Mayfest 2015: Morphest

9:15-9:30 Welcome (Jeffrey Green)
9:30-10:15 Alec Marantz (via Skype) - “Taking morphology seriously: MEG studies of morphological representations”
10:15-11:00 Ariel Goldberg - “Experimental investigations of the morphology-phonology interface”
Break
11:15-11:45 Ellen Lau - Commentary
2:00-2:45 Maria Gouskova - “How to tell phonologically aggressive affixes from exceptional ones”
2:45-3:30 Robert Henderson - “Mayan morphology at the syntax-prosody interface”
Break
3:45-4:30 Vera Gribanova - “On the role of head movement in word formation”
4:30-5:10 Bill Idsardi - Commentary

9:30-10:15 Charles Yang - “The Third Factor in Morphology”
10:15-11:00 Karen Miller - “Crosslinguistic studies on children's acquisition of morphology”
Break
11:15-11:45 Jeff Lidz - Commentary
1:15-2:00 Ruth Kramer
“The Morphosyntax of Gender and Number: Morpho or Syntax?”
2:00-2:45 Mark Aronoff - “Competition and the body: Two non-computational factors in the organization of linguistic morphological systems”
Break
3:00-3:45 Jonathan Bobaljik - “Cyclic Spell Out in Words”
3:45-4:25 Alexander Will
4:40-5:40 (MMH Maryland Room) - 
Moderated by Omer Preminger

For decades, linguists have framed the study of language in terms of a language faculty, a specialized cognitive ‘organ’ unique to humans. In the last decade, even the most stalwart proponents of this view have come to acknowledge the existence of other factors in the
organization of human languages. In this talk, I will concentrate on two of these factors,
embodiment and competition, drawing examples from the morphology of spoken and
signed languages. Neither is unique to language, nor especially human or cognitive in
nature. Their role in the structuring of languages points to a research paradigm in the study
of language in which no single factor is privileged and the importance of any one of them
is gauged by the insights it provides, not by its uniqueness to language.

This talk looks at the mapping of syntactic structure to morphological from the perspective of structural locality constraints on morphological operations (cf. Embick 2010, Bobaljik 2012, Moskal in press). One topic to be discussed is whether the word (maximal X-zero) is a privileged locality domain. I discuss evidence that the word is both too small and too large. On the one hand, the word is too small in that triggers of purely morphological operations, such as suppletion, may be located outside the complex X-zero containing the node that undergoes the operation (Bobaljik and Harley 2014 on participant-number-governed suppletion; cf. Thornton in prog for an alternative). On the other hand, the word is too large - there is evidence for phase-like cyclic domains within words, restricting which heads may interact with one another (Bobaljik 2012 on adjectival suppletion). Somewhat programmatically, I will present a conception of how the interaction of head movement and Spell-Out may yield the result that cyclic domains are defined in roughly the same way for both syntactic and morphological concerns, but leaves open sufficient flexibility to understand why syntactic phases and morphological domains do not always neatly line up.

Traditional linguistic descriptions of morpho-phonological phenomena comprise a phonological grammar that acts over symbolic representations of phonological and morphological information. This approach has had enormous success in accounting for many patterns observed in the world’s languages though it is not without its limitations. Specifically, it does not address the fact that morphologically complex words are produced and comprehended by modality-specific cognitive processes. These processes may influence not only a word’s morphological and phonological structure but even grammatical computations themselves. It also assumes that all apparent patterns are the explicit result of learned grammatical processes. In this talk, I argue for a holistic approach that fundamentally acknowledges the joint psycholinguistic and grammatical nature of morpho-phonological phenomena. Within this perspective, psycholinguistic data can be used to inform linguistic theories of morphology and phonology. At the same time, it raises the possibility that certain phenomena may have their origins in processing principles rather than in learned grammatical computations.

In this talk, I present data from psycholinguistic and cognitive neuropsychological investigations of spoken production. I argue that these data constrain grammatical theories, for example placing limits on the degree to which morphological and phonological information may interact. Moreover, these data suggest that certain cases of morphologically-conditioned underapplication may be due to performance factors relating to how morphemes are combined into single representations rather than learned grammatical computations. Quantitative data on two English allophonic patterns are presented that are consistent with these claims. Finally, I discuss how this approach allows us to integrate information from domains that are even farther afield (e.g., written production) to inform linguistic theories.

