The linearization of agreement morphemes

My primary source of evidence for the functional sequence is compositionality. But as agreement markers have no semantics, they don’t have scope either. Baker (1996, p. 30.) writes: "agreement morphemes, unlike tense and aspect, are semantically vacuous; thus, there is no way of locating them in a syntactic tree by investigating their scope with respect to other items". This in turn means that I will not be able to use my main source of evidence to locate agreement morphemes in f-seq. Instead, I will have to rely on the evidence presented by morpheme order.

In the approach that I adopted in Section 6.3.2, an important but frequently forgotten or ig-nored problem arises for the linearization of agreement morphemes. In particular, this approach entirely underdetermines where the agreement morphemes added to the functional head in question linearize. With one agreement feature added to a functional head, three possibilities arise. The functional head and the agreement morpheme may be spelled out by a single morpheme, a port-manteau. This is the easy case, as the linearization issue doesn’t arise. In this case the phonological form of the functional head is expected to vary according to the value of the agreement morpheme.

For instance, if an uNum agreement morpheme is added to a head, then the spellout of the head is expected to reflect whether uNum is valued as singular or plural by the goal. Examples where this possibly happens are furnished by North American Indian languages, in which suppletive verb forms cross-reference the number of intransitive subjects and transitive objects (data from Booker, 1982, p. 15.).

(35) singular dual plural

Mikasuki coko:l- wi:k- i:l ‘sit’

11The Nanosyntactic approach to case markers is explored in detail in Caha (2009). The interested reader is invited to consult this work.

(36) singular plural

Biloxi toh´o t^hi ‘lie’

Chitimacha nu:p- tuw- ‘die’

Diegue˜no pam n9k9mic ‘get there’

Hopi paki yungya ‘enter’

On the other hand, if the functional head and the agreement are spelled out by two different morphemes, then the issue of linearization arises. There are two logical possibilities in this scenario:

the agreement may be linearized either in front of the spellout of the functional head, or behind it. It is usually tacitly assumed that agreement comes after the spellout of the head, but nothing forces this in the theory. Verbal and nominal agreement markers that are likely to spell out in front of the head they are added to are provided in (37) and (38).¹²

(37) Zeeme you-pl

nawazi knife

ze-puunanai 2ps-buy

‘You all bought a knife.’ (Baker, 1985, p. 397., ex. 54. a.) Huicol

(38) `a-ˇc’k⁰’9n-c⁰a det-boy-pl

r9-y⁰n-k⁰`a

poss.3pl-house-pl

‘the houses of the boys’ (Wratil and Gallmann, 2011, p. 11., ex. 9. b.) Abkhaz

In Chapters 8 and 9 I will argue that Hungarian also features agreement morphemes that linearize in front of the functional head they are added to. In particular, I will analyze the plural marker on demonstratives and strong third person plural pronouns as an uNum agreement feature added to K. This agreement precedes K in the order of suffixes, so it linearizes in front of the hosting head.

The space of options is even bigger when not one but two agreement features are added to a head, and both agreement features have their own spellout separate from the spellout of the other agreement and the hosting head. In Bolivian Quechua, for instance, possessive agreement cross-references both the number and the person of the possessor, and the two agreement features have their own spellout. On the assumption that uPerson and uNumber reside on the same head, Bolivian Quechua instantiates the scenario described above.¹³

(39) wasi-kuna-n-ku-pi

house-plural-3.poss-plurality.of.possessor-in

‘in their houses’ (Myler, 2009, p. 52., ex. 21.) Bolivian Quechua

In this scenario, both agreement morphemes may spell out in front of the exponent of the head (in either order), or both may spell out after it (again in either order), or one may spell out in front it and the other behind it (yet again two options depending on which goes where).

The linearization of agreement morphemes is going to be a recurrent issue in the thesis. Let me foreshadow that I will not have a generalization or solution to offer on this point. At the appropriate places I will remark on the position of morphemes that I identify as agreement, but I will not go beyond descriptive adequacy on this particular point. Whether there are overarching generalizations about the linearization of agreement morphemes internally to particular languages or cross-linguistically is a question that must be settled on the basis of a wide and broad empirical basis. This thesis looks at the functional sequence of the Hungarian xNP. This domain of inquiry is too small to settle the matter.

There are indications that the linearization of agreement morphemes is possibly subject to broad generalizations or at least to general tendencies. For instance, the order of agreement morphemes

12Assuming that the Huicol verbal agreement feature is hosted by the verb, it spells out in front of the hosting category indeed. I cannot exclude the alternative that the agreement is hosted in some higher functional head. In that case this conclusion doesn’t follow. The same considerations apply to the Abkhaz agreement marker. I am not in a position to determine the exact position of the agreement in these examples. They are just included here as cases where agreement linearization in front of the host is an analytical option.

13It is possible, of course, that uPerson and uNumber reside on discrete heads. I am not in a position to determine whether this is so. This is not necessary for the present concerns, however, as the purpose of this example is merely to show the logical possibility.

on a modifier often replicate the order of suffixes on the modified head. The latter, in turn, reflect the functional sequence. Some examples are given below. (40) shows that adjectival agreement morphemes in Tundra Nenets (Samoyedic) line up in the same order as the suffixes on the noun.

(41) from Old Georgian shows that the Suffixaufnahme type of plural and case marking on the modifyee replicate the order of the same suffixes on the nounkey.

