The functional sequence meets the agreement problem
6.1 The problem of deciding whether a morpheme is agree- agree-ment or not
Agreement is a ubiquitous phenomenon of natural language that has intrigued scholars for a long time. In generative grammar many approaches to agreement have been conceived, and, as it is usually the case, these approaches often embrace opposing conclusions and solutions. However, before we turn to the theoretical and empirical challenges surrounding agreement as well as the variety of analyses offered to deal with them, it is worth pointing out that a question that often arises is whether a certain piece of morphology is the actual exponent of a functional category or merely agreement with it, the actual exponent being phonologically zero.
In certain cases, conventional wisdom has it that agreement of some sort is clearly involved in the structures. A relevant example is DP-internal agreement between the noun and its modifiers for number, gender and case, as in the Modern Greek example in (1).1
(1) a. i
the-fem.sg.nom
amerikanid-a
American-fem.sg.nom
ginek-a
woman-fem.sg.nom
‘the Americal woman’ (genitive)
1The DP-internal agreement displayed by (1) is known as ‘concord’. Given the widely held assumption that
‘regular’ agreement and concord are derived by the same mechanism (Carstens, 2001, 2008; Koopman, 2006), concord will not receive a separate discussion here. See, however, de Chene (2004) and Giusti (2008) for claims that Agreement and Concord are different notions.
153
b. ton
the-fem.pl.gen
amerikanid-on
American-fem.pl.gen
ginek-on
woman-fem.pl.gen
‘the American women’ (genitive) (Alexiadou, 2001, ex. 17.) Modern Greek
A different type of example is provided by the elaborate suffixaufnahme system of Kayardild (Tangkic, Australian), detailed in Evans (1995) and Round (2009). Kayardild has five types of cases: i) adnominal case relates NPs to NPs, ii) relational case relates core arguments to the verb (nominative) or peripheral arguments to the clause (location, destination), iii) modal case indicates TMA, iv) associating case links NPs with nominalized verbs, and v) complementizing case applies to clauses, indicating that they are an argument of the matrix or that marked coreference relationships hold between matrix and subordinate (Evans, 1995, ch. 3.4). These cases appear in the order indicated in (2) (associating and complementizing case fill the same slot in (2) and don’t co-occur).
(2) stem + adnominal + relational + modal + associating/complementizing
Crucially, all five types of cases appear on every word in their domain. A simple PP is shown in (3), where the instrumental appears both on the actual instrument and its owner.
(3) dangka-karra-nguni man-gen-instr
mijil-nguni net-instr
‘with the man’s net’ (Evans, 1995, ex. 3.47) Kayardild
In (4) the above PP is embedded in a clause with the TMA indicator modal ablative case (mabl) na. This morpheme stacks on every word in its domain, including those in the PP.
(4) maku woman
yalawu-jarra catch-pst
yakuri-na fish-mabl
dangka-karra-nguni-na man-gen-instr-mabl
mijil-nguni-na net-instr-mabl
‘The woman caught some fish with the man’s net.’ (Evans, 1995, ex. 3.48)
(5) minimally differs from (4) in also containing the complementizing oblique case (cobl) ntha.
This yields altogether four cases ondangka ‘man’.
(5) maku-ntha woman-cobl
yalawu-jarra-ntha catch-past-cobl
yakuri-naa-ntha fish-mabl-cobl
dangka-karra-nguni-naa-ntha man-gen-instr-mabl-cobl mijil-nguni-naa-nth
net-instr-mabl-cobl
‘The woman must have caught fish with the man’s net.’ (Evans, 1995, ex. 3.50)
That the Modern Greek and Kayardild examples above contain agreement morphemes is sup-ported by both morphological and interpretational considerations. As for the morphological consid-erations, only nouns come from the lexicon with gender features baked into them, therefore in the Greek example the gender feature of the adjective and the article must be agreement morphemes that merely reflect the value of the gender on the noun. As far as interpretation is concerned, it is clear that a feature like singular or a case like Instrumental is interpreted only once, on the nominal head and on the Instrument respectively, therefore their other occurrences must be non-interpreted agreement morphemes. All this is conventional wisdom.
However, in a variety of cases it is subject to debate whether a certain piece of morphology represents a contentful functional category or merely agreement. This is because the number of occurrences of a morpheme/feature does not provide a sensible diagnostic for agreementhood.
