1
Compounding in European Portuguese – a Distributed Morphology approach
Catarina Loureiro Soares
Supervisor: dhr. dr. Jan Don Second reader: dhr. dr. Mauro Scorretti
Abstract
This thesis provides an account of compounding in European Portuguese within the framework of Distributed Morphology, drawing mainly on Harley (2009). It provides a novel account for the ban on DP incorporation in English based on a genericity restriction on synthetic compounds (the Genericity Principle), which correctly derives the Portuguese and English data. The analysis of Portuguese primary departs from Harley’s in that it relies on separate workspaces and direct merger instead of incorporation. A new account of prepositional compounds is proposed which makes a
distinction between argumental and modificational prepositional
compounds, and assigns each subtype a different syntactic structure.
1. Introduction
The aim of this thesis is to provide an account of compounding in European Portuguese within the framework of Distributed Morphology. Compounding is understood here according to Harley’s (2009) definition:
2
(1) Compound (Harley 2009:130)
A word-sized unit containing two or more Roots.
This means that compounds are morphologically complex structures in the sense that they contain more than one Root, and furthermore that these structures behave
syntactically1 and/or phonologically like a word (hence word-sized). I say “and/or”
here, differently from Harley, who just uses the conjunction, because European Portuguese compounding does not seem to have phonological repercussions - we may expect different phonological behaviour for compounds, but the literature on Portuguese seems to suggest otherwise (Rio-Torto and Ribeiro 2012:126). However, we should keep in mind that any potential phonological differences might be translated into rather subtle and hard to notice phonetic differences.
The thesis is organized as follows. Section 1.1 provides the core tenets of the framework (Distributed Morphology). Section 2 presents Harley’s (2009) account of compounding in Distributed Morphology, specifically, of synthetic (2.1), modificational synthetic (2.2), primary (2.3) and phrasal (2.4) compounds, along with some preliminary discussion of some of the issues of her proposal. The following sections are then centered on Portuguese compounds. Section 3 deals with Portuguese synthetic compounds, the analysis of which brings about problems for Harley’s theory. I propose that synthetic compounds are subject to a semantic constraint such that their interpretation must be generic, and argue that this correctly derives the Portuguese and English data (which Harley 2009 is based on). Section 4 analyses Portuguese primary compounds and discusses the need for different structures for synthetic and primary compounds (contra Harley 2009). Section 5 provides an account of prepositional compounds in Portuguese which divides this type of compound in two subclasses
3
(argumental and modificational prepositional compounds), and which proposes different structures for each. Section 6 concludes.
1.1 Introduction to Distributed Morphology
Distributed Morphology (DM) is a theoretical framework which posits that the architecture of grammar is such that the mechanism responsible for generating word structure is the same mechanism responsible for generating phrase structure (i.e syntax). This is known in the literature as syntax-all-the-way-down – syntax operates both above and below the word-level, which means that (part of) word formation is syntactic. In this theory, morphology applies on the output of syntactic derivations – the syntax builds complex heads, which are then sent to Spell-Out, at which point several morphological operations may occur. The resulting structures are subsequently sent as input to the phonology. This is illustrated in (1), taken from Bobaljik (2017:1).
(2) Architecture
Syntactic derivation
Output (Spell-Out)
Morphology
Phonology Semantics
The fact that word formation is syntactic means that, in the simplest of cases, morphological structure corresponds to syntactic structure. In more complex cases, additional morphological operations apply at the PF branch. It is in this sense that
4
morphology is distributed in this framework – word formation processes are distributed over several points of the architecture of grammar.
The objects manipulated by the syntax into morphological structures (functional morphemes) are abstract features, which means they have no phonological content. The process of assigning phonological content to the syntactic terminal nodes (vocabulary insertion) takes place post-syntactically. This is called Late Insertion and is one of the key hypotheses of DM.
The Lexicon is also distributed in this framework, since what is usually thought of as the “lexicon” is in this theory comprised of three distinct lists: first, there is a list of syntactic atoms – these are manipulated by the syntax and correspond to features (e.g [PAST]) or bundles of features (which can be language specific, e.g [INFL]) which project to a syntactic node; secondly, there is a list that pairs morphosyntactic features with their phonological exponents, called the Vocabulary; finally, there is a third list containing idiomatic meanings of morphemes - the Encyclopedia.
In DM there are no categories such as verbs or nouns present in the Lexicon – these categories are derived in the syntax by merging Roots with categorizing heads. Roots “are sequences of complexes of phonological features, along with, in some cases, non-phonological diacritic features” (Embick and Noyer 2005:5), such as √CAT or √BUTTON, for instance. They become categorized when merged with categorizing functional head such as nº (which derives a noun), vº (which derives a verb) and aº (which derives an adjective). Taking the Root √BUTTON as an example, it can be used as a noun in sentences like A shirt has buttons, which means its structure is
[√BUTTON]√ n]n, or as a verb in sentences like I buttoned the shirt, in which case the
5
heads can be merged on top of other categorizing heads. Consider, for instance, the word pennilessness – the Root √PENNY is first merged with a categorizing nº head realized as ∅, then with categorizing aº -less, and finally with categorizing nº ness, yielding the structure [[[[√PENNY]√-∅]n-less]aness]n.
Vocabulary Insertion (VI) or exponence is the process applying on the output of syntactic derivations, which pairs the abstract features in syntactic terminal nodes to phonological underlying representations (subject to further derivation in the phonology). Each specific instantiation of such pairing corresponds to a Vocabulary Item (VI), exemplified in (3):
(3) [X] ⇔ y
The object to the left of the arrow corresponds to an abstract feature (e.g [PLURAL]), while the object to the right of the arrow corresponds to a phonological exponent that can realize this abstract feature. In English, for instance, this would correspond to (4):
(4) [PLURAL] ⇔ -z
(4) would add the phonological exponent /-z/ to a syntactic terminal node containing the feature [PLURAL]. The mechanism which rules whether a given vocabulary item (or exponent) is eligible to spell out a terminal node is the Subset Principle, which states that a vocabulary item can spell out a terminal node if it matches all or a subset of the grammatical features specified in that node.
(5) Subset Principle (Halle 1997:128)
The phonological exponent of a Vocabulary item is inserted into a morpheme in the terminal string if the item matches all or a subset of the grammatical features
6
specified in the terminal morpheme. Insertion does not take place if the Vocabulary item contains features not present in the morpheme.