In theories of lexical accent, accentual dominance of affixes is sometimes tied to their syntactic status as heads vs. adjuncts: heads can be dominant, whereas adjuncts cannot (Revithiadou 1999 and others). This talk argues that the facts are more complex than this: dominance has heterogeneous sources and does not align neatly with morphosyntactic head status. Morphosyntactic head status does, however, align with the ability to rob other morphemes of their special phonological behavior. The talk discusses three types of dominance. The first is analytically the simplest: an affix imposes a predictable stress pattern onto the word, for example by pulling stress onto itself even when the stems are lexically specified as stressed. Some Russian diminutive affixes fall into this category, and they are demonstrably adjuncts based on morphosyntactic diagnostics. Either heads or adjuncts can follow this pattern--it is just a manifestation of exceptionality within the stress system. More complex types of dominance include deaccenting and conditional dominance. In both of these types, an affix has the ability to take away another morpheme's exceptionality. In deaccenting dominance, found in Japanese, an affix erases the stress of its stem without introducing its own accent. In conditional dominance, found in Slovenian, an affix takes away another affix's dominance. A unified analysis of these patterns is developed in Lexical Maximum Entropy theory: morphemes can be specified for special scaling factors (exceptionality) or regularization factors (aggression); only heads can be regularizing. This proposal is extended beyond lexical accent to other cases of morphologically conditioned loss of exceptional patterning, including nonlocal blocking of alternations and morphologically conditioned nativization effects in loanword phonology. 

Despite the usefulness of head movement over the past 30 years in deepening our
understanding of a range of empirical phenomena — verb second effects, verb-initial
clause structures, and lexical decomposition, among others — the modular status of head
movement has long been at the center of controversy, since at least Chomsky 2000. In part 
this is because aspects of the Minimalist program — a commitment to building structures
at the root, for example — have made it difficult to integrate head movement theoretically.
Further, the evidence weighing in favor of any particular position is subtle and can be
difficult to verify. In this talk I focus on a set of arguments that has direct consequences for this
discussion. These arguments draw on configurations involving verb stranding ellipsis (vse),
in which there is verb movement out of a constituent that is elided. A general schema is in
(1), and a representative Russian example of verb- stranding TP ellipsis (Gribanova, In
progress) is in (2). 

An important question surrounding such constructions involves the identity condition on ellipsis and its interaction with head movement: must the parts of the verb that originate inside the ellipsis site match their respective parts in the antecedent verb? Investigations of Hebrew (McCloskey, 2011, 2012) and Irish (2005, b,a) demonstrate that there is in fact a strict matching requirement on the stranded and antecedent verbs, even under discourse conditions like contrastive verbal focus, which might favor mismatch. Some discussions of this required matching effect, known as the Verbal Identity Condition (vic), have suggested that it forces a conclusion about the modular status of head movement: if the movement is post-syntactic in nature, the relevant heads will be in the ellipsis site at narrow syntax, making them relevant for any isomorphism requirement imposed by ellipsis licensing
(Schoorlemmer and Temmerman, 2012).

This talk demonstrates that Schoorlemmer and Temmerman’s (2012) conclusion is neither
necessary nor sufficient once a broader set of facts is considered. Drawing on data from
Russian vse, I show that not all languages behave identically with respect to the vic:
Russian vse constructions require matching as a default, but permit mismatches between
the antecedent and stranded verb’s verb stems in discourse situations involving lexical contrast on the relevant verbs. I argue that the Russian pattern is completely expected under fairly standard assumptions: if the head movement is syntactic and may leave a trace, then prominent theories of ellipsis licensing stemming from Heim 1997 will permit mismatch of the extracted verbal parts under contrastive focus, just as this is permitted in cases of phrasal movement out of ellipsis sites (e.g. I know which boy the teacher praised, but not which girl the teacher praised). Finally, I use this contrast — between Russian on the one hand and Irish and Hebrew on the other — as the starting point for a deeper, open-ended investigation of the potential sources of a matching requirement like the vic, as well as variation among languages in the degree to which it is obeyed.

All theories of the syntax-phonology interface must answer the question of what information is accessible to each module across the boundary? Serial theories of the interface must go one step further and ask when is this information available in the course of a derivation. Mayan languages provide an ideal empirical testing ground for such theories because they present a wide range of syntactic and morphosyntactic phenomena whose adequate description seems to make reference to prosody, e.g., intonational phrase final clitics, phonological phrase and intonational phrase final allomorphy, etc. Against this backdrop, the present talk uses data from two Mayan language—K’iche’ and Tz’utujiil—to defend the following claim about the syntax-prosody interface: Individual morphemes can make direct reference to high level prosodic constituents, like intonational phrases. The effects of this are manifested in two ways: (i) Idiosyntactic morphological facts can alter the default way clause-level syntactic structure is mapped to prosodic structure, and (ii) high-level prosodic structure can affect Vocabulary Insertion. The result is a PF-interface with richer interactions than is commonly assumed. In particular, the construction of highlevel prosodic constituents needs access to low-level, idiosyncratic morphological facts, and the phonological content of morphemes cannot be inserted until almost all prosodic structure is built.

