CrISP

In this presentation Dr. Neeleman will develop a theory in which person features are more abstract than usually assumed: they do not refer to speaker or addressee, but are rather used to navigate a 'person space'. The theory is confronted with two typological problems. (i) Why is the inventory of persons so limited? Why aren't there 30 persons? (In this context 30 is not a random number, but represents the number of potential persons.) (ii) What explains the typological observation that syncretism between first and third person is much rarer than syncretism between either first and second, or second and third person (Baerman et al. 2005, Baerman and Brown 2011)? If time allows, he will discuss also Dutch as a case study. In this language there are two person endings that arrange themselves in such a way that there is a 2-3 syncretism in the regular case, a 1-2 syncretism under subject-verb inversion, and an optional 1-3 syncretism with a particular lexical class of verbs (modals).

This talk addresses the question: What is the best way to model phonologically conditioned morphology? I will present the empirical facts, to the extent they are known, regarding three different phenomena where a phonological property affects or potentially affects a morphological process -- namely, infix placement, phonologically conditioned suppletive allomorphy, and phonologically conditioned affix order. I will then summarize the arguments from Paster 2005a,b, 2006a,b, 2009, to appear, regarding why the original Optimality Theory (OT) account of phonological effects in morphology (McCarthy and Prince 1993a,b) is unsatisfactory. I will discuss the extent to which more recent constraint-based analyses and other alternative models are better equipped to model phonologically conditioned morphology, ultimately arguing in favor of an account in which phonology and morphology are distinct components of the grammar, incorporating aspects of Lexical Phonology (Kiparsky 1982a,b, Mohanan 1986), morphological subcategorization (Lieber 1980, Kiparsky 1982, Orgun 1996, Yu 2003), templatic morpheme ordering (Bloomfield 1962, Zwicky 1985, Anderson 1986, Simpson and Withgott 1986, Speas 1990, Stump 1992, Inkelas 1993, Hyman and Inkelas 1999, Good 2003), and the Scope principle (Rice 2000) and Mirror Principle (Baker 1985).

Present-day English has two methods for forming comparative adjectives: a synthetic (prouder) and an analytic (more intelligent) route. I examine the variation within single adjectives (prouder vs. more proud, intelligent vs. *intelligenter) and across syntactic positions (pre-nominally, post-nominally, post-verbally) and argue that prosodic structure is relevant both at the word and at the phrase level in modulating the distributional patterns of the comparative. In particular, I argue that high frequency lexical items (fast) have less complete prosodification than infrequent items (vast); this prosodification is visible to morphological operations like suffixation. As a consequence of extrametricality, -er attaches more readily to high frequency stems (faster vs. more vast) because there is less segmental material between the main stress and the cohering suffix -er. In terms of variation by syntactic position, I argue that adjuncts before a head are preferentially exponed synthetically for prosodic reasons, largely to satisfy a principle that lexical stress be realized as phrasal stress as much as possible. This promotes structures with heavier material later in the utterance, in which periphrastic forms are increasingly preferred (i.e., in post-nominal or post-verbal positions). Crucially, however, I contend that word-level variation patterns, such as those resulting from sensitivity to token frequency, remain surface-apparent in the midst of prosodic constraints that apply over larger syntactic domains. With short adjectives like fast, there is a higher rate of use of -er in post-nominal or post-verbal positions than a longer adjective like happy, for example. The gradient nature of this effect suggests that the distributional patterns of word-like units independently influence the preference for synthetic or analytic forms. Blocking of periphrastic forms by synthetic forms is recast in gradient terms as a process that is conditioned by multiple factors operating across morphological or syntactic levels.

English -ing varies between two phonologically distinct allomorphs, [iŋ] and [in]. Across different varieties of English this variation has been shown to depend on gender, speaking style, and socio-economic factors (Fischer, 1958; Labov, 1972; Trudgill, 1972). Phonological context has also been shown to be relevant (Houston, 1985): the allomorph [in] is more likely when a coronal segment follows. Strictly localist theories of morphology (e.g., Bobaljik, 2000; Embick, 2010) predict that the phonological context should only be able to affect allomorph selection under syntactic locality conditions. Globalist theories (e.g., theories of allomorph choice formulated within standard optimality theory) predict that in principle any information in a linguistic representation could affect allomorph choice. This paper reports on experimental data involving -ing-allomorphy that seems incompatible with both types of theories.

As illustrated in (1) and (2), we crossed the syntactic environment (local vs. non-local) with the phonological environment (a-[ə] vs. the-[ð]), using a syntactic contrast familiar from studies of prosodic phrasing (e.g., Itzak et al. 2010):

(1)
Local:
a. Whenever the boy was browsing a book the game would fall off the table.
b. Whenever the boy was browsing the book the game would fall off the table.

(2)
Non-Local:
a. Whenever the boy was browsing a book would fall off the table.
b. Whenever the boy was browsing the book would fall off the table.

Localist theories predict that the phonological context should be able to affect the choice of allomorph when the word providing the phonological environment is syntactically local as in (1), but not when it is part of the next sentence (2). Globalist theories predict that phonological context should be relevant in both types of cases.

The results show an effect of phonology both in (1) and (2). This is unexpected under the localist account. However, the effect is much smaller in (2), which is unexpected under the globalist account.

