|Friday April 5|
|1:30 – 1:45||— Opening Remarks —|
|1:45 – 2:30||Sharon Inkelas||Modeling scalar vowel strength in Q Theory|
|2:30 – 3:15||Rachel Walker||Gradient feature activation and the special status of coronals|
|3:15 – 4:00||Caitlin Smith||Partial Vowel Height Harmony and Partial Transparency via Gestural Blending|
|4:00 – 4:30||— Coffee Break —|
|4:30 – 5:15||Sam Tilsen||Motoric mechanisms for the emergence of non-local phonological patterns|
|5:15 – 6:00||Scott Moisik||Phonology begins in the body: Explorations of laryngeal quantality|
|Saturday April 6|
|9:00 – 9:30||— Breakfast —|
|9:30 – 10:15||John Kingston||Variability and contrast [CANCELLED]|
|10:15 – 11:00||Stephanie S Shih & Hayeun Jang||Categoricity in gradience|
|11:00 – 11:45||Florian Lionnet||Phonetically grounded gradient faithfulness: the case of featural affixation in Laal|
|11:45 – 1:30||— Lunch —|
|1:30 – 2:15||Paul Smolensky, Matthew Goldrick & Eric Rosen||Gradient symbolic computation in phonological theory|
|2:15 – 3:00||Eva Zimmermann||Gradience in Phonology: The Argument from Exceptions|
|3:00 – 3:30||— Coffee Break —|
|3:30 – 4:15||Claire Moore-Cantwell||Emergence of lexical idiosyncrasy in language change: an iterated learning simulation|
|4:15 – 5:00||Anne Pycha||Perception and memory of individual morphemes in spoken words: Two experiments|
|5:00 – 7:00||— Reception —|
“Modeling scalar vowel strength in Q Theory”
This paper, based on joint work with Karee Garvin, Myriam Lapierre, and Martha Schwarz, uses the Q-Theoretic scale of subsegmental quantity to model scalar vowel strength effects, as manifested in phenomena ranging from quantity-sensitive stress to faithfulness effects. Using the novel possibility within Q Theory (e.g., Shih & Inkelas 2019) of representing a vowel with one, two, three or even four subsegments, quantity-sensitive constraints are developed within Optimality Theory which generate implicational predictions across and within languages. Briefly, vowels with more subsegments are predicted to be better stress hosts, and to resist alternations more strongly, than vowels with fewer subsegments. Implications of subsegmental quantity for vowel quality contrast possibilities are also discussed.
“Variability and contrast”
In this talk, I appeal to variability across contexts within a language and across languages in the pronunciations, i.e., the phonetic correlates of laryngeal contrasts between obstruents to argue that laryngeal realism demands more concrete definitions of phonological contrasts than can be supported by the facts. The gist of the argument is that laryngeal realism focuses too narrowly on one phonetic measure, voice onset time, in one context, utterance-initial position, to provide a comprehensive phonetic definition of the contrasts. It also relies on too liberal a definition of passive voicing, together with ad hoc phonetic implementation procedures, in its account of the realization of these contrasts in other contexts. Curiously, laryngeal realism is also unrealistic in arguing that some member of the contrast is not specified phonologically and not defined phonetically. I propose an alternative, articulatory explanation for the evidence that has been argued to show that one member of the contrast is specified and not the other, namely, that voice onset time varies with speaking rate for the specified but not the unspecified member. Time permitting, I also challenge the arguments for laryngeal realism from diachrony.
“Phonetically grounded gradient faithfulness: the case of featural affixation in Laal”
I describe and analyze the unusually complex multifeatural plural affix /L [+high, +round]/ of Laal (endangered isolate, Chad), focusing on its most intriguing property: the realization of the [+round] subexponent, which is conditioned by the presence of a labial consonant in the base. Instrumental evidence shows that the conditioning factor is the rounding coarticulatory effect exerted on the vowel by the adjacent labial consonant. This is evidence that coarticulation has a role to play in phonology. Specifically, I show that the degree of faithfulness to a feature value [αF] can be weakened if the realization of that feature is affected by coarticulation. I propose an analysis utilizing both subfeatural representations and scalar faithfulness constraints, both of which are shown to be necessary. This analysis is shown to both confirm and supersede Steriade’s (2009) P-map hypothesis. The subfeatural analysis is compared to an Agreement by Correspondence alternative, shown to be less satisfactory.