The first part of the talk revisits data in Henderson 2012, which concern so-called phrasefinal morphology in K’iche’. In particular, I will focus on a set of suffixes that indicate verb class membership known as status suffixes [5]. Only when the verb is clause-final do status suffixes appear; compare (6-a) with the status suffix -ik and (6-b) without. Crucially, there is no difference in interpretation. When there are multiple clauses, there are multiple phrase final morphemes (2).

(1) a. Iwir xinkos-ik
yesterday I.tired-SS
I got tired yesterday.
b. Xinkos iwir.
I.tired yesterday
I got tired yesterday.
(2) a. We kin-kos-ik kin-war-ik.
if INFL-tire-SS INFL-sleep-SS
If I get tired, I sleep.
b. Aree taq kin-kos-ik kin-war-ik.
when INFL-tired-SS INFL-sleep-SS
When I get tired, I sleep

While phrase final morphemes track clause boundaries, their distribution is best understood as ιP-final, where ιPs are mapped nonrecursively onto CPs. The strongest argument comes from syntax-prosody mismatches which allow status suffixes to sometimes appear non-adjacent to a clause boundary, e.g., when a preposition taking a CP complement intervenes (3).
(3) a. Xinkos-ik rumal xinchakun-ik. b. (V-ik rumal (V-ik)CP)CP I.tired-SS because I.worked-SS c. (V-ik)ιP (rumal V-ik)ιP I am tired because I worked.

The most straightforward analysis of these data would take status suffixes to be bundles of
features that are only associated with phonological content under Vocabulary Insertion when adjacent to an ιP-boundary. But this would obviously only be possible if Vocabulary Insertion were to take place after intonational phrases are built, which is our first argument that morphology and high level prosodic structure interact. The second argument is based on previously unanalyzed data from Tz’ujutiil, which like K’iche’, has phrase final status suffixes. And just like in K’iche’, they appear utterance finally and at CP edges.

(4) Dayley 1981: 113,120
a. X-war-i.
INFL-sleep-SS
He slept.
b. X-war iiwiir.
INFL-sleep yesterday
He slept yesterday.
(5) Dayley 1981: 359
a. Ja taq nuu-miix xee-kam-i qas ee taq utz
The PL my-cat INFL-die-SS truly INFLPL good
My kitties that died were very pretty

Unlike K’iche’, though, Tz’utujiil status suffixes must appear before certain nominals, in particular, ones that bear the definite article

(6) Dayley 1981:113
a. X-war-i jar aachi.
INFL-sleep-SS the man
The man slept.
b. X-war jun aachi.
INFL-sleep a man
A man slept.

I argue that this cannot be due to general syntactic or semantic properties of the definite article, and must be due to lexical idiosyncracies. In particular, the definite article simply requires an ιP-boundary to its left. In this way, baroque facts about particular lexical items can trump the default syntax-prosody mapping at the highest level of prosodic structure building, creating the environment for the prosodically sensitive vocabulary insertion of phrase-final status suffixes. It is through the interaction of these two phenomena that we see an elaborate case of morphological and prosodic structure-building co-constraining one another.

The gender system of a language is usually determined by inspecting the agreement patterns of its singular nouns. For example, in Amharic, one set of singular nouns appears with the definite determiner –u, while the complement of that set appears with the definite determiner –wa. Therefore, Amharic has two genders: one with –u and one with - wa. However, plural nouns do not always make the same cut, i.e., they do not always show the same gender distinctions as singular nouns. For example, in Amharic, all plural nouns appear with –u, regardless of which definite determiner they appear with in the singular. This talk aims to investigate two particular types of gender-number relationships from the perspective of Distributed Morphology: convergent gender-number systems and crossed gender-number systems. Both types turn out to have implications for morphological theory, especially on how to draw the line between a morphological phenomenon and a syntactic phenomenon.

A language is convergent when it makes fewer gender distinctions in the plural than in the singular, like Amharic definite determiners. Gender-number convergence often holds across multiple paradigms of a language (Bobaljik 2002, Harley 2008), and I discuss how to distinguish between a morphological analysis of this effect (i.e., as a metasyncretism) and a syntactic analysis (i.e., as a difference in how plural feature bundles are structured). The underspecificational approach to syncretisms in DM also leads to predictions about possible types of convergent systems, and I show that these predictions are confirmed through case studies of three convergent languages: Coptic, Maay and Tamil. In a crossed language, some nouns (appear to) change gender from the singular to the plural, e.g., masculine in the singular and feminine in the plural. I argue that crossed systems are heterogeneous; some are due to morphological idiosyncrasy (e.g., Hebrew), some are due to syntactic properties (e.g., Somali) and some are neither morphological nor syntactic but due to a separate phenomenon entirely (e.g., Romanian). The talk concludes with brief discussion of the accuracy of, and potential explanations for, Greenberg’s (1966) Universal 37 that no language makes more gender distinctions in the plural than they do in the singular.