The interaction between phonology and syntax suggests that syntactic locality might be relevant after all. However, within the syntactic conditions, there is a quantitative correlation between the strength of the prosodic boundary separating the verb and its complement and the liklihood of a phonological effect of the following word. In other words, whether the phonological form of the following word has an influence on allomorph choice depends gradiently on the strength of the prosodic boundary separating the two words even within the same syntactic condition. Once these quantitative measures of boundary strength are taken into account, the effect of between syntactic conditions vanishes: the difference between (1) and (2) in the size of the phonological effect is completely explicable as a result of the difference in boundary strength between the two structures.

The pattern of phonological conditioning can be accounted for by a model of allomorph selection that is constrained by the locality of production planning. The segmental content of an upcoming word can have an effect on allomorph choice if its phonological form is already available at the time of vocabulary insertion. The strength of a prosodic boundary negatively correlates with the availability of the following word, and can thus serve as a proxy measure for the locality of production planning.

The data suggests that the phonological effect on allomorph choice, at least in this case, can be stated in purely segmental terms. The apparent effect of syntax on the phonologically conditioning of allomorph choice can be explained by its indirect effect on the likelihood that the phonological material of the upcoming word is already planned out at time when allomorph selection happens. This suggests a more modular view of the syntax/morph-phonology interaction across word boundaries than current approaches that assume an interleaving of phonology and syntax.

The account in terms of the locality of production planning provides a potential explanation why individuals in our experiment and the dialects described in the literature only seem to vary in the proportion with which they choose the allomorphs (from almost always [in] to almost always [iŋ]), but none seem to show a complementary distribution according to phonological or syntactic context: the reason is that the conditioning environment is only probabilistically available depending on how much planning is been possible, and this varies depending on the structure of sentence and other factors. In other words, there might be a reason why ing-allomorph selection is consistently a variable process: reliably planning out an entire utterance in all its phonological detail is difficult if not impossible. Other cases of phonologically conditioned allomorphy are considered and their amenability to an account in terms of the locality of production planning is discussed.

In Richards (2010) I posited a universal condition on the prosody of wh-questions, which was intended to predict whether a given language would move its wh-phrases or leave them in situ. The condition requires a wh-phrase to be in the same prosodic domain as the interrogative complementizer which Agrees with it. Whether a language has to move its wh-phrases then depends on how its prosody is organized. Some languages can leave wh-phrases in situ and manipulate the prosody of the sentence to satisfy the prosodic requirement; others cannot do this, and must move the wh-phrase to make it sufficiently prosodically close to C.

In this talk I will generalize the prosodic requirement I posited for the relation between C and wh-phrases, applying it to all pairs of syntactic objects that are related either by Agree or by selection. Data handled by the resulting theory include a variety of facts about the placement of adverbs in languages like English and French (traditionally accounted for via claims about the structural height of verbs), the Final-over-Final Constraint of Biberauer et al (2010), and the requirement that clauses with English quotative inversion cannot have auxiliaries.

This talk explores the perennially troubling and well-known problem of Russian prefixal bracketing paradoxes as a way of shedding light on, and distinguishing between, two approaches to the interface between morphosyntax and phonology at the word and sub-word levels. In these paradoxes, the prefix and stem behave semantically and morpho-syntactically as a unit to the exclusion of suffixal inflection (1a), but the phonology --- in particular, the (non)realization of jer vowels, which must be posited in underlying representations but are not always realized in surface forms --- treats the inflected stem as distinct from the prefix (1b).

1a. Morphosyntax/semantics:
[[vy bros]-il]
[[out throw]-3sg.m]
[[prefix stem] inflection]

1b. Phonology:
[vy [bros-il]]
[out [throw-3sg.m]
[prefix [stem inflection]]

One approach to this problem, couched in the framework of Distributed Morphology and embodied most recently by the work of Embick (2010), proposes that words are built syntactically, with functional, category-defining phase heads playing the crucial role of ÒfixingÓ both morphophonological and semantic information before further word formation takes place. Such approaches typically involve a direct interface to rule-governed phonology, with each morphosyntactic phrase defining a cyclic phonological domain. I demonstrate that, because it syncs up morphosyntactic and phonological domains, this theory is interestingly too restrictive to effectively capture the pattern represented in (1). In particular, a reasonable attempt to capture the facts in this framework leads to the conclusion that certain morphological items must be characterized as post-cyclic in only one of the two modules (as in Matushansky 2002). Further, such an account fails to capture:

- the generalization that jer realization is conditioned by sonority sequencing.
- the generalization that the phonological bracketing is part of a broader pattern of prosodic alignment in the language.

Building from this result, I develop and explore an account based on a different conception of the morphosyntax-phonology interface, in which word composition is carried on partially in the syntax and partially in a separate morphological module (Ackema and Neeleman, 2004). This model involves an indirect interface -- mediated via the building of prosodic structures -- to a constraint-based phonology. I argue that these theoretical tools are required if the two generalizations missed by the DM-based theory (above) are to be captured.

References
Ackema, Peter, and Ad Neeleman. 2004. Beyond Morphology: Interface Conditions on Word Formation. Oxford: Oxford University Press.
Embick, David. 2010. Localism versus Globalism in Morphology and Phonology. Cambridge: MIT Press.
Matushansky, Ora. 2002. On formal identity of Russian prefixes and prepositions. MIT Working Papers in Linguistics 42:217--253.