“Phonology begins in the body: Explorations of laryngeal quantality”
In this talk I look at the flip-side of the phonological gradience issue: categorical phonetics. Building on recent elaborations of Stevens’ (1989) Quantal Theory in the domain of quantal articulatory-biomechanical relations, I explore computational modelling of the larynx and vocal tract imaging data (including laryngoscopy and MRI) to characterize the states and interactions underpinning phonological laryngeal phenomena, both articulatory and phonatory.
“Emergence of lexical idiosyncrasy in language change: an iterated learning simulation”
Across languages, higher-frequency words deviate more from the phonological grammar, and exhibit more extreme and more idiosyncratic behavior than lower-frequency words do. In this talk, I present the results of an iterated learning simulation using Representational Strength Theory, which conceives of lexical entries as having gradiently weighted phonological representations. The gradience simulates gradient memory resources allocated to features of words: some features are highly predictable and require fewer memory resources, while others are unpredictable or idiosyncratic, and require more memory resources. The simulation mimics the transmission of language across generations, and shows that the transmission process itself, iterated over many generations, can cause high-frequency words to deviate from the phonological grammar and develop idiosyncratic preferences in their behavior.
“Perception and memory of individual morphemes in spoken words: Two experiments”
I report the results of two experiments that investigate how people perceive and remember individual morphemes in spoken American English words. Experiment 1 was a noise-rating task that probed perception. Participants heard stimuli such as the prefixed word inward or the suffixed word drinker, where strikethrough indicates the presence of white noise at a controlled signal-to-noise ratio. The task was to assign a rating indicating the loudness of the noise. Results showed that for complex words, the target morpheme (root versus affix) and the type of word (prefixed versus suffixed) affected ratings; for simple control words such as insect and glitter, however, no significant effects occurred. Experiment 2 was an eye-fixation task that probed memory. Participants were first exposed to a spoken stimulus such as in-ward or drink-er while associating each individual morpheme with a different shape on the computer screen. Immediately afterwards at test, they saw the associated shapes again, and they were asked to look at the shape associated with a particular morpheme, such as -er. Results showed that for complex words, no significant effects occurred. For simple words, however, the target morpheme and type of word affected times-to-first-fixation. Taken together, these studies support the conclusion that listeners perceive and remember the parts of complex words differently from the parts of simple ones. This suggests, in turn, the potential need for phonological representations which are sensitive to morphological constituency.
Stephanie S Shih & Hayeun Jang
Categoricity in gradience
With the recent increased acceptance of gradience in phonological behaviours, we argue in this talk that categoricity is still crucial in phonological representations. After all, categorization is central to human cognition (e.g., Harnad 2017). What is possible now, versus before, is to revisit our traditional assumptions of categorical symbols from the bottom up: categoricity necessarily emerges from gradience. We highlight cases from our recent work that show, across a number of phonological domains, that categories are emergent from surface-gradient data: e.g., lexical and grammatical word categories can be learned from the natural speech stream; and the coordinations of separate muscle activations give rise to symbolic features in phonological contexts.
“Partial Vowel Height Harmony and Partial Transparency via Gestural Blending”
In the Gestural Harmony Model (Smith 2016, 2018), vowel and vowel-consonant harmony is the result of the extended activation of sub-segmental gestures as defined in Articulatory Phonology (Browman & Goldstein 1986, 1989). In this model, harmony is triggered when a gesture extends its period of activation and overlaps the gestures of other segments in a word. Transparent segments are those that include a gesture that is antagonistic to a harmony triggering gesture, allowing that segment to resist the effect of the harmonizing gesture via the mechanism of gestural blending (Saltzman & Munhall 1989). Partial transparency (Smith 2017), in which a segment appears to have neither fully undergone harmony nor fully resisted it, arises when blended antagonistic gestures are of equal strength, resulting in the production of a segment that is intermediate between them. In this talk, I extend the gestural blending analysis to cases of partial height harmony. In partial height harmony, a non-high vowel target approaches the height of a high vowel trigger, but does not necessarily take on its exact height specification. In some cases, low target vowels are raised to mid vowels while mid vowels are raised to high vowels. I analyze such raising patterns as the distinct results of the blending of antagonistic trigger and target vowel gestures of different heights. In particular, the blending of the gestures of a low vowel target and a high vowel trigger are viewed as a case of partial transparency to height harmony: the blending of equally strong high and low vowel gestures results in the production of a vowel whose height is intermediate between the two. One advantage of adopting this gestural model of harmony is that cases of partial height harmony need not be analyzed as synchronic chain shifts, avoiding the need for additional grammatical architecture to account for them. It also provides a unified account of partial transparency and partial height harmony, both of which involve some segments undergoing harmony to only an intermediate degree.