Vocabulary Insertion can be less straightforward, however, in situations in which there is a mismatch between the list of abstract syntactic features and the Vocabulary, or, in other words, when there is no one to one mapping from one list to the other. This can happen in one of two ways: if there is a many to one mapping, in which case there is more than one feature in the syntax being realized by a single phonological exponent:
(6) [X Y Z] ⇔ w
Underlying (in the syntax) there are more contrasts than realized (in the morphophonological form). There can be mismatches between the lists in the opposite direction – a one to many mapping, in such cases the same abstract feature can be realized by more than one phonological exponent. This can be due to the fact that there are several Vocabulary Items specified for the same abstract feature that apply in different contexts (allomorphy) or that the VIs share abstract features. The fact that there can be more than one VI to realize a given syntactic node stems from the Subset Principle - when the morphology is trying to spell out a terminal node, all exponents matching a subset of the grammatical features specified for that node are eligible for spell out. This means that there is competition for insertion between VIs, which is regulated according to the Elsewhere Principle (EP):
(7) Elsewhere Principle (Bobaljik 2012:14)
If two (incompatible) rules R1, R2 may apply to a given structure, and the
context for application of R2 is contained in that of R1, then, R1 applies and R2
7
The Elsewhere Principle (EP) operates under the assumption that Vocabulary Insertion (VI) Rules can be either context-free or context-sensitive. Context-sensitive rules are obviously more specific than context-free rules because they apply exclusively in the contexts they are specified for, whereas context-free rules apply generally (‘elsewhere’). More formally, a rule is “more specific” if and only if it applies to a proper subset of the contexts to which a less specific rules applies.
2. Harley’s theory of compounding in DM
As should be clear from section 1.1, within a framework such as DM compounds are formed in the syntax. Harley (2009) provides a fleshed out analysis of compounding in DM, which I will go over in the following subsections.
Harley distinguishes four different kinds of compounds: synthetic compounds, modificational synthetic compounds, primary compounds and phrasal compounds. I will now go over her analysis of each one of these kinds.
2.1 Synthetic compounds
Synthetic compounds are structures in which there is an argumental relation between the head and the non-head (much like there would be in a verbal construction), such that the non-head is an argument of the head. Examples of such compounds in English are truck driver, book stacker, law abiding (citizen). As mentioned earlier, compounding under Harley’s (2009) view corresponds to incorporation of the non-head into the head of the compound; in the case of synthetic compounds, that means incorporation of an
8
argument into the selecting head. It is therefore necessary to establish what the selecting head is – in a compound like truck driver, for instance, there are two options: either truck is selected for by the verb to drive (i.e the result of the merger of √DRIVE with a
categorizing terminal node vº) or by the root √DRIVEitself. To answer this question,
Harley analyses evidence from one-replacement effects (Harley 2009:133-135), from which she concludes that arguments are merged with Roots before the Roots are categorized. I will follow Harley in this respect in my analysis. Considering that being
an argument of something is equivalent to being its first sister (Roeper and Siegel
1978), this means that in synthetic compounds the complement of the head is the first sister of the (uncategorized) Root. The complement is a noun, which means that it Merges with a categorizing nº node before incorporating into the Root. This incorporation, like all head-movement, must be feature-driven; in Harley’s analysis it is assumed that the feature driving head-movement of the non-head into the Root of the head is Case. The structure is as follows (adapted from Harley 2009:136).
(8)
√TRUCKMerges with nº and incorporation of the Root into the categorizing node is
assumed. The nP then Merges with the Root √DRIVE, and head-moves into it. The final step is Merger of this structure with a categorizing agent-flavoured nº (realized as –er),
9
and head-movement into it. Vocabulary Insertion realizes the final structure as truck driver.
From this, Harley concludes that incorporation in general occurs when the elements within √P incorporate first within themselves (i.e the non-head Root into its categorizing node, and then this structure into the uncategorized Root of the head) and then into the categorizing node of the head.
A point of interest in synthetic compounds in English is that they do not seem to allow incorporation of DPs. If the argument of √DRIVE was a DP like the truck or trucks, incorporation into the Root would not be possible - *the-truck driver, *trucks driver (Harley 2009:136) – and the argument would be stranded to the right of the Root: driver of the truck, driver of trucks. Harley’s solution to this issue draws on the featural requirements on head-movement in the current proposal: as mentioned earlier, it is assumed in Harley’s paper that incorporation of the argument into the Root is case-driven, i.e. there is a case feature on the argument that must be checked, and this achieved by head-movement of the argument into the Root. In the case of DPs, Harley argues that head-movement of the argument (nP) into Numº (e.g *trucks driver) or into Dº (e.g. *the-truck driver) would satisfy the case requirement of the argument DP-internally, meaning that the case feature would no longer be available to drive movement and the argument would be stranded to the right of the head.
Interestingly, the stranded examples require the addition of the preposition of; Harley attributes the insertion of of in the structure as a case-licenser in a Last-Resort operation (i.e. a phenomenon that must appear in contexts in which the derivation would otherwise fail, and furthermore can only appear in these contexts) in cases where
10
incorporation does not occur (Harley 2009:136, footnote 5). We will come back to this issue in section 3 as it will be relevant for the analysis of Portuguese compounds.
2.2 Modificational synthetic compounds
Modificational synthetic compounds consist of a deverbal adjectival head and an adjectival non-head, such as quick-acting (baking powder) or fast-falling (snow). In these types of compounds the non-head is not an argument of the head, unlike what happens with synthetic compounds; however, like synthetic compounds, modificational synthetic compounds "can only be formed from verb-modifier pairs where, in the corresponding verb phrase, the modifier would be the 'first sister' of the verb” (Harley 2009:137), as is shown by Roeper and Siegel (1978). This may seem conflicting with what was said earlier of synthetic compounds – the non-head is the first sister to the Root of the head because it is an argument of the Root, so claiming that the modifier is also the first sister of the Root seems counterintuitive. In the case of modifiers in verb phrases with an internal argument, the argument is Merged first with the Root that selects for it (i.e. and so is the first sister of the Root) and the modifier Merges secondly to an extended projection of the Root. If there is no internal argument, however, the modifier will be the first element Merged to the Root and so will be its first sister. Harley’s proposal is, then, that modifiers that are in the position of first sister to the Root may incorporate into it, since they share the same syntactic properties as an argument would – both are first sisters to the Root, and both are governed by it. The structure proposed for quick-acting is thus as follows:
11 (9)
[adapted from Harley 2009:138] Like we saw before for synthetic compounds, elements within the √P incorporate first (√QUICK into aº, then this structure into √ACT), and only then to other categorizing
heads; in this case, the structure [[[[√QUICK]√ a]a √ACT ]√ Merges first with a
categorizing vº because aº -ing affixes only to verbs and not to Roots (e.g. acting is possible, but *quicking is not, which shows that the category of the element that –ing affixes to is relevant, and, moreover, that that category must be v). The final step of the formation of the compound quick-acting is then to Merge the categorizing aº head realized as –ing on top of the structure, with subsequent head-movement into it. Structurally, then, there is no difference in this analysis between synthetic compounds and modificational synthetic compounds; the difference resides in the relation that the head and the non-head would establish in the corresponding verb phrase, i.e. verb-complement (synthetic compounds) and verb-modifier (modificational synthetic compounds).
