Postersession 3
Poster #: 102
Topic: Speech and language (incl. deficits)
Friday, Sep 11, 2015
1st floor

Temporo-spatial decomposition of MMN reveals underspecified phoneme representations

Arild Hestvik1 & Karthik Durvasula2

1Linguistics and Cognitive Science, Universit of Delaware, Newark, DE, United States
2Linguistics, Michigan State University, Germany

The theory of phonological underspecification (Halle, 1959; Hall, 2007) states that the phonemes that make up words in long-term memory are only coded for a subset of the articulatory features needed for pronounciation. Recent psychological support for underspecification comes from acquisition studies (Fikkert and Levelt, 2008) and ERP studies (Eulitz and Lahiri, 2004; Lahiri and Reetz, 2010). Our study tested a theoretical proposal by (Iverson and Salmons, 1995) that English voiceless stops are coded for a voicelessness feature in lexical representations, but that voiced stops are not coded (i.e. underspecified) for voicing specification. We tested this by utilizing the multi-standard MMN paradigm (Phillips et al., 2000), which causes a phonetic representation to be compared to a phonological representation, allowing us to test a fully specified feature matrix against a partially specified feature matrix. We present here both new data and a reanalysis of results presented at the previous MMN conference, by using difference waveforms as input to temporo-spatial PCA (Dien and Frishkoff, 2005; Dien, 2012); which captures spatial differences in topography of MMN for /d/ vs. /t/. As predicted by underspecification theory, we found a greater MMN to a phonetic oddball [d] in two multiple-standard MMN experiments (with and without attention to the stimuli). Furthermore, we predicted, and observed significant MMNs for both /d/ and /t/ in a single-standard MMN paradigm (but /d/ and /t/ MMNs were observed via ICA to have different spatial distributions). The results support a theory with underspecification at the phonological, but not phonetic level.