Depending on the phenomenon and the analysis, a morpheme appearing only once in the containing maximal XP may still be agreement, and multiple exponents of the same feature in the containing maximal XP do not necessarily mean that agreement is involved either.
The plural occurring with quantified nouns is an example of a morpheme that is often taken to represent agreement even if it appears only once in the containing maximal projection. As is well known, whether a noun modified by a numeral or quantifier bears plural morphology or not is subject to cross-linguistic variation: English and Norwegian require overt plural marking in these
cases, while Hungarian and Kurdish reject it.2
(6) a. seven book-s English
b. sju seven
eple-r apple-pl
‘seven apples’ Norwegian
(7) a. h´et seven
k¨onyv-(*ek) book-pl
‘seven books’ Hungarian
b. mamoste, teacher,
du two
mamoste teacher
‘a teacher, two teachers’ (Ortmann, 2000, ex. 6.) Kurdish
Some works take the plural co-occurring with numerals and quantifiers to be the actual exponent of a functional category such as Num or Div (Borer, 2005 is a prominent representative of this group).
Others view it as merely agreement (c.f. Farkas and de Swart, 2010, Ionin and Matushansky, 2004, 2006, the latter two explicitly argue that agreement is with the entire extended NP and is semantic in nature). Yet others recognize the existence of both ‘real’ and agreement plurals (Borer and Ouwayda, 2010). Thus the fact that the plural occurs only once in (6) does not guarantee that it is not an agreement morpheme.3
Examples of the converse case, where a certain morpheme or feature has multiple appearances but agreement is not necessarily taken to be involved, is presented by Romance, Bantu and Semitic clitic doubling phenomena and verbal subject or object cross-referencing affixes (‘verbal agreement’) in non-pro-drop languages.4 Incorporated or phonologically bound pronouns, clitics, and agreement markers are infamously difficult to tease apart on empirical grounds, especially because of the well-known grammaticalization cline in (8) (but see Zwicky and Pullum, 1983 for some pointers).
(8) pronoun > clitic > agreement marker
As an example of a clitic doubling construction, consider the Spanish examples below. Pronominal direct objects are clitic doubled in all varieties of Spanish (9), while non-pronominal direct objects are optionally clitic doubled in the R´ıo de la Plata, Argentina dialect (and some others), as shown in (10).5,6
(9) La her
llamaron 3pl-call
a A
ella her
‘They calledher.’ (Su˜ner, 1988, ex. 4. a.) Spanish
(10) Juan Juan
la
it.fem.cl sac´o got-3sg
la the
nota grade
sin without
esfuerzo.
effort
‘Juan got the grade without effort.’(Franco, 2000, ex. 38.) Southern Cone Spanish
A definitive analysis of clitics, and more importantly for us, of clitic doubling constructions has proven elusive. In the following summary of the most influential approaches to clitic doubling and agreement markers, I include in brackets the language which provided the empirical basis of the investigation, because, as suggested by some sources, it is likely that different languages require
2There may also be intra-language variation in this respect, c.f. Englishevery apple andall apples. I will ignore this issue here.
3The nature of Hungarian plural marking will be extensively discussed in Chapter 9.
4These are often argued to be the same phenomenon, c.f. for instance Manzini and Savoia (2002, 2004a,b, 2007, 2009). This chapter follows this line of thinking and makes no important theoretical distinction between clitics and agreement affixes. Where they are referred to under different names, it is only to follow descriptive traditions.
A further relevant phenomenon is clitic left dislocation, which involves clitic doubling with a full DP on the left periphery of the clause but not in focus. For reasons of space, the literature on CLLD is not reviewed here.
5Theaof (9), not glossed by Su˜ner (1988), has various functions. In the example at hand it marks Accusative.
Thanks to Antonio F´abregas for discussion and clarification regarding these examples.
6Depending on the language and the dialect, clitics may not co-occur with full NP doubles (French objects), clitics may double subjects (Italian varieties), direct and indirect objects (Italian, Spanish, Bantu) and there may be a divide on the basis of whether the double is pronominal or not (Standard Spanish objects).
different analyses.
