BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20250801T163203EDT-0645uz9G8I@132.216.98.100
DTSTAMP:20250801T203203Z
DESCRIPTION:Grammar\, Frequency\, and Speech Rate in Phonological\nVariatio
 n\nAndries Coetzee (University of Michigan)\nOver the past two decades\, v
 ariation has been promoted from the\nmargins of phonological theory to its
  center. The success of a\nphonological model is now measured\, inter alia
 \, by how well it\naccounts for phonological variation. Much progress has 
 been made\,\nand there are currently multiple competing models of phonolog
 y\, any\nof which can account for variation with seemingly great success\n
 (Anttila 1997\, Boersma & Hayes 2001\, Coetzee 2006\, Kimper 2011\,\netc.)
 .\nIn this presentation\, however\, I will argue that the success of\nthes
 e models is only apparent. All of these models are exclusively\ngrammatica
 l – they do not allow for factors other than grammar to\ninfluence the rea
 lization of phonological variation. Although\nphonological grammar undoubt
 edly contributes to variation\, there\nare many other factors that also in
 fluence how variation is\nrealized. In a model where grammar alone account
 s perfectly for\nvariation\, grammar is therefore doing more than its fair
  share of\nthe work. Rather than an example of a successful model\, this w
 ould\nbe an example of a model with a too powerful a grammar.\nI will prop
 ose an augmented version of the noisy Harmonic\nGrammar model of phonologi
 cal variation (Coetzee & Pater 2011).\nIn this augmented version\, both gr
 ammatical and non-grammatical\nfactors are incorporated\, and co-determine
  how variation is\nrealized. I will rely on two non-grammatical influences
  on\nvariation as examples: usage frequency and speech rate.\n\nUsage freq
 uency. Many variable phonological processes\napply more often to frequent 
 than infrequent words. The variable\ndeletion of word-final t/d from conso
 nant clusters in\nEnglish\, for instance\, applies more often to frequent 
 just\nthan infrequent jest. The traditional grammatical models\nof phonolo
 gical variation do not differentiate between words based\non their frequen
 cy\, and would predict the same deletion rate for\nboth just and jest. The
  graph to the left serves\nas an illustration. It plots the t/d-deletion r
 ate in\nphrase-final position in the Buckeye Corpus (Pitt et\nal. 2007) ag
 ainst the log frequency of words as measured in\nCELEX (Baayen et al. 1995
 ). The dotted line marks\nthe prediction of a classic\, non-augmented Harm
 onic Grammar model\,\nand the solid line the prediction of the proposed\nf
 requency-augmented model. The augmented model accounts better for\nthe act
 ual deletion rates observed for individual words.\nSpeech rate. Some varia
 ble processes\, such as English\nschwa deletion\, apply more frequently at
  faster than slower speech\nrates. The word potato is more likely to be re
 alized as\np_tato at faster than slower speech rates (Patterson\net al. 20
 03). I will show that listeners use this\ncorrelation between speech rate 
 and deletion during speech\nperception. They are more likely to “perceive”
  an absent schwa\n(i.e. to identify p_tato as an utterance of\npotato) whe
 n listening to faster than slower speech. I\nthen develop a model of the l
 istener’s perceptual grammar in the\nsame augmented version of noisy Harmo
 nic Grammar used above\,\nshowing how this augmented model accounts better
  for listeners’\nactual performance than a classic non-augmented model.\nG
 iven the data sources (speech corpora\, data collected in the\nphonology l
 aboratory\, etc.) that we currently have at our disposal\,\nwe have access
  to more realistic information on actual linguistic\nbehavior. The mathema
 tically and computationally sophisticated\ntheories of grammar currently a
 vailable also enable us to develop\nmore realistic models of grammar. We a
 re at an exciting nexus in\nthe development of linguistic theorizing where
  we can start\nintegrating traditional generative/competence models with a
 ctual\nproduction data.\n
DTSTART:20120127T200000Z
DTEND:20120127T220000Z
LOCATION:Education Building\, CA\, QC\, Montreal\, H3A 1Y2\, 3700 rue McTav
 ish
SUMMARY:9I��������� Linguistics Phonology Job Talk - Andries Coetzee (University
  of Michigan)
URL:/channels/event/mcgill-linguistics-phonology-job-t
 alk-andries-coetzee-university-michigan-213484
END:VEVENT
END:VCALENDAR