Harley notes that there is an issue related to the analysis of these compounds. Due to the fact that -ing affixes only to verbs, there must be a vº head above the √P phrase, which means there must be incorporation into the vº head. This, however, seems to entail that
12
verb-incorporation would be possible in English, which would produce verbs like *to quick-act, which are impossible, however. Harley’s solution to this problem is to posit a constraint on English vº such that it cannot host complex heads containing more than one Root for Spell-Out in its base position (Harley 2009:141-142). This constraint on English vº accomplishes two goals: it correctly rules out verbs such as *to quick-act,
while also allowing for –ing to affix to structures such as [[[[√QUICK]√ a]a √ACT ]√ , as
the “complex head containing more than one Root” would head-move up from vº to incorporate into aº, which at Spell-Out would be the head containing the two Root Vocabulary Items prohibited for vº. It should be noted that this is the kind of prohibition that is parameterizable (i.e. not universal), since other languages allow for verb-incorporation (e.g Mohawk (Harley 2009:141)). This means that languages can differ with respect to which category-creating heads (i.e. nº, vº or aº) are capable of hosting incorporation.
A problem with this solution, however, is that there seem to be cases in English where verb-incorporation does in fact occur. Consider the following examples:
(10) a. to fat-shame b. to body-shame c. to photo-bomb d. to fact-check e. to proof-read f. to double-check
13
The examples above show that verb-incorporation is in fact attested in English, which contradicts Harley’s statement that no verb-incorporation is possible. The existence of these compound verbs is puzzling, because the data that led to Harley’s postulation of a ban on verb-incorporation (i.e. a parameter attaching to vº prohibiting the Spell-Out of complex heads containing more than one Root) does in fact seem to point towards an impossibility of verb-incorporation in English – it is not possible to form verbs such as *to quick-act and *to truck-drive. While it could be argued that modifiers cannot incorporate into vº, due to the fact that (at least) none of the examples above contains a modifier (even though this would be a difficult claim to make, given that there is no structural difference in this proposal between a modifier and an argument if both are first sisters to the Root), this would not explain why to fact-check is possible, for instance, but to truck-drive is not.
In Portuguese verb-incorporation is not possible, even though as we will see in section 3 there are cases of compounding in which there is movement through vº up into another head (much like in quick-acting), which means that the Portuguese data could be amenable to a solution like the one provided by Harley; however, for the English case such a solution fails to account for the existence of compound verbs such as to fact-check. I leave the status of verb-incorporation in English for future work.
2.3 Primary compounds
Unlike in synthetic compounds, in primary compounds (also known as “root compounds”) there is no argumental relationship between the Root of the head and the non-head. This has interpretative consequences – whereas in synthetic compounds the interpretation of the compound is always unambiguous, in primary compounds the
14
interpretation of the compound is ambiguous. That ambiguity is solved via Encyclopedic and pragmatic information. To illustrate this point, consider the synthetic compound truck driver on one hand, and the primary compounds nurse shoes and alligator shoes (Harley 2009:139) on the other. Truck driver can only ever mean ‘a person who drives trucks’, which is entirely predictable from the meaning of the Roots contained within the compound and the argumental relationship between them. This is not so with primary compounds – our pragmatic and/or Encyclopedic knowledge favours a different interpretation for nurse shoes than for alligator shoes: nurse shoes are shoes made for nurses, whereas alligator shoes are shoes made of alligator. While our knowledge favours these interpretations, it is still possible to conceive of nurse shoes as shoes made of nurses and of alligator shoes as shoes made for alligators; the relationship between the head and the non-head is strictly modificational, which means that there is nothing that forces an unambiguous interpretation for primary compounds.
Harley proposes that the non-head is in a direct sisterhood relationship with the Root of the head (as was the case with synthetic and modificational synthetic compounds) for two main reasons. First, it was already argued before for modificational synthetic compounds that modifier incorporation is possible if the modifier is Merged first with the Root (i.e. is the first sister); the second argument has to do with the interpretation of these kinds of compounds – Harley argues that it is possible to have the modifier be the first sister to the head noun’s Root and still get a non-compositional interpretation of the compound as long as there is no argument structure related to that Root (if the Root selects for arguments then there will be restrictions concerning what kind of modifiers it can take and the interpretation of the compound would therefore be unambiguous). A final meta-theoretical argument is that, since under Harley’s view the data can be
15
accounted for using the same structural properties as for other kinds of compounds, then you should use those properties so as to keep the theory as simple as possible.
The structure proposed is thus as follows:
(11)
√NURSE incorporates into the categorizing head nº to become a noun, which
subsequently incorporates into the Root √SHOE. Finally, the structure [[[[√NURSE]√
n]n √SHOE]√ incorporates into the categorizing head nº, realized as ∅ at Spell-Out.
This structure is exactly the same as the ones proposed for synthetic and modificational synthetic compounds – the non-head of the compound is introduced as first sister to the Root of the head, then the elements within √P incorporate within themselves, and finally the resulting structure incorporates into minimally one categorizing head. This is
problematic because, as we will see in section 4, the class of synthetic compounds2 and
the class of primary compounds have different properties, and as such are subject to different constraints; considering that in a theoretical framework such as Distributed Morphology words are built in the syntax (and compounds are word-sized units), it must be the case that these differences are due to a different syntactic structure; in other
2 I use “class of synthetic compounds” here to include both synthetic and modificational synthetic
16
words, if both types of compounds are formed in the syntax, but do not have the same properties, then the way in which they are formed must also not be the same.
2.4 Phrasal compounds
English allows for the formation of phrasal compounds, structures like bikini-girls-in-trouble genre and stuff-blowing-up effects. Considering what we have seen above about the ban on incorporation of DPs, the fact that phrasal compounds exist in the language is very surprising – “they include both DPs and vPs, to say nothing of CPs and PPs (though they cannot themselves be a DP; Lieber 1992:12))” (Harley 2009:142). If DPs cannot incorporate, you would expect that a structure containing DPs (and possibly other phrases other than nPs) would equally not be able to incorporate.