(40) (pidør^◦) you.sg

serako-m-t^◦ white-acc-2sg

te-m-t^◦

reindeer-acc-2sg

‘your white reindeer’ (Nikolaeva, 2003, ex. 9.) Tundra Nenets

(41) k.li¸te-n-i key-pl-nom

sasupevel-isa kindgom-gen

ca-ta-jsa-n-i

heaven-obl.pl-gen-pl-nom

‘(the) keys of the kingdom of (the) heavens’

(Plank, 1995, p. 14. ex. 9.) Old Georgian

These particular examples involve phrasal modifiers, and one may want to argue that in Tundra Nenets, for instance, there is a separate KP merged on top of adjectives, too, and the identical order results from identical movements in xNP and xAP. But with agreement on functional heads (as opposed to agreement on phrasal modifiers) this argument is inapplicable. The following Brazilian Portuguese examples, for instance, feature Number agreement on a functional head, D. Agreement on D follows D, just like the number morphology follows N.

(42) o

the-m.sg

cachorro dog-m.sg

‘the dog’ (King and Dalrymple, 2004, p. 91. ex. 52. a.) (43) os

the-m.pl

cachorros dog-m.pl

‘the dogs’ (King and Dalrymple, 2004, p. 94. ex. 59. a.) Brazilian Portuguese

How strongly the suffix order on phrasal modifiers and functional heads correlates with the order of suffixes on the modified head is an empirical issue that awaits future research.

As far as the relative order between different agreement morphemes is concerned, there are indications of tendencies or generalizations in this area, too. Trommer (2003) examines a sample of 58 languages that feature separate agreement markers for the subject’s person and number features. He finds that these languages exhibit 80 ordering patterns: 10 cases in which both are a prefix, 30 cases in which both are a suffix, and 40 mixed cases (one prefix, one suffix). He also finds that irrespective of the prefix/suffix issue, in 70 cases (87.5%) person agreement precedes number agreement, and in only 10 cases (12.5%) does number agreement precede person agreement. In such broad surveys, it is of course difficult to control for phonologically bound clitics, let alone the issue of whether the two agreement features reside in the same head or not. The preference for person preceding number is highly significant, but without further inquiry it is not possible to tell whether it means that person probes tend to reside in higher heads in f-seq, or person and number probes tend to be in the same head with a preference for person to linearize as first.

Svenonius (2007) touches upon the relative order of agreement with the Agent and the Patient.

Based on the sample in Haspelmath et al. (2005), he finds 172 languages which have separate, co-occurring agreement markers for these categories. Of these, 95 (56%) feature agent agreement before patient agreement, 57 (33%) do the other way around, and 19 (11%) have both orders.

Again, there is a clear preference for A before P, but phonologically bound clitics and languages in which these agreement features reside in separate heads are not controlled for.

To summarize, the approach that treats agreement morphemes as spellouts of agreement fea-tures added to interpretable heads runs into the problem of not predicting how the agreement linearizes with respect to the exponent of the head itself or other agreements on the same head.

As the thesis adopts this approach, it also inherits this problem.

However, it is important to point out that the problem has a generality which goes beyond this particular view, and the positioning of agreement morphemes is not straightforward in other approaches either. In the theory of Fuß (2005), terminals for agreement are adjuncts to other

terminals. Fuß argues that via the directionality of adjunction, his approach can capture both agreement prefixes and suffixes. While this is true, his theory also has the problem of describing rather than predicting the position of agreement with respect to the hosting head. DM’s treat-ment of agreetreat-ment in terms of dissociated morphemes runs into exactly the same problem: adding morphemes post-syntactically makes no predictions about whether those morphemes will end up as prefixes or suffixes. This is an important point. If dissociated morphemes could capture the linearization of agreement morphemes in an insightful manner, it would constitute a strong argu-ment in favour of a morphological component of grammar. This, in turn, would seriously challenge the adopted spellout algorithm and all of Part I. But dissociated morphemes do not explain the prefix or suffixhood of agreement morphemes in a principled way, and so agreement doesn’t require rethinking Chapters 2 through 5.

It may appear that the only approach that can actually predict the order of agreement mor-phemes with respect to each other and functional heads is the one which operates with dedicated AgrPs. This is a misinterpretation, however: this approach also stipulates the order, only it stip-ulates it via the functional sequence. Consider why. Large chunks of the functional sequence can be motivated by compositional semantics. For instance, if the goal is to build a nominal f-seq, then the sequence had better start with a nominal core. If one wants to have multiple units, it is motivated to make units first, and apply counting subsequently (Num > Cl). It is also motivated to merge Num below K, as Num is an‘internal’ business of xNP, while K serves to embed xNP into the larger sentence context. But as already pointed out above, agreement markers have no scope.

This means that the position of AgrPs in f-seq can be motivated only on the basis of word order, precisely the thing that the position of AgrPs is intended to capture. Differently put, the argument is circular: the position of agreement markers is captured by the position of AgrPs in f-seq, and the position of AgrPs in f-seq is motivated (and supported) only by the position of agreement markers.

This is the very definition of stipulation. This means that re-introducing AgrPs into f-seq would bring no empirical payoffs. This is significant in the context of this work. I would have to loosen up the syntax-semantics mapping I assume if AgrPs were the only sensible way to capture the position of agreement morphemes. This is not the case, however, and I will continue to assume the tight mapping between the two components in Parts II and III as well.

Having noted the problem with the linearization of agreement morphemes and the fact that this thesis will not have any insights to offer on this matter, I will turn to the mechanism of Agree used in Part II.