Learning Prosodic Focus from Continuous Speech Input:A Neural Network Exploration

Abstract

This study uses connectionist modeling to explore whether and how infants might learn prosodic focus directly from continuous speech input. Focus is a communicative function that serves to put emphasis on a particular part of an utterance, and it is mainly encoded by pitch variations. The acquisition of focus entails two major difficulties. The first is that focus-related pitch patterns are confounded by other linguistic functions that also use pitch for their encoding, such as lexical tone in a tone language. Second, speakers have different pitch ranges, which further confounds the focus related pitch patterns. In three simulations using self-organizing neural networks, we explored how focus may be learned from continuous acoustic signals in Mandarin that were produced with cooccurring lexical tones and by multiple speakers. We used sentence-sized F 0 contours as well as their velocity profiles (D 1 ) as training input. Results show that both F 0 and D 1 contours provide information for focus learning, but only the D 1 -trained network adequately handled the variability introduced by cross-gender differences. The recognition rate was analogous to human performance. Implications of these findings for theories of language acquisition and adult speech perception are discussed.

Interconnected growing self-organizing maps for auditory and semantic acquisition modeling

“…Their research on semantic modeling revealed that SOM has the ability to detect the “logical similarity” between words and group similar words into clusters. In recent years, the topography-preserving ability and the self-organizing ability of the SOM have been increasingly applied to tasks involving the modeling of acquisition, especially in linguistic field, covering many areas including tonal acquisition (Gauthier et al, 2007a , b , 2009 ), lexical acquisition and development (Li et al, 2004 , 2007 ; Hernandez and Li, 2007 ), bilingual lexical development (Zhao and Li, 2007 , 2008 , 2010 ; Li, 2009 ; Zinszer and Li, 2010 ), grammatical acquisition (Li and Zhao, 2009 ; Zhao and Li, 2009 ), semantic representation (Zhao et al, 2011 ), the relation between sensory, and motor states (Kröger et al, 2006a , b ), the acquisition of vowel and consonant auditory categories (ger et al, ), the acquisition and development of articulatory movements (Kröger et al, 2011a , b ; Warlaumont et al, 2013 ), etc. Li and Zhao ( 2013 ) provide an excellent review of SOM-based language models.…”

Modeling intrinsic intelligibility variation: vowel-space size and structure

“…From those 15 sentences the average pitch range magnitude was calculated for each talker, 4 as well as the mean absolute value of rate of change in pitch ("pitch dynamicity, " cf. measures of "pitch velocity" used in Gauthier et al 2007aGauthier et al ,b, 2009. Pitch dynamicity was included as a measure of how dramatic the rises and falls in ƒ₀ were across each sentence, irrespective of overall ƒ₀ downtrend.…”

O Uso De Redes Neurais Auto- Organizáveis Na Visualização De Formação De Agrupamentos a Partir Do Conhecimento Acentual De Aprendizes Brasileiros De Inglês

“…Todavia, os modelos de redes neurais auto-organizáveis (Self-organizing maps -SOM), comumente chamados de rede de Kohonen, ao longo dos anos têm sido também usados como um modelo exploratório primário de fenômenos lingüísticos (Cf. FARKAS;CROCKER, 2008;GAUTHIER, 2009; LI ET LI ET AL., 2007;MIIKKULAINEN, 1997).…”

“…Igualmente, ao se utilizar do SOM, Gauthier et al (2009) investigou se e como as crianças poderiam aprender foco prosódico diretamente de um insumo (input) contínuo da fala. Os autores exploraram como o foco poderia ser aprendido por meio de sinais acústicos contínuos em Mandarim, que eram produzidos com tons lexicais co-ocorrentes e por vários falantes.…”

“…A despeito de alguns trabalhos prévios sobre modelagem conexionista envolvendo traços prosódicos para o desenvolvimento e identicação da linguagem (BLANC;DOMINEY, 2003;CUMMINS et al, 1999;GAUTHIER et al, 2009), ao que parece, ainda não houve nenhuma investigação sistemática ou análise exploratória do fenômeno de transferência negativa de padrão acentual de L1 para L2 por meio de um modelo conexionista utilizando a rede SOM.…”