The first thing to note about phrasal compounds is that even though the whole modifying phrase incorporates to form the compound, yielding the order bikini-girls-in-trouble genre, and not genre (of) bikini-girls-in-bikini-girls-in-trouble, there is no incorporation within the phrase itself, otherwise you would get trouble-in-girls-bikini genre (Harley 2009:142), which aligns with what was previously said about the impossibility of incorporation of DPs. The second thing is that these phrases can appear in contexts other than compounds – it is possible to attach derivational morphemes like -ish, -y, or – ness directly onto the phrase (e.g. the general 'bikini-girls-in-trouble'-ness of it all) (Harley 2009: 143). This shows that, in fact, the phrase is behaving like a nominal in these contexts: it allows for derivational morphemes to attach directly to it like any other nominal would, and it can equally participate in primary compounding. Furthermore, the interpretation of these phrases seems to be the result of deriving an abstract conceptualization of the meaning that the internal syntax determines
17
compositionally. In other words, the syntax builds the phrase, and its meaning is compositionally determined by the way in which the phrase was built; the meaning of the phrase when it occurs in compounds or when it is attached derivational morphemes is then the result of abstracting a concept from the compositional meaning the phrasal syntax determined. Harley’s explanation for why these phrasal elements behave like nominals (i.e. like they have already undergone categorization) follows Sato (2007) – the phrasal elements undergo zero-derivation to a nominal category (Harley 2009:143); in DM, this means that the phrase merges with an nº to become an nP. As we saw in section 1.1, categorizing heads can have different “flavours” – some are agentive, like – er in a noun like driver, for instance, while others seem to be semantically neutral, in
the sense that they just nominalize the structure they attach to, like ∅ in cat-∅. The nº
that categorizes complex phrases like bikini-girls-in-trouble fulfils two functions in this proposal: it categorizes the phrase as a noun, and reifies the content of the phrase, which corresponds to the step of abstracting a concept from the compositional meaning determined by the syntax (so this new concept is not compositionally determined by the syntax, but rather ‘evoked’ by it). For the specifics of how denotation of the complex XP is able to compose with the reifying nº head see Harley (2009:143-144).
This analysis explains why it makes sense that these complex phrases should be able to participate in primary compounding – by merging with the reifying nº head they become semantically and syntactically an nP – but it does not explain how come they are able to incorporate. As we saw earlier, when these phrasal elements appear as modifiers (i.e. non-heads) in primary compounds the order of the modifier and the head
shows that they have incorporated into the Root ([bikini-girls-in-trouble]MOD
[genre]HEAD); this means that the complex XP must have incorporated into nº, after
18
what we have seen before for DPs, such incorporation should not be possible. I account for this issue in the following section.
3. Synthetic compounds in Portuguese
3.1 The data
In this section I will analyse the structure of synthetic compounds in Portuguese and see to what extent the Portuguese data fits Harley’s proposal for compounding.
Portuguese compounds are usually left-headed, unlike English compounds, which are right-headed. In synthetic compounds, the head is a deverbal noun, as can be seen below in (12) where a list of some Portuguese synthetic compounds is provided. This kind of compounds mainly produces nouns for instruments (e.g. quebra-nozes (‘nut cracker’)), and occasionally for agents (e.g. guarda-costas (‘body guard’)).
(12) a. guarda-jóias ‘jewel box’ (lit. ‘keep-jewels’)
b. guarda-vestidos ‘wardrobe’ (lit. ‘keep-dresses’)
c. papa-formigas ‘anteater’ (lit. ‘eat-ants’)
d. tira-nódoas ‘stain remover’ (lit. ‘take-stains’)
e. limpa-chaminés ‘chimney sweeper’ (lit. ‘clean-chimneys’)
f. quebra-nozes ‘nut cracker’ (lit. ‘crack-nuts’)
g. guarda-costas ‘body guard’ (lit. ‘guard-backs’)
h. saca-rolhas ‘corkscrew’ (lit. ‘pull-corks’)
i. lança-chamas ‘flamethrower’ (lit. ‘throw-flames’)
j. afia-lápis ‘pencil sharpener’ (lit. ‘sharpen-pencils’)
k. abre-latas ‘can opener’ (lit. ‘open-cans’)
19
m. faz-tudo ‘jack of all trades’ (lit. ‘do-everything’)
n. quebra-mar ‘sea wall’ (lit. ‘break-sea’)
o. sabe-tudo ‘know-it-all’ (lit. ‘know-everything’)
The first thing to notice about Portuguese synthetic compounds is that there is no overt categorizing nº head, unlike in English synthetic compounds, in which the categorizing head is usually realized by an overt morpheme ( –er in the case of nouns (e.g. truck driver) and –ing in the case of adjectives (e.g. car chasing)). As a consequence, the
head of the compound does not resemble a noun, but rather a verb; specifically, the 3rd
person singular of the Present tense, which in Portuguese is formed by the Root of the verb and the theme vowel associated with the verb class the Root belongs to (plus some phonological adjustments).
The second thing to notice about these compounds is that there can be variation in the non-head with respect to inflection – in examples a through k the non-head is plural, whereas in examples l through o the non-head is singular – and in particular, the non-head is very frequently plural (Villalva 1992:15); in other words, the non-head is, in the majority of the cases, a DP. This is in contrast with the English case, in which DPs are never present in synthetic compounds. In order to get at a structure of Portuguese synthetic compounds it is therefore necessary to answer the following questions:
i. Are Portuguese synthetic compounds formed by incorporation (as proposed in Harley 2009)?
ii. What is the exact nature of the head and the non-head (i.e. what categorizing and functional heads are contained in their structure)?
20
I will answer these questions in the following subsections.
3.2 Incorporation
In order to answer the first question it is necessary to have a way to differentiate between a structure in which the argument has incorporated into the Root of the head, and one in which the step of incorporation has not happened. Fortunately, as we saw before in section 2.1, synthetic compounds always have a phrasal counterpart – incorporated structures like synthetic compounds are right headed (e.g. truck driver) whereas structures in which incorporation did not occur are left headed (e.g. driver of trucks) because the argument is stranded to the right of the head. One important difference between structures with and without incorporation is the presence of the Pº head of in the latter, but not in the former, for reasons I will go over in what follows.
