UD Conference on Stress and Accent
Program Registrations Venue Travel & Lodging Accepted Research Abstracts Organization
The aim of this workshop is to bring together researchers and scholars interested in the nature of stress and accent in the world’s languages. It is a follow-up conference to the workshops on Stress and Accent in 2010 and 2011 at the University of Connecticut, and is supported by grant no. 1123692 from the National Science Foundation.
The program is here. There will be a conference dinner on Friday night at Homegrown, a local restaurant.Back to the top
In order to plan more effectively, and to help cover costs of the event, we do ask that participants register for the conference. The registration fee will be $15 for students and $30 for faculty.
Please register by clicking here before November 26, 2012.Back to the top
The conference is being hosted by the Department of Linguistics and Cognitive Science at the University of Delaware. We are located at 125 E. Main Street.Back to the top
Here are instructions describing how to get to the University of Delaware.
Here are all the lodging options near the University of Delaware.
Here are the lodging options that are close enough to walk to the venue or offer shuttle service to the University of Delaware (please check with hotel staff about how their shuttle service works).
Courtyard Newark at the
University of Delaware
400 David Hollowell Drive
Newark DE 19716 http://www.udel.edu/hotel
South 654 S. College Ave.
Newark DE 19713
Homewood Suites by Hilton - Newark/Wilmington South
640 S. College Ave.
Newark DE 19713
The following hotels don't offer shuttle service but are within walking distance to the following University of Delaware shuttle routes: Express, Evening 1, and Campus Loop, South Campus.
Howard Johnson Inn & Suites and Conference Center
1119 S. College Ave.
Newark DE 19713
630 S. College Ave.
Newark DE 19713
This hotel is close enough to walk.Super 8 Motel
Back to the top
Covert representations and the acquisition of lexical accent
B. Elan Dresher
Lexical accent poses in a particularly sharp form two basic problems (Dresher 1999) that learning models must overcome: the Credit/Blame Problem—which parameters or constraints are responsible for any given success or failure?—and Meno’s Problem—how can we find the solution if we don’t know what we’re looking for?
Dresher & Kaye (1990) attempted to address these problems in the context of a learning model for metrical phonology. This model assumes that every parameter is associated with a cue (that is, a trigger), something in the data that signals to the learner how that parameter is to be set. Further, parameter setting proceeds in a (partial) order: this ordering specifies a learning path (Lightfoot 1989) which allows later parameters to depend on the results of earlier ones. As a consequence, the categories of ‘overt’ and ‘covert’ structures are fluid, and not fixed throughout acquisition. For example, which syllables are heavy and which are light in a quantity-sensitive (QS) system may be covert at the outset, in the sense that the learner does not know the proper distinctions. At a certain point they become overt, and aid in the acquisition of further covert structure, which in turn becomes overt, etc.
In tackling the problem of how lexical accent is acquired, we first have to decide how it is represented in the grammar. I adopt the SBG theory of Idsardi (1992), Halle & Idsardi (1995), Halle (1999). Dresher (1994) modifies the Dresher & Kaye model to apply to SBG representations, but stops short of trying to incorporate lexical accent. In this talk I show how the learning path of Dresher (1994) can be extended to account for lexical accent. I argue that any learning algorithm would have to have to instantiate, in some form, the basic ingredients of this model.[Top | Abstracts Top]
Conflicted metrical prominence: Evolutionary paths and synchronic analysis
Most languages of the world display stress systems that can be captured in standard metrical theories through a cohesive set of rules or constraints encoding factors such as proximity to a word edge, rhythmic alternation of strong and weak syllables, and syllable weight. There are, however, some languages in which diachronic developments have conspired to create multiple types of conflicting prominence that are not readily amenable to formal analysis given standard assumptions about metrical representations. For example, Vaysman (2009) discusses Uralic languages in which the reconstructed trochaic stress pattern of the proto-language has been replaced by a quantity-sensitive system but segmental alternations appear to be sensitive to the original rhythmic system. Similarly, the Northern Iroquoian languages Seneca and Cayuga (Chafe 1977, Michelson 1988) have maintained to a large extent the penultimate lengthening pattern that is a vestige of the original stress system but have also instituted a rhythmic stress pattern originating at the left edge of a word. This paper will present results of a cross-linguistic study of these hybrid metrical systems with the goals of exploring their sources of diachronic development and investigating how they might be most effectively treated in metrical theories not designed to model orthogonal sources of metrical prominence within the same language. [Top | Abstracts Top]
Harry van der Hulst
In many so-called stress languages, the rhythmic profile of words results from two separate procedures: accent and rhythm. The accentual module selects a specific syllable which occupies the position of primary stress and which functions as an important reference point for rhythm. In van der Hulst (2012) it is argued that the burden of irregularity is carried by the accentual module which belongs to the lexical phonology. Subsequently a rhythmic module provides the complete rhythmic ‘wordscape’. I will argue that rhythm is typically (and perhaps always) post-grammatical (not just post-lexical, but also post-syntactic) and as such fully regular.
