Acquisition of Initial /s/-stop and stop-/s/ Sequences in Greek

Syrika, Asimina; Nicolaidis, Katerina; Edwards, Jan; Beckman, Mary E.

doi:10.1177/0023830911402597

“…For example, #sT clusters are often simplified in L1 acquisition (Gerlach, 2010 ). While the most common outcome is deletion of [s] (which results in the #TV sequence), deletion of the stop is robustly attested as well in L1 acquisition (resulting in #sV), both in the general population and in children with speech disorders (Catts and Kamhi, 1984 ; Ohala, 1999 ; Gerlach, 2010 ; Syrika et al, 2011 ). While this deletion likely involves articulatory factors that are lacking in our model, the fact that segmental units can be deleted from the output and recombined in L1 acquisition resembles the deletion in the Generator's innovative outputs, such as the #sV sequence.…”

Section: Discussionmentioning

confidence: 99%

Generative Adversarial Phonology: Modeling Unsupervised Phonetic and Phonological Learning With Neural Networks

Beguš

¹

2020

Front. Artif. Intell.

View full text Add to dashboard Cite

Training deep neural networks on well-understood dependencies in speech data can provide new insights into how they learn internal representations. This paper argues that acquisition of speech can be modeled as a dependency between random space and generated speech data in the Generative Adversarial Network architecture and proposes a methodology to uncover the network's internal representations that correspond to phonetic and phonological properties. The Generative Adversarial architecture is uniquely appropriate for modeling phonetic and phonological learning because the network is trained on unannotated raw acoustic data and learning is unsupervised without any language-specific assumptions or pre-assumed levels of abstraction. A Generative Adversarial Network was trained on an allophonic distribution in English, in which voiceless stops surface as aspirated word-initially before stressed vowels, except if preceded by a sibilant [s]. The network successfully learns the allophonic alternation: the network's generated speech signal contains the conditional distribution of aspiration duration. The paper proposes a technique for establishing the network's internal representations that identifies latent variables that correspond to, for example, presence of [s] and its spectral properties. By manipulating these variables, we actively control the presence of [s] and its frication amplitude in the generated outputs. This suggests that the network learns to use latent variables as an approximation of phonetic and phonological representations. Crucially, we observe that the dependencies learned in training extend beyond the training interval, which allows for additional exploration of learning representations. The paper also discusses how the network's architecture and innovative outputs resemble and differ from linguistic behavior in language acquisition, speech disorders, and speech errors, and how well-understood dependencies in speech data can help us interpret how neural networks learn their representations.

show abstract

“…For example, #sT clusters are often simplified in L1 acquisition (Gerlach, 2010). While the most common outcome is deletion of [s] (which results in the #TV sequence), deletion of the stop is robustly attested as well in L1 acquisition (resulting in #sV), both in the general population and in children with speech disorders (Catts and Kamhi, 1984;Ohala, 1999;Gerlach, 2010;Syrika et al, 2011). While this deletion likely involves articulatory factors that are lacking in our model, the fact that segmental units can be deleted from the output and recombined in L1 acquisition resembles the deletion in the Generator's innovative outputs, such as the #sV sequence.…”

Section: Parallels In Human Behaviormentioning

confidence: 99%

Generative Adversarial Phonology: Modeling unsupervised allophonic learning with neural networks

Beguš¹

2019

Preprint

0

View full text Add to dashboard Cite

Training deep neural networks on well-understood dependencies in speech data can provide new insights into how they learn internal representations. This paper argues that acquisition of speech can be modeled as a dependency between random space and generated speech data in the Generative Adversarial Network architecture and proposes a methodology to uncover the network's internal representations that correspond to phonetic and phonological properties. The Generative Adversarial architecture is uniquely appropriate for modeling phonetic and phonological learning because the network is trained on unannotated raw acoustic data and learning is unsupervised without any language-specific assumptions or pre-assumed levels of abstraction. A Generative Adversarial Network was trained on an allophonic distribution in English, in which voiceless stops surface as aspirated word-initially before stressed vowels, except if preceded by a sibilant [s]. The network successfully learns the allophonic alternation: the network's generated speech signal contains the conditional distribution of aspiration duration. The paper proposes a technique for establishing the network's internal representations that identifies latent variables that correspond to, for example, presence of [s] and its spectral properties. By manipulating these variables, we actively control the presence of [s] and its frication amplitude in the generated outputs. This suggests that the network learns to use latent variables as an approximation of phonetic and phonological representations. Crucially, we observe that the dependencies learned in training extend beyond the training interval, which allows for additional exploration of learning representations. The paper also discusses how the network's architecture and innovative outputs resemble and differ from linguistic behavior in language acquisition, speech disorders, and speech errors, and how well-understood dependencies in speech data can help us interpret how neural networks learn their representations.

show abstract

“…Sonority is one of the factors predicting the sequence ordering mastered by young children (Ohala & Kawasaki-Fukumori 1997;Pater 2009), and the rate and type of errors observed in individuals with developmental or acquired language impairments (Bastiaanse et al 1994;Béland et al 1990;Buckingham 1986;Christman 1994;Romani & Calabrese 1998;Romani & Galluzzi 2005;Romani et al 2002;Stenneken et al 2005). Studies have supported the role of SSP in adult speech perception (Berent et al 2008;Berent et al 2007), continuous speech segmentation (Ettlinger et al 2012), and child language acquisition (Ohala & Kawasaki-Fukumori 1997;Pater 2009) Several studies have tested the sonority related-grammatical markedness through assessing production and perception of SSP-violating onset clusters versus SSP-adhering ones in typically developing children including Greek, Dutch, English, Hebrew, and Norwegian language (Berent et al 2011;Clements 1990;Ohala 1999;Syrika et al 2011;Yavaş et al 2008). It is widely debated whether the sonority related-grammatical markedness is innate or learned (Parker 2008).…”

Section: Sonority and Sonority Sequencing Principlementioning

confidence: 99%

Sonority’s Effect as a Surface Cue on Lexical Speech Perception of Children With Cochlear Implants

Hamza

¹

,

Okalidou

²

,

Kyriafinis

³

et al. 2018

Ear &Amp; Hearing

3

0

View full text Add to dashboard Cite

Children with CIs performed better in lexical tasks relying on the sonority perceptual prominence cues, as in SS-SS condition, than on SSP initial relying conditions as NS-NS and NS-SS. Template-driven word learning, an early word-learning strategy, appears to play a role in the lexical access of children with CIs whether matched in hearing age or not. The SS-SS condition acts as a preferred word template. The longer RTs brought about by the highly accurate SS-SS condition in children with CIs is possibly because listening becomes more effortful. The lack of RTs difference between the children groups when matched on hearing age points out the importance of oral-language exposure period as a key factor in developing the auditory processing skills.

show abstract

Acquisition of Initial /s/-stop and stop-/s/ Sequences in Greek

Cited by 15 publications

References 30 publications

Generative Adversarial Phonology: Modeling Unsupervised Phonetic and Phonological Learning With Neural Networks

Generative Adversarial Phonology: Modeling Unsupervised Phonetic and Phonological Learning With Neural Networks

Generative Adversarial Phonology: Modeling unsupervised allophonic learning with neural networks

Sonority’s Effect as a Surface Cue on Lexical Speech Perception of Children With Cochlear Implants

Contact Info

Product

Resources

About