In Portuguese incorporation does not change the locus of the head, which means the structure remains left-headed; however, as we can see in (13) the presence of the similar Pº head de is also mandatory:
(13) abridor de latas ‘can opener’ (lit. ‘opener of cans’)
quebrador de nozes ‘nut cracker’ (lit. ‘cracker of nuts’) tirador de nódoas ‘stain remover’ (lit. ‘taker of stains’) guardador de vestidos ‘wardrobe’ (lit. ‘keeper of dresses’) quebrador de mar ‘sea wall’ (lit. ‘breaker of sea’)
From these examples we can see that Portuguese differs from English in three more significant ways, the most straightforward of which is that the head of the structures with no incorporation bears an affix that the head of the incorporated structures does
21
not, namely –(d)or3. We take it that this is the realization of a categorizing head nº that
attaches to verbs (i.e. elements Merged with vº) to form deverbal nouns. It is similar to English –er (e.g. condutor - driver) with the difference that whereas –er can affix directly onto Roots, -(d)or must affix onto verbs. In English, both the structures with and without incorporation bear the affix –er (e.g. truck driver – driver of trucks); in Portuguese only the structure without incorporation bears the corresponding affix – (d)or.
The second, less obvious difference between the two languages has to do with the non-head of the compound. In Portuguese, both structures remain the same with respect to the non-head: if the non-head is plural in the incorporated structure (e.g. abre-latas) then it is also plural in the non-incorporated structure (e.g. abridor de abre-latas); conversely, if the non-head is singular in the incorporated structure (e.g. quebra-mar), then it is also singular in the non-incorporated structure (e.g. quebrador de mar). In English, however, this is not the case - while the non-head of a synthetic compound cannot be Merged with any DP material (cf. section 2.1), the non-head of its phrasal counterpart must be Merged with DP material: *driver of truck is ungrammatical, but driver of trucks or driver of the truck, for instance, are well formed.
Finally, the third difference has to do with the syntactic status of the incorporated and non-incorporated structures in the two languages. In English, an incorporated structure such as truck driver is a compound, i.e. it behaves syntactically like an Xº. A non-incorporated structure, on the other hand, does not, and so it is just a regular noun phrase (NP). The same does not hold for Portuguese – the non-incorporated structure has a different syntactic behaviour than an apparently similar noun phrase would, and
3 The -d- is probably inserted in the phonological component whenever –or attaches to a vowel – compare
22
furthermore this behaviour is consistent with that of a compound. This means that the non-incorporated structure is an Xº, which I will now show. (14) compares the syntactic behaviour of a regular noun phrase containing the preposition de (livro de história (‘history book’)) with that of a non-incorporated structure like abridor de latas, the counterpart to abre-latas (‘can opener’).
(14) i. Modification of the head
a. o livro de história (‘the history book’)
a’. o [livro velho] de história (‘the old history book’) b. o abridor de latas (‘the opener of cans’)
b’. *o [abridor velho] de latas (‘the old opener of cans’) b’’. *o [abre velho]-latas (‘the old can opener’)
ii. Modification of the non-head
a. o livro de [história portuguesa] (‘the book of Portuguese history’)
b. *o abridor de [latas azuis] (‘the opener of blue cans’) b’. *o abre-[latas azuis] (‘the [blue can] opener’) iii. Impossibility of Dº material in the non-head
a. o livro da história (de Portugal) (‘the book of the history (of Portugal)’)
a. */?? o abridor das latas (‘the opener of the cans’)
a’. *o abre-[as latas] (‘the [the cans] opener’) iv. Ellipsis
23
a. Comprei um livro de física e um de história. ‘I bought a physics book and a history one.’
b. *Comprei um abridor de latas e um de frascos. ‘I bought an opener of cans and one of jars.’
b’. *Comprei um abre-latas e um de frascos. ‘I bought a can opener and one of jars.’
(14) i and ii show the differences with respect to modification between a noun phrase containing the preposition de (examples a’ and a, respectively) and a compound (b’’ and b’, respectively). Within a noun phrase, both head and non-head can be modified; within compounds, however, it is impossible to modify either – modification can only apply to the whole compound (e.g. o [abre-latas velho] (‘the [old can opener]’), which of course is a consequence of compounds behaving syntactically like Xºs. Examples i.b’ and ii.b show that the syntactic behaviour of the non-incorporated counterpart of the compound abre-latas (‘can opener’) with respect to modification patterns with that of the incorporated compound, and not like that of the noun phrase livro de história (‘history book’), as would be expected if this structure was not a compound. The same holds for iii and iv: whereas the noun phrase can host determiners on the non-head (iii.a) and participate in ellipsis (iv.a), the compound cannot (iii.b’ and iv.b’, respectively), with the non-incorporated structure again patterning with the compound. Moreover, (iii) shows that it is not possible to refer to the non-head of these structures, as this would require a determiner to exist in the structure. Based on this evidence, I conclude that the non-incorporated counterpart of Portuguese synthetic compounds is itself also a compound. It should be noted that not every structure with a deverbal noun with the affix –(d)or and the preposition de is a compound; take the sentence Onde está
24
o condutor do carro? (‘Where is the driver of the car?’), for instance. Here the modifier do carro (‘of the car’) is referential, as can be seen by the presence of the determiner o contracted with the preposition (de + o = do). These structures are only compounds when the non-head is the argument of the Root of the head, and the non-head has not incorporated.
This means that in Portuguese incorporation is not strictly necessary for the formation of synthetic compounds. The structures for incorporated and non-incorporation synthetic compounds are given in (15) and (16), respectively:
(15)
25
In (15) the Root √LATA Merges with a categorizing head nº and incorporates into it; next, Numº Merges onto the structure to make it plural, and the noun lata incorporates into it. The resulting DP Merges as first sister of the head √ABR, and subsequently incorporates into it. This structure then Merges with vº and incorporates into it, where
√ABR is categorized as a verb and gets its theme vowel4
. Finally, the categorizing head nº Merges with the vP, and the structure incorporates into it. (16) differs from (15) in the incorporation step of the DP into the Root of the head –√LATA incorporates into nº and then into Numº, at which point movement of the non-head stops; √ABR follows the same derivational process as in (15), incorporating into vº and then into nº. The resulting structure is not grammatical by itself (*abridor latas) and so the Pº head de must be inserted in the structure. As we saw at the end of section 2.1, this is a Last-Resort operation which inserts de as a case licenser to prevent the derivation from crashing, yielding abridor de latas.
It should be noted that verb-incorporation is not possible in Portuguese. Unlike English, which as we saw in section 2.2 allows for some verb-incorporation, in Portuguese no verbs can be formed via incorporation into vº (e.g. *abrir-latas). There is a vº in the structure of synthetic compounds, and in the formation of these compounds there is incorporation into it, but verbs cannot be formed in this way. This is exactly the kind of situation which is amenable to Harley’s explanation for English, i.e. that there is a prohibition on vº such that it cannot host complex heads containing more than one Root for Spell-Out in its base position – that means it is possible for the structure below vº to incorporate into it, so long as it head-moves out of it afterwards. This way the head containing more than one Root at Spell-Out will not be vº, but whichever one was
4
The theme vowel of verbs of the third class (to which abrir belongs) is -i-; however, Portuguese phonology prevents words from ending in [i], so the phonology changes it to [ɨ] (orthographically e). In abridor the theme vowel does not change because it is not word-final.
