BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//conference-hub.linguistic-society.com//athens-2026//spea
 ker//A39ENP
BEGIN:VEVENT
UID:pretalx-athens-2026-XEC9MD@conference-hub.linguistic-society.com
DTSTART:20260424T123000Z
DTEND:20260424T124500Z
DESCRIPTION:Abstract\nThis study aims to determine the influence of L2 pros
 odic features on automatic accent classification.\nSpecifically\, we inves
 tigated which features—durational\, melodic\, intensive\, or voice quali
 ty—are most\neffective at classifying accents as either 'Native' (L1-Eng
 lish) or 'Foreign' (L2-English and L1-Brazilian\nPortuguese). Our hypothes
 is is that a multidimensional matrix of prosodic features is necessary to 
 finely\ndistinguish L1- L2 linguistic differences and enhance automatic ac
 cent classification. For Methodology\,\nthis research integrates principle
 s from phonetics\, L2 prosody\, and Artificial Intelligence (AI). The data
 set\ncomprised 160 read-speech samples from three groups: 80 L1-English (L
 1E) American speakers\, 40 L2-\nEnglish (L2E) proficient Brazilian speaker
 s\, and 40 L1- Brazilian Portuguese (L1BP) speakers. Samples\nwere based o
 n a phonetically balanced text (an Aesop’s fable). Acoustic processing i
 nvolved forcedalignment \nvia Montreal Forced Aligner (MFA)\, manual corre
 ction\, re-alignment into prosodic-level units\,\nand automatic feature ex
 traction using a Praat script. Statistical analysis included a Kruskal- Wa
 llis test\nfollowed by a Dunn test for pairwise comparisons (L1E-L2E\, L1E
 -L1BP\, L2E- L1BP). Finally\, L1E was\ncategorized as ‘Native’ and L2E
 /L1BP as ‘Foreign’ targets for the Automatic Speech Recognition (ASR)\
 nsystem. Preliminary results indicate that long- term spectral (and other 
 intensive) features of voice quality\,\nfollowed by durational features\, 
 were consistent in differentiating the language groups globally and in\npa
 irwise comparisons. These features also showed a moderate-to-high influenc
 e on the accent\nclassification performance. The Machine Learning algorith
 ms achieved classification accuracy levels\nranging from 71% to 100%. Vari
 ables related to duration and intensity were found to be significant\npred
 ictors in the accent classification models. The major conclusion is that p
 rosodic acoustic features\,\nparticularly those related to intensity and d
 uration\, are highly influential in the automatic classification of\nforei
 gn accent. The significance of this study extends to L2 pedagogy and to th
 e L2 Forensic field. For\nthe former\, the predictive power of prosodic fe
 atures suggests that current practices in pronunciation\nclasses that prio
 ritize a segment-narrow-focus approach should be revisited and re-prioriti
 zed to align\nwith suprasegmental instruction (e.g.\, teaching stress\, rh
 ythm\, intonation\, and voice modulation). For the\nlatter\, the identific
 ation of reliable\, automatically-extracted prosodic features provides new
 \, measurable\nacoustic parameters that can be applied to speaker profilin
 g and (foreign) accent characterization in\nunknown or disguised speech sa
 mples. This alignment enhances both the technical performance of ASR\nmode
 ls and the potential for improving L2 communication effectiveness and fore
 nsic speaker analysis.\nKeywords: L2 Prosody. ASR. Acoustic Phonetics. L2 
 Pronunciation Pedagogy. L2 Forensic Field
DTSTAMP:20260419T081020Z
LOCATION:Online Session
SUMMARY:L2 Prosodic Features in Automatic Accent Classification - Leonidas 
 Silva Jr
URL:https://conference-hub.linguistic-society.com/athens-2026/talk/XEC9MD/
END:VEVENT
END:VCALENDAR