.Firstly, I have characterized pelirrojo as an inalienable-only construction with its own conceptual profile, as opposed to non-inalienable compounds or other inalienable structures.Secondly, I have stated the correlation between the presence of the functional category i and the inalienable nature of pelirrojo.
I have showed that precisely the kind of relational categories to which i has been traditionally related, that is, prepositions like de 'of', are required to codify inalienable external possessor relationships whenever the possessee is not an argument of the verb.That happened in constructions such as I hit John on the head, it would be the case of María es ancha de cadera.'Mary is wide of hip.' and, presumably, the adequate analysis for pelirrojo.

Conclusions
In this paper, I have argued that the structure of lexical compounds like pelirrojo is syntactic.As such, we can differentiate structures with different categorial projections and variable degrees of complexity.It follows from this kind of approach that compound types can be characterized according to their distributional properties and their structural semantics in unprecedented ways.
Concretely, I have argued that pelirrojo and red-haired are completely distinct from the point of view of their syntactic, semantic and phonological properties.Differences arise even when they are compared with Classical compounds such as canceriforme and English complement-head compounds such as airsick.All of them differ from pelirrojo because they are not inalienable-only constructions.
Even the lexicalist approaches frequently acknowledge that the grammatical relationships inside a compound are syntactic-like.What they cannot possibly explain, however, is the layered and complex structure that a lexical compound like pelirrojo seems to possess, according to our data, and the presence of relational categories with the properties we have attributed to i.
I have tried to demonstrate that the relationship between the possessor and the possessee inside pelirrojo may be considered as an argument against some lexicalist frameworks.Should the difference between phrases and compounds be troublesome for establishing certain kinds of long-distance relationships?The answer is no, according to Uriagereka (1999).The reason why the internal constituents of a compound do not usually participate in coreference relationships, undergo movement, or internal modification is not that they are formed using different rules, but the fact that the relevant syntactic features are absent.

(
se cortó las venas.('John cut *the/his veins.')b.Pluto se perseguía el rabo.('Pluto went after *the/its tail.')(44) a. Juan se llevó al padrino.('Juan took his/their/the best man.') b.María me trajo el coche.('Mary brought my/Mary's/anybody's car.') golpeé la cabeza (A-binding) *I him i .hit the head i 'I hit his head.' b.I hit John on the head (A'-binding) su rebanada.'He i ate his i slice.' b.Me i corté la mano i .(*I me i cut the hand i ) 'I cut my hand.' c.Golpeé a Juan i en i la cabeza.'I hit John i on i the head.' Juan es piernilargo porque (él) tiene las piernas largas.
b. María le golpeó la cabeza a Juan.(The head can only be John's) Mary him kicked the head to John 'Mary hit John on the head.' c.Juan trajo el coche.(The car can be John's or anybody's) John brought the car 'John brought the car.' But some problems remain with lexically specifying the "special inalienable condition" of any of the nouns in (38): All of them allow non-inalienable uses, however metaphoric, (39), since manos/hands are not possessed by cartas/cards, colas/queues are not possessed by metros/undergrounds and padres/fathers are not possessed by vidas/lives. ).23