26
Merged last on the structure. This explanation correctly accounts for the Portuguese facts: it rules out verb-incorporation while allowing for incorporated elements to move up through vº.
One issue that still needs to be addressed is why the categorizing head nº that Merges last on the derivation is realized as ∅ in incorporated structures (e.g. abre-∅-latas) and as -(d)or in non-incorporated ones (e.g. abridor de abre-∅-latas). There are two possibilities – either there is a different nº Merging with each structure, or the nº is the same in both structures, and one is an allomorph of the other. If it were the case that each structure Merges with a different nº, then we would expect there to be differences between the two structures in meaning, and possibly in morphological properties such as gender. However, what we find is that there is no real difference in meaning between
∅ and -(d)or5
, nor in morphological properties – nouns categorized by -(d)or are always masculine, as are all incorporated synthetic compounds, even when the non-head is feminine, like in the following examples: o abre-latas, ‘the can opener’, o guarda-costas, ‘the bodyguard’, o papa-formigas, ‘the anteater’. It would seem quite coincidental that two different nºs would have such similar semantic and morphological behaviour.
An explanation relying on allomorphy of the nº seems more appropriate. This is supported by Beard (1995, 1996), who explains the difference between incorporated (nº realized as ∅) and non-incorporated (nº realized by -(d)or) synthetic compounds based on a generalization on the order of the head and the non-head within endocentric synthetic compounds. He claims that “[t]he order of the modifier and the head in an endocentric compound is almost always the same as the (default) order of an attributive
5 Perhaps -(d)or could be used more often to designate agents, but ∅ works as well (e.g. guarda-costas,
27
adjective and the noun that it modifies within a noun phrase.” (Snyder, in press: 8). In Portuguese (and in Romance in general) the default position of an attributive adjective in a noun phrase is to the right of the noun that it modifies, e.g. o vaso azul (‘the blue vase’, lit. the vase blue). The suffix -(d)or (and –eur in French, for instance) is, according to Beard, subject to two requirements: it must affix to a V (in the current framework: a structure categorised by vº), and, furthermore, it must affix to the rightmost edge of the word (Snyder, in press:13). In English, where the default order of an attributive adjective and the noun it modifies is N-V, the suffix –er can fulfill both requirements – it affixes to V and this occurs at the rightmost edge of the word. In Portuguese, however, these two requirements conflict with each other because the default order is V-N, and so affixing to V comes at the cost of not affixing at the rightmost edge of the word, hence the suffix is left unpronounced.
Beard’s explanation presents some problems within the current proposal – if it is the case that -(d)or affixes to the rightmost edge of the word, then a structure like abridor de latas contains more than one word, and so is not a compound, contra what was shown earlier. The problem can be solved, however, if we posit a restriction on – (dor) such that it cannot attach to complex heads containing more than one Root6; another option is Snyder’s suggestion that the absence of the suffix (in this case, French –eur) is due to a Last-Resort operation that deletes it when incorporation occurs (see Snyder, in press:13-15).
3.3 Conciliating the English and Portuguese data
28
Now that it has been established that there is in fact incorporation in Portuguese synthetic compounds, it is necessary to provide an account that accommodates both the Portuguese and the English facts:
(17) Both Portuguese and English synthetic compounds incorporate:
abre-latas – abridor de latas Portuguese
truck driver – driver of trucks English
(18) English only allows incorporation of nPs, never DPs:
truck driver but *trucks-driver
(19) Portuguese allows incorporation of both:
abre-latas (lit. ‘open-cans’) quebra-mar (lit. ‘break-sea’)
(20) The non-head of the non-incorporated counterpart of English synthetic
compounds cannot be an nP:
driver of trucks/ driver of the truck but *driver of truck
In the examples above we saw that there are instances of synthetic compounds in Portuguese with a plural non-head (such as abre-latas) and a singular non-head (such as quebra-mar). By itself, this does not mean that Portuguese allows for incorporation of nPs and DPs indiscriminately – if that were the case, then it would be possible to form similar compounds with nPs instead of DPs, and vice-versa. However, this is not what we find:
29
b. quebra-mar - ??quebra-mares (lit. ‘break-seas’)
The nP in (21a) gets a referential interpretation – instead of being interpreted as a ‘can opener’ (i.e. instrument used to open cans) it is interpreted as an opener for a specific can (i.e. instrument used to open a specific can; cannot open any can). The same observation can be made in (21b), even though in this instance it can be considered more or less acceptable. So, the plural is mandatory in most synthetic compounds in Portuguese (so that those that take a plural non-head must take the plural and not the singular), and pluralizing the non-head of synthetic compounds that take an nP as complement can be considered somewhat acceptable. It seems as though there is something specific about the plural in Portuguese that causes it to occur so frequently in synthetic compounds; this makes it relevant to look again at the examples in (12l-o) (repeated below for ease of exposition) to determine what it is about the non-head in these compounds that allows it to be singular:
(12) l. passatempo ‘pastime’ (lit. ‘pass-time’)
m. faz-tudo ‘jack of all trades’ (lit. ‘do-everything’)
n. quebra-mar ‘sea wall’ (lit. ‘break-sea’)
o. sabe-tudo ‘know-it-all’ (lit. ‘know-everything’)
In the first example, passatempo (‘pastime’), the non-head tempo (‘time’) cannot be pluralized7, and so it could only appear in a compound as an nP. The same holds for tudo (‘everything’) in examples (12.m and n), as it is invariable (does not inflect). (12.n) is a bit different, because it does have a plural: mar – mares (‘sea’ – ‘seas’). The same is true of the example above cata-vento (‘wind vane’): vento-ventos (‘wind’ – ‘winds’). Why, then, do these examples use the singular form and not the plural, like the majority
30
of synthetic compounds in Portuguese? And why does Portuguese allow incorporation of DPs, when English does not? The answer to this question lies, I believe, in the semantic restrictions of synthetic compounding. We will see below exactly how that works.
Let us examine Harley’s reasoning for the ban on incorporation in English again, step by step (the argument itself consists of 1 and 2; 3 is a consequence):
1. The feature driving head-movement of the non-head nP into the Root of the head is Case – the nP checks its case feature via incorporation into the head Root; 2. DPs do not incorporate in English because the case requirements of their nPs are
satisfied within the DP – the nP checks its case feature within the DP, which means there is no feature driving head-movement of the nP out of the DP and into the Root, and consequently incorporation does not occur;
3. When incorporation fails, a case-licenser is needed8 – in the case of English, this
corresponds to Last-Resort of, deriving structures like driver of trucks.