Although contemporary linguistics investigates the knowledge of language underlying language comprehension and production, cognitive neuroscientists often approach language with the assumption that speaker-listeners may process linguistic stimuli without crucially accessing the representational knowledge attributed to them by linguists. For example, although linguists demonstrate that words necessarily decompose into constituent morphemes, cognitive scientists often assume as a default that words are recognized first as wholes, and perhaps decomposed afterwards, if a task demands this. Two MEG experiments will be presented as case studies arguing that linguistic representations are necessarily and crucially exploited in language use. Both experiments demonstrate that speakers decompose apparently simple words into roots and affixes as a necessary step in word recognition. An experiment on Arabic demonstrate that the recognition of speak sounds within words is modulated by predictions for upcoming sounds based on the frequency of the Arabic roots, not the frequency of the words being recognized. An experiment on visual word recognition in English shows that syntactic category assignment (e.g., is “hammer” a noun or a verb?) involves the combination of word root with an unpronounced category affix, rather than the choice of a whole noun or verb word from a mental lexicon. An important conclusion from these and related experiments is that button press reaction times in behavioral experiments can be a misleading guide to the stimulus variables crucial to early stages of language processing. As the button press is the end of the line for a complex series of mental computations, RTs reflect most strongly the last stages of the computation, while the MEG signal reveals properties of earlier stages.

This talk focuses on children’s acquisition of morphology that occurs variably in the input. Data from two empirical domains will be presented: (1) plural morphology and number agreement in four dialects of Spanish (Chilean, Mexican, Andalusian, and Dominican) and (2) the root infinitive stage in working-class and middle-class English. The Spanish plural marker is /s/ and all elements in the DP agree in number (e.g. La-s niña-s inteligente-s). Studies on the acquisition of number morphology have concluded that Spanish-speaking children (from Madrid, Spain) have mastered plural marking on the noun before 24 months of age and shortly after extend the plural marker to other constituents in the DP (Marrero & Aguirre 2003). However, most other varieties of Spanish have a phonological process of syllable final /s/ lenition that impacts both morphological and non-morphological /s/ (Lipski 1994). In such varieties, the plural
marker is sometimes produced as [s] and/or [h] and sometimes it is omitted, and the level of omissions varies within speakers (e.g. style-shifting), across speakers (e.g. sex, SES) and across varieties of Spanish (e.g. Dominican, Chilean). 

1. Mexican Spanish (Mexico City): No /s/ lenition, plural marker produced as [s]
2. Chilean Spanish: /s/ Lenition, plural marker produced as [s], [h], v. omission
3. Dominican Spanish: /s/ Lenition, plural marker produced as [s] v. omission
4. Andalusian Spanish: /s/ Lenition, plural marker omitted, instead laxed vowel

In this talk I will present child production and comprehension data on children’s development of plural morphology in these varieties of Spanish and situate the findings within Yang’s (2002) Variational Learning Model. I will also present data from a second study on acquisition of subject-verb agreement marking. Non-agreeing don’t (e.g. She don’t have her books) has been associated to the Root Infinitive Stage in English child language (Schutze 2010, Guasti & Rizzi 2004), as absence of third singular –s in both forms appears to overlap in timing. This previous research focused primarily on the Sarah and Nina spoken corpora from the CHILDES database – yet did not take into consideration that Sarah (a working-class child) was exposed to nonagreeing don’t in the input, while Nina was not. In this talk I will reanalyze this longitudinal 
data and argue that while there are timing differences in the acquisition of agreement morphology for Sarah and Nina, the stages through which the two children pass are the same, regardless of input-type.
 

It is generally recognized that exceptions are part and parcel of morphology. Theorists, who have access to a wide range of data and empirical methods, can identify exceptions as such, which are then isolated from the more systematic aspect of word formation. (Think "marked", "diacritics", "in the syntax" vs. "in the lexicon".) But how does the child learner find exceptions from a relatively small corpus of positive data? In other words, how does a morpheme find ITS way, on its own, to the appropriate place in the grammar where the theorist says it should go?

In this work, we develop a calculus which the child can deploy to find the exceptions (as well as rules) in morphology. The primary motivation comes from computational complexity considerations, that the quantity of lexical exceptions imposes cost to the online and real time process of word formation. As a result, exceptions can be predicted on purely numerical measures of vocabulary composition. We provide several empirical case studies of how productive morphology can tolerate a certain amount of exceptions. Critically, we tackle the problem of gaps in word formation: gaps arise, and can be predicted, when none of the alternations defined over a morphological domain reaches the requisite quantity to tolerate the others as exceptions.