In addition to accent and alternating rhythm, I will adopt a third ‘player’ in the rhythmic structure, namely a ‘polar beat’ that provides prominence to the edge opposite to the edge of the lexical accent. I first provide a brief overview of the accentual module (based on van der Hulst 2012), after which this talk focuses on the rhythmic module which is presented in terms of a grid-only approach. I provide a typology of rhythmic systems, based on various discussions in the literature and the available evidence from the StressTyp database. A distinction is made between simple rhythms and complex rhythms, the latter mostly involving so-called bidirectional systems or dual systems. The proposal is made here that bidirectionality is a consequence of an Edge Prominence rule which places a polar beat on the edge opposite to the accent that underlies the primary stress, creating a ‘hammock pattern’. Subsequently, rhythm operates in the valley between these two prominence peaks and can echo (i.e. ripple away from) either one or the other. I also discuss a subclass of the complex rhythms occurring in so-called clash systems, proposing that these systems too can be seen as having two opposite prominence peaks with rhythm bouncing into the lesser, polar beat. For the specifics of rhythm assignment I compare three alternative theories, concluding that the simplest theory, one that has no iambic or trochaic bias but instead operates with ‘free beat addition’, is sufficient and thus preferred. Overall, I propose the following set of rhythm parameters.
I here included the polar beat under the rhythm parameters, although the point will be argued that this kind of ‘edge prominence’ is an independent submodule in the post-lexical phonology, preceding alternating rhythm. Parameter (b) indicates whether rhythm is echoing the lexical accent or, if present, the polar beat. Parameter (c) decides whether rhythm is weight-sensitive and parameter (d) decides whether rhythm is binary or ternary. Parameter (e) decides whether the final syllable is provided with a rhythmic beat or not. I will show that these parameters explain the variety of attested rhythmic patterns, including the symmetries and asymmetries that have been attested in the literature. [Top | Abstracts Top]
Learning Phonology with Hidden Structure
Computational models of learning with violable constraints have led to significant progress in understanding how learners acquire the complex system of knowledge that is phonology. However, a number of significant challenges remain. This talk addresses one of the major outstanding problems, learning in the face of hidden structure, such as the structural ambiguity created by metrical structure. I examine Robust Interpretive Parsing (Tesar and Smolensky 1998), a well-known approach to structural ambiguity in OT, and show that its extension to probabilistic constraint-based grammars (Boersma 2003) is problematic. I propose two modifications to the learning algorithm and show that both lead to improvements in the performance of Stochastic OT and noisy HG learners. The experiments also highlight computational differences between OT and HG with implications for learnability, typology, and the modeling of variation. [Top | Abstracts Top]
Is there a default in lexical stress systems?