Let us first assume that 1 is true – the non-head nP incorporates into the Root to check its case feature. Next, let us assume that 2 is also true – nPs can check their case features DP-internally. This predicts two things: first, it (correctly) predicts that the nP does not head-move out of the DP, as there is no unchecked feature in the nP that matches anything outside of it; second, it predicts that the nP does not need any element outside the DP to check its case features, as they have already been checked within the DP. This means that as far as that nP is concerned, there is no reason for the derivation to fail, since its case features have been checked. The first prediction is consistent with the facts; the second and third, however, clash directly with 3 above, namely that failure of
31
incorporation results in insertion of Last-Resort of as a case-licenser. The second prediction has to do with the need for a “case-licenser” itself – if the case features of the nP have already been checked within the DP (which, as a consequence, prevents the DP from incorporating), then there is no need for an extra case-licenser outside the DP. A consequence of the second prediction has to do with the fact that this case-licenser is “Last-Resort of”. In section 2.1 we saw that Last-Resort operations must appear in contexts in which the derivation would otherwise fail, and furthermore can only appear in these contexts. In this particular instance, this means that if of had not been inserted then the derivation would have failed, but as we have just seen there is no reason for the derivation to fail. So assuming the truth of 1 and 2 has led to an inconsistency – there is no reason for Last-Resort of to be present in the structure (or any case-licenser outside of the DP, for that matter) and yet the structure would indeed be ungrammatical without it (e.g. *driver-trucks). There are two possible explanations for this: either 1 is true and 2 is false (i.e. incorporation is case-driven, but the DP does not assign case to its nP) or 1 is false, and therefore 2 is inconsequential (i.e. if case does not drive head-movement of the nP, then case assignment within the DP has no bearing on incorporation constraints). Let us begin by exploring the first possibility – 1 is true and 2 is false. By itself, the fact that incorporation is case-driven is not sufficient to exclude DPs from incorporating in English; however, it makes the correct predictions for the case of failure of incorporation – if the nP does not get case by incorporating into the head’s Root, then a Last-Resort case-licenser is necessary to prevent the derivation from crashing. So, this option brings about the right consequences for failure of incorporation, but fails to explain why incorporation of DPs should fail in the first place.
32
The second possibility is that 1 is false – incorporation is not driven by case requirements, but by some other feature. A challenge that this approach faces is that it becomes necessary to provide a new feature to drive movement, and the use of this hypothetical feature here must be consistent with syntactic theory in general. This would have to be a feature that the head’s Root is Probing for, and that the non-head (i.e. the Goal) is also valued for. Furthermore, failure of incorporation would imply that these features have already been checked DP-internally, or are somehow unavailable for checking.
Finally, Harley’s solution fails to explain why failure of incorporation involves insertion of Last-Resort of – if case is not involved in incorporation, then why would its failure require the insertion of a case-licenser? From this I conclude that case is indeed the feature that drives incorporation of the non-head into the head’s Root (which is to say that 1 is true). That being established, it is still necessary to account for the fact that
DP incorporation is not possible in English9, but widely attested in Portuguese synthetic
compounds. As we have seen above, saying that DPs cannot incorporate because the nP’s case features are checked DP-internally (2 above) is not a satisfying explanation because it leads to an inconsistency, and moreover it would predict that any nP that merges with DP material cannot incorporate, which the Portuguese data contradicts. I therefore conclude that the fact that incorporation of DPs is allowed in Portuguese but not in English indicates that there is some parameterizable property of the non-head that can lead to variation between languages (at least) with respect to synthetic compounding (or, more generally, to compounding involving incorporation; I will go into what kinds of compounds are formed in this manner and which ones are not in the next section).
33
So, what is this parameterizable property? The answer, I believe, lies in the interpretation of the ungrammatical compound that results from attempting to incorporate an element not allowed in the language. Let us take a closer look at such cases in English (22) and in Portuguese (23):
(22) a. *trucks-driver (23) a. *guarda-vestido (‘wardrobe’)
a’. *the-truck(s) – driver a’. *guarda- o(s)-vestido(s)
b. *drugs-pusher b. *cata-ventos (‘wind vane’)
b’. *the-drug(s) – pusher b’. *cata- o(s)-vento(s)
c. *cans-opener c. *abre-latas (‘can opener’)
c’.*the-can(s)-opener c’. *abre- a(s)-lata(s)
Given Harley’s claim that incorporation of the nP of the non-head into any DP material will cause incorporation to fail in English, I included both examples of ungrammatical compounds containing only Numº material (examples a-c), and Dº material (examples a’-c’). In the latter case, I added the option of including Numº as well, as a way to determine whether there is any interaction between Numº and Dº. This, however, does not obtain – both singular and plural forms are ungrammatical in examples a’-c’. This
means that the presence of Dº10 is enough to make incorporation impossible in the two
languages. The problem with examples a’-c’ is their interpretation.The interpretation of
a compound such as can opener (or abre-latas in Portuguese) is generic, in a similar sense than that of generic sentences. Generic sentences attribute a property to a class of
10
Please note that for the sake of space I used only examples with the definite determiner (the in English and o(s)/a(s) in Portuguese), but the result would be the same with an indefinite determiner (e.g. *a-can-opener, *abre- uma(s)-lata(s)).
34
individuals without attributing it to any individual of the class in particular11 (this is only a working definition; for a comprehensive definition of generic sentences, see for example Leslie and Lerner 2016). Using the example above, we can see that can opener is likewise generic because its interpretation corresponds to something like “X that opens cans” – the property of “opening cans” is attributed to the class of objects that opens cans, from which it does not follow that given any particular can, a can opener must be able to open it (or, in other words, the instrument “can opener” does not stop being a can opener just because there is a can that it cannot open). The same holds for synthetic compounds in general – their interpretation must be generic, and so there is a
further semantic restriction on the non-head12, i.e. that it be such that its interpretation
within the compound be generic. Returning to examples a’-c’, we can now see how their ungrammaticality can be put in terms of genericity – the presence of Dº in the structure (i.e. the determiner) makes a generic interpretation of the compound impossible. Turning now to examples (22a-c) and (23b), we can see that their ungrammaticality is due to the non-head being plural (which means their grammatical counterparts would contain singular forms in the non-head), and that conversely, the ungrammaticality of examples (23a,c) is due to the non-head is singular. It is now possible to interpret this ungrammaticality in terms of genericity: in the English examples, having a plural non-head is always ungrammatical because, in this context, the plural cannot yield a generic interpretation. That is not to say that in English it is impossible to get a generic interpretation from a plural; comparing (24a, b, c), it is clear that the context determines whether or not the plural gets a generic interpretation:
(24) a. The potato is a starchy vegetable.