Paul Smolensky, Matthew Goldrick & Eric Rosen
“Gradient symbolic computation in phonological theory”
Harmonic Grammar and Optimality Theory are grammar frameworks that arise from a theory of neural computation in which both optimization and gradience play key roles. This theory predicts the existence of mental representations in which symbols in structures have a gradient degree of presence (or ‘activity’), and multiple gradiently-active symbols can blend together in a single structural position. These Gradient Symbol Structures can serve as inputs and/or outputs of the grammar. I will illustrate with analyses that highlight differences between gradient and discrete representations in phonological theory. I will also discuss learning algorithms that compute the gradient activity level of phonological elements, and the gradient strengths of grammatical constraints, that are implicated in a given empirical pattern.
“Motoric mechanisms for the emergence of non-local phonological patterns”
The origins of many local phonological patterns can be well understood as a consequence of gestural overlap, in the framework of Articulatory Phonology. But patterns which involve temporally distant sounds are not as easily modeled in that framework. Indeed, there is disagreement regarding whether such patterns require non-local mechanisms, or whether all long-distance patterns can be reduced to temporal extension of articulatory gestures. A number of researchers have argued that at least some long-distance patterns—in particular, consonant agreement harmonies—cannot be reduced to feature spreading and thus are not readily analyzed with temporally extended gestures (see e.g. Hansson 2001: Theoretical and Typological Issues in Consonant Harmony). This talk argues that non-local agreement patterns originate via a mechanism which differs from spreading, but which can nonetheless be interpreted as local. This mechanism requires several revisions to the framework of Articulatory Phonology, including the allowance that gestures which are part of a speech plan, even when inactive, can exert an influence on the state of the vocal tract. Various typological differences between spreading and agreement harmonies are argued to follow from the proposed model.
“Gradient feature activation and the special status of coronals”
Overview. In languages that show asymmetries by place of articulation, coronal consonants tend to exhibit weaker markedness effects and/or a weaker degree of faithfulness in comparison to labials and dorsals (de Lacy 2002, 2006). In this work, I propose that this dual behavior arises from a weaker activation of the feature [Coronal] than [Labial] and [Dorsal], using gradient symbolic representations (GSRs) (Smolensky & Goldrick 2016).
Coronal behavior. It is well established that coronal consonants exhibit special behavior in phonological patterns in contrast to other supralaryngeal places of articulation, labial and dorsal (e.g. Paradis & Prunet 1991). Coronals may show a special status in two ways (see (1)): (a) coronals may be exceptionally permitted or inactive in some context where non-coronals are not, and (b) coronals alone capitulate in some phonological phenomenon.
(1) Special behavior of coronals
|a.||Pattern: Coronals are exceptionally permitted / inactive||Ex. language||References|
|i.||Coronals are exempted from phonotactic restrictions||English, German||Clements 1990, Hall 2002|
|ii.||Coronals are exempted from OCP effects||English, German||Davis 1991, Coetzee 2004|
|iii.||Coronals are exempted from role of trigger and/or target in place assimilation||Sri Lankan Portuguese Creole||Smith 1978, Hume & Tserdanelis 2002, de Lacy 2006|
|iv.||Epenthetic consonants are coronal||Axininca Campa||Payne 1981, McCarthy & Prince 1993|
|b.||Pattern: Coronals alone capitulate||Ex. language||References|
|i.||Coronals alone are neutralized||Yamphu||de Lacy 2006|
|ii.||Coronals alone are targets of place assimilation||Catalan||Mascaró 1976, Kiparsky 1985, Wheeler 2005, de Lacy 2006|
Patterns of type (1a) are argued to reflect a higher degree of markedness for [Labial] and [Dorsal] than [Coronal] (McCarthy 1993, Prince & Smolensky 1993, Smolensky 1993, de Lacy 2002, 2006, Lombardi 2002, cf. Boersma 1998). For patterns of type (1b), it has been proposed that [Labial] and [Dorsal] are subject to a higher degree of faithfulness than [Coronal] (de Lacy 2002, 2006, Jun 2004).