Stress is not always calculated by means of a phonological rule which operates on the basis of syllable count or weight sensitivity. Lexical stress systems are reported in the literature as systems with highly unpredictable stress (see Bat-El 1989, 1993; Idsardi 1992; van der Hulst 1999; Halle & Idsardi 1995, a.o.). In such systems the computation of stress heavily relies on the lexically pre-specified information morphemes may be endowed with. In accentual conflicts, the actual position of stress is determined by a grammar-specific principle (e.g., edgemostness, headedness, etc., cf. Halle, 1973, 1997; Kiparsky & Halle, 1977; Melvold, 1990; Alderete 1999, 2001a, 2001b; Revithiadou 1999, 2007). The examples in (1) and (2) from Russian and Greek, respectively, are instructive. In (1b) & (2b), the inherently accented inflectional ending (/ á/ and /-ón/, respectively) surfaces with primary stress. However, in (1d) and (2d) the same ending loses prominence to the inherent stress of the root.
Theoretical analyses of lexical accent systems designate the non-lexically inflicted stress as the phonological default (examples (1a) and (2a) for Russian and Greek, respectively). However, the issue of which stress pattern represents the default is not an easy one to answer. Several nonce-probe experiments on Russian revealed stem final stress (Nikolaeva, 1971; Crosswhite, Alderete, Beasley & Markman, 2003) or penultimate (PU) stress (Andreev, 2004; Fainleib, 2008; Lavitskaya & Kabak, 2011a, 2011b) to be the speakers’ statistically preferred choice. Similarly in Greek, the default antepenultimate (APU) stress was found to be strongly dispreferred in reading tasks (Protopapas et al., 2006) and to be highly marginal in bare/suffixless words, e.g. acronyms (Revithiadou et al., 2011; Topintzi & Kainada, 2011). It is evident, therefore, that, as far as the phonological aspect of lexical stress systems is concerned, we still walk on murky ground. In this paper, we address in detail the split between theoretical and experimental research in determining the nature of the default, using Greek as a case study.
Under the Stress Deafness Hypothesis on acquisition (Peperkamp & Dupoux, 2002; Dupoux & Peperkamp, 2002, et seq.), speakers of lexical stress systems are expected to have richer representations of stress engraved in their Mental Lexicon, because they are naturally exposed to more stress contrasts during acquisition compared to speakers of predictable stress. Based on this hypothesis, we expect such speakers to predominantly rely on this stress encoding mechanism when having to assign stress on unknown/less frequent lexical items, and hence to produce lexically-inflicted stress patterns, and not the language-specific default.
Interestingly, this expectation is borne out in Greek. We report on the results of two perception experiments (Revithiadou, Lengeris & Ioannou, 2012), which confirm that Greek speakers are indeed not stress deaf. 260 pseudonouns, spoken by a male native speaker of Standard Modern Greek in his thirties with equal prominence (Exp1: syllables were equally stressed, Exp2: syllables were equally reduced/unstressed, 60 nouns served as fillers and retained their original stress prominence), were presented to 20 native speakers of Greek. Participants heard each pseudoword in random order and chose, by clicking on a label on the computer screen, between three (or four in the case of the trisyllabic pseudowords) options: “Stress on the first (syllable)”, “Stress on the second (syllable)”, “Stress on the third (syllable)”, and “It is not clear where the stress falls”. The results of both experimental tasks revealed that Greek speakers could still ‘hear’ stress prominence on some syllable of the pseudoword (the choice of “It is not clear where the stress falls” was 15% in Exp1, a finding that is confirmed by the preliminary results of Exp2). More importantly, speakers showed a strong preference for PU (=non-default) stress for most noun classes (i.e. -as, -a, -ineut, -ifem), with the exception of trisyllabic nouns in -os and -o, where APU (=default) stress was opted (55% and 56%, respectively). Ultimate (U) stress was the least preferred choice. These results were coupled by a production experiment conducted on 32 speakers of Greek who were prompted to read aloud 150 sentences that contained pseudonouns (with no stress diacritic) (Apostolouda, Revithiadou & Papadopoulou, 2011; Apostolouda, 2012).