11
It should be noted that a generic sentence such as “Birds fly” is not refuted by the observation that penguins are birds and yet they do not fly – the property of “flying” is attributed to the class of birds, but not to specific individuals within that class.
35 b. Potatoes are a great side-dish. c. can-opener
Both the singular and the plural form of potato can be used in generic sentences, but only the singular can be used in synthetic compounds (in English).
In Portuguese, on the other hand, the non-head is most frequently plural, and in the examples above (23a, c) were ungrammatical due to their non-head being singular. The reason for their ungrammaticality is the same as for English – in synthetic compounds, the plural is necessary to convey a generic interpretation of the compound. The ungrammaticality of example (23b), in which the grammatical compound’s non-head is singular, stems from the fact that in this particular case, the singular form of vento (‘wind’) is the one that allows for a generic interpretation by virtue of it being a mass noun; the same can be said of the example quebra-mar (‘sea-wall’) in (12n).
In view of these data, we propose that the “parameterizable property” of non-heads that was mentioned before, corresponds to a genericity restriction – the non-head of a compound must allow for a generic interpretation, which I formulate here as the Genericity Principle:
(25) Genericity Principle
The interpretation of synthetic compounds must be generic.
In English, this means that plural non-heads are disallowed; in Portuguese, it means that (generally) only plural non-heads are allowed, with occasional exceptions due to their lexical properties. In both languages, Dº-merged non-heads are disallowed, since this would immediately impede a generic reading. In this theoretical approach, the technical implementation of this semantic restriction would be that the derivation does
36
not crash in the syntax (as was the case in Harley’s proposal), but in LF (i.e. the semantic component).
3.4 Interim summary
In this section, I show how data from Portuguese synthetic compounds conflicts with Harley’s proposal for the ban on incorporation of DPs in English, and propose an alternative account based on genericity of synthetic compounds. The interpretation of synthetic compounds must be generic across languages – the specific way in which these languages implement this requirement is parameterizable, and in this particular case it was shown that for English, this means that synthetic compounds must take singular non-heads, while Portuguese generally takes plural non-heads.
In the following sections, I propose an account for two other kinds of compounds in Portuguese, namely primary (section 4) and prepositional (section 5) compounds.
4. Primary compounds in Portuguese
Section 2.3 provided an overview of Harley’s proposal of primary compounding in DM, which essentially consisted of the same structure as the one proposed for synthetic compounds, with the additional requirement that the head’s Root not have an associated argument structure. This means that the only difference between synthetic and primary compounds in this proposal has to do with the properties of the Roots themselves – whether or not they have argument structure (in the affirmative case, synthetic compounds are yielded, in the negative case, primary compounds) determines what kind
37
of compound is possible. As already mentionedin that section, having the syntax build
structures in exactly the same way and obtaining results with differing properties is problematic in a theoretical framework such as DM. In this section, I will expand on the differences between synthetic and primary compounds, which will in turn motivate an accordingly different structure.
4.1 Primary compounds vs synthetic compounds
Harðarson (2017:1-2) points out several properties in which primary and synthetic compounds differ (based partly on Marchand 1969, Roeper and Siegel 1978, Selkirk 1982, Lieber 2004 and Giegerich 2009), some of which have already been mentioned in previous sections.
The interpretation of synthetic compounds is consistent with the compositional
meaning of the structure built in the syntax (26); the same does not hold for primary compounds (28) (see section 2.3):
(26) truck driver =someone who drives trucks
(27) nurse shoes =shoes made for nurses/shoes made of nurses
Synthetic compounds always have a synonymous phrasal counterpart (28),
primary compounds only sometimes (29):
(28) a. truck driver = driver of trucks
b. book stacker = stacker of books
(29) a. nurse shoes =shoes for nurses
b. daughter languages ≠language of daughters
38
Synthetic compounds are restricted in what elements can be combined (primary
compounds are not):
(30) a. grim-acting (31) a. *grim-wanting
b. fast-mover b. *fast-finding
c. wage-earner c. *fool-looker
d. skin-grafting d. *doctor-grafting
[Roeper and Siegel 1978:207 (adapted from Harðarson 2017:2)]
In Harley’s proposal these differences are due exclusively to the properties of the Roots themselves, namely whether they have an associated argument structure or not. This, however, is not enough in a DM framework – such disparate properties must be due to different syntactic structures associated with each kind of compound, i.e. the difference between the two compounds must stem from the way in which they are formed in the syntax, not in differences in the properties of Roots. Harley’s proposed structure for synthetic compounds works because the argumental relationship between the head (more specifically, the Root of the head) and the non-head is present in the syntax via First Merge (i.e. the head’s Root and the non-head stand in a direct sisterhood relationship). The compositional interpretation of synthetic compounds (26-27) thus follows from the predicate-argument relationship established in the syntax, as do the selectional restrictions these compounds are subject to (30-31), while the incorporation step of the proposal explains the existence of synonymous phrasal counterparts when incorporation fails (28).
Conversely, the structure of primary compounds may not establish a predicate-argument relationship, as the evidence above shows that the formation of these compounds does not involve any kind of syntactic or semantic selection (29-31).
39
Moreover, if primary compounds were also formed through incorporation it would be expected that they would also (always) have phrasal counterparts; considering that this is not so, it must be the case that these compounds are not formed through incorporation. I will therefore follow Fenger and Harðarson (2018:6) in assuming that primary compounds are formed by merger of the non-head directly to the head of the compound, with each element being previously formed in separate workplaces (cf. Chomsky 1970, Borer 2003, Lieber 1992, Roeper et al. 2002, Baker 1988, Piggott and Travis 2013). This means that both the head and the non-head merge with their respective categorizing nodes prior to merging with each other:
(32) Workspace 1 Workspace 2
The non-head [√SHOE]√ n]n then merges with the head [√NURSE]√ n]n in the syntax,
producing the compound nurse shoes:
(33)
Harðarson (2017:3-8) shows that the non-head does not undergo spell-out prior to merger with the head, contra Piggott and Travis (2013). So, now we have two different structures for synthetic and primary compounds: the former relies on direct sisterhood between the head’s Root and the non-head to establish a predicate-argument relationship, as well as in subsequent incorporation of the non-head into the Root to