Proposal. Employing GSRs (Smolensky & Goldrick 2016), place features are proposed here to have a gradient degree of activation such that [Dorsal], [Labial] > [Coronal]. The weaker activation of [Coronal] is present in both input and output representations, predicting that [Coronal] will incur a lesser violation of markedness and faithfulness constraints for Place than non-coronal features. The approach is illustrated with application to two patterns of nasal place assimilation. In Sri Lankan Portuguese Creole, /n/ is exempted as a target for assimilation due to its lesser violation of markedness, while in Catalan, /n/ is solely targeted by assimilation because of its lesser violation of faithfulness. The special behavior of coronals in both of these patterns follows from a unified source, namely, weaker activation in the representation. This approach eliminates a need for mirrored markedness and faithfulness constraint scales over place features (de Lacy 2002, 2006).
Implications. In terms of the treatment of gradience in representations, this study contributes to a growing body of work utilizing GSRs at the featural level (Jang 2019, Lee 2019, cf. Boersma 1998, Lionnet 2016, 2017). Moreover, it lends support to GSRs with partial activation in the output (Faust 2017, Faust & Smolensky 2017, Zimmermann 2017, 2019).
Discussion will be brought to the origin of lesser activation of [Coronal]. The concept finds some common ground with proposals for coronal underspecification (e.g. Kiparsky 1985, Avery & Rice 1989, Paradis & Prunet 1989), although an underspecified feature is fully absent. In this vein, it would be valuable in future work to examine gradience in patterns involving coronals and their analysis using GSRs.
“Gradience in Phonology: The Argument from Exceptions”
The assumption of Gradient Symbolic Representations (Smolensky and Goldrick, 2016; Rosen, 2016) that phonological elements can have different degrees of activation allows a unified explanation for the typology of phonological exceptions. The crucial theoretical mechanism for exceptional behaviour are gradient constraint violations: Elements which are weakly active are, first, not fully protected by faithfulness constraints and, second, don’t fully create markedness problems. I argue that this simple mechanism predicts the attested typology of exceptional a) non-undergoers, b) undergoers, c) non-triggers, and d) triggers for otherwise regular phonological processes within a language (Pater, 2006; Finley, 2010).
Furthermore, the assumption that morpheme-specific phonological behaviour within one language arises from gradient differences in the activity of phonological elements makes some predictions that set this model apart from alternative approaches to exceptionality (as autosegmental defectivity (e.g. Lieber, 1987; Tranel, 1996; Zoll, 1996) or lexically indexed constraints (e.g. Pater, 2006; Flack, 2007; Mahanta, 2012)). First, it predicts that exceptional elements can be exceptional for multiple processes. This is shown with a case study of tone spreading in Molinos Mixtec where certain tones are exceptional non-undergoers of one process and exceptional non-triggers for another. A representational account where the gradient activity of the tones is the explanation for exceptional behaviour predicts exactly such an accumulation of exceptional behaviour. Second, it predicts different degrees of exceptionality. This point is illustrated with a case study of ghost consonants in Welsh. There are at least two classes of exceptional ghost consonants that differ a) in whether they appear to resolve a markedness problem or whether they disappear to avoid a markedness problem and b) in the markedness thresholds that trigger their exceptional behaviour. The assumption of three different activity levels for segments in Welsh straightforwardly explains the regular and the two exceptional classes. Third, it predicts implicational relations between exceptionality classes within a language. If, for example, one morpheme class is an exception and fails to trigger/undergo a process P2 but regularly triggers/undergoes a process P1, then it is impossible under the gradience account that yet another morpheme class is only exceptional for P1 but not P2 if both refer to the same phonological structure. The typology of exceptions seems to confirm such general restrictions.