Doubling clitics have been analyzed as agreement for instance by Rizzi (1986) (Trentino Italian), Poletto (1995) (Veneto Italian), Sportiche (1996) (French dative clitics), Su˜ner (1988) and Franco (2000) (Spanish object clitics). On the other hand, Borer (1984) (Hebrew genitives, Spanish, French) and Jaeggli (1986) (Spanish) have analyzed clitics as non-agreemental base-generated elements on the head V or N, with the full DP double occupying the argument position. According to a different proposal, doubling clitics are D elements. They either project their own DP that is flanked by VP shells (Torrego, 1995, Spanish, the full DP is merged in spec, VP) or this D is in the extended projection of the object argument (Uriagereka, 1995 for Spanish, the full DP doubles occupy spec, DP). Yet a different suggestion is that clitics are independent heads that project their own CliticP in the clausal spine above the verb phrase (Sportiche, 1996; Cocchi, 2000; Manzini and Savoia, 2002, 2004a,b, 2007, 2009; Ouali, 2011) or inside the vP (Papangeli, 2000). Sportiche (1996) (French accusative clitics) and Ouali (2011) (Tamazight Berber object clitics) suggest that the full DP doubles are merged in the vP-intenal argument position; while Manzini and Savoia (2002, 2004a,b, 2007, 2009) (Italian subject and object clitics) and Cocchi (2000) (Spanish and Bantu object clitics) argue that the doubles are merged as focus or topic. Papangeli (2000) (Modern Greek object clitics) proposes that the clitic phrase is merged as the complement of the verb and takes the DP double as its complement.
Turning to morphemes that are generally characterized as subject or object ‘agreement’ in descriptive terms, we find the same debate and the same types of proposals. In polysynthetic languages, for instance, both Baker (1996) and Jelinek (1984) take the full DP arguments to be adjuncts. But while Baker analyzes Mohawk argument cross-referencing markers as spellouts of the head’s case feature (essentially, as agreement, with an analysis that is equivalent to Borer’s 1984 treatment of clitics) and suggests that there are null pronouns in the argument positions, Jelinek (1984) argues that the cross-referencing markers in Warlpiri (as well as in Spanish and pro-drop languages in general) are clitic pronominal arguments of the verb. More recently Adger et al. (2010) proposed that the agreement markers of Kiowa are the spellout of uninterpretable φ features on the argument introducing verbal heads. They are prosodic clitics valued by the full DP arguments introduced in the regular argument positions (that is, they are real agreement morphemes).
As we can see from this cursory overview, doubling of the same set of features within the clause does not entail the conclusion that agreement is involved, and there is more than one way to accommodate the full DP doubles into the syntax.
That the number of occurrences of a feature (or feature bundle) has no bearing on its status as agreement or contentful functional head/phrase is clearly shown by Zeijlstra’s (2004) analysis of n-words and Manzini and Savoia’s (2007) analysis of Italian subject markers. Zeijlstra (2004) argues that negative concord arises when the presence of a (c)overt negative operator is signaled by the uNeg feature of n-words. In strict negative concord languages the interpreted operator is covert. If there are multiple n-words in a sentence in such a language, then a feature occurs and receives overt spellout multiple times but none of the pronounced instances correspond to the actually interpreted feature (i.e. all of them are agreement).
Manzini and Savoia (2007) examine the subject markers of the Castellazzo Bormida example in (11), among others. In this northern Italian dialect clitic doubling co-occurs with verbal inflection:
(11) contains a full lexical subject DP, an expletive subject clitic (ClS) and a verbal inflection cross-referencing the subject (infl).
(11) iR the
maý"næi children i ClS
"dRwOm-u sleep-infl
d9 in
"l6 here
‘The children sleep in there.’
(Manzini and Savoia, 2007, ch. 2., ex. 22. a.) Castellazzo Bormida Manzini and Savoia argue that neither subject-related morpheme is agreement. D is projected in three different bits of the functional sequence of the clause, VP, IP and CP. The three subject-related morphemes spell out these Ds. They share the same reference because their referential properties are compatible with each other, and this allows an interpretation in which they are in a chain construal. In other words, they are assigned the same reference and the same argumental slot (but crucially the chain does not involve movement).
To sum up this section, before we think about the syntactic representation of agreement as such, we must carefully examine each and every morpheme suspected to be agreement, and decide whether it is agreement indeed or rather a phonologically weak exponent of a contentful functional category or phrase.