The experimental findings evidence that Greek speakers dynamically employ their in-built stress representation mechanism when assigning stress. However, it is an open question how stress patterns are distributed among the various morphological classes. Between the lexically-inflicted patterns, PU stress is preferred over U stress in most morphological classes, even when the reverse frequencies hold in the Lexicon, e.g. U > PU. This suggests that speakers apply phonological-grounding by adjusting their outputs towards a less marked (in terms of foot structure) option. The only instance in which speakers’ outputs faithfully reflect lexical statistics involves the elevated results of the APU default in the fossilized os- and o-nouns (54% and 60%, respectively, in the Lexicon based on the Anastassiadis’ Reverse Dictionary of Modern Greek), which suggests that, as far as Greek stress is concerned, lexical knowledge is given enough leeway to breathe in only in less productive noun classes.
To conclude, in this paper we show how the findings of experimental research can provide valuable insights in the nature and processing of stress in lexical stress systems, and propose ways in which such insights can be integrated in their formal analysis. [Top | Abstracts Top]
Generative models and typological frequency
Linguists have long endeavored (i) to describe the typology of human languages and furthermore (ii) to construct theoretical models that explain or motivate this observed typology. The success or failure of a model with respect to the second of these goals is often characterized in terms of undergeneration (i.e., failure to predict/generate an attested pattern) and overgeneration (i.e., the prediction/generation of unattested patterns that are deemed implausible). In recent years, however, the growth of databases like StressTyp2 is making it increasingly possible to give a more nuanced analysis of under- and over-generation in terms of confidence and frequency. In this talk I examine the relationship between generative models of stress systems and typological frequency. I ask whether there is a natural way to incorporate confidence (based on attested frequency) into the evaluation of models of stress and--in order to evaluate this first question--I also ask if there is a natural way to associate generative models with predictions about frequency of typological attestation. [Top | Abstracts Top]
The Acoustic Properties of Prominence and Phonological Contrasts in a Language
While there is much theoretical debate about the terms associated with linguistic prominence (e.g. stress, accent, focus), there is also substantial debate regarding the acoustic properties of these phenomena. The most commonly measured coustic properties are duration, F0 and amplitude, however, several other properties including phonation types and spectral patterns are also often found to correlate with prominence.
One particularly interesting aspect of the acoustic study of prominence is the fact that the same properties used to express prominence (just mentioned) are often used in languages for phonemic contrasts (e.g. contrastive vowel length, tone). Since it is crucial for speakers of a language to perceive both prominence and contrastive information, the question arises as to how this would be possible if they are expressed with the same mechanisms.
The proposed research addresses this issue on the basis of data from Hungarian, Greek, Spanish and Turkish, and proposes that the “Functional Load” of the properties in the phonological system of a language (i.e. as contrastive or not) plays a substantial role in determining the extent to which they are used in the manifestation of prominence, both lexically and sententially. [Top | Abstracts Top]
Multiple dimensions of stress faithfulness in Tagalog
Tagalog exhibits two types of productive stress faithfulness (Sabbagh 2004): segment-for-segment faithfulness in nouns, where the stress remains on the same segment if possible (palít, palít-an); and template faithfulness in verbs, where stress remains either final (bantás, bantas-án) or penultimate (bása, basá-hin).
In most two-syllable reduplication, there is no conflict between the two types of stress. For example, the reduplicative prefix in [patíd-patíd] or [jákap-jákap] is faithful to the base in both senses. But for a base like [doséna], there is a conflict; in this word, it's resolved in favor of template faithfulness ([dóse-doséna]), but in others it's resolved in favor of segmental faithfulness ([bihí-bihíra?]). There is also a conflict between faithfulness to the immediate base of reduplication and faithfulness to any other member of the paradigm (Steriade 1999's lexical conservatism)
Besides the interest of the coexistence of these different types of faithfulness, what's interesting about this case is that a good match to the data can be obtained by training a (maximum entropy) learner on only the simplest cases (patíd-patíd, jákap-jákap). The learner attributes equal weight to all of the faithfulness constraints that could have been responsible. As in Ryan 2010, learners may not track detailed patterns of variation; the treatment of non-basic cases is a side effect of learning the basic pattern. [Top | Abstracts Top]