Automatic sociophonetics: Exploring corpora with a forensic accent recognition system

Area of Science:

• Computational Linguistics
• Sociophonetics
• Phonetics

Background:

• The York ACCDIST-based automatic accent recognition system (Y-ACCDIST) is primarily designed for forensic casework.
• Sociophonetic research can leverage Y-ACCDIST to analyze linguistic variation within corpora.
• The PEBL (Panjabi-English in Bradford and Leicester) corpus provides a dataset for exploring Y-ACCDIST's capabilities.

Purpose of the Study:

• To demonstrate Y-ACCDIST's utility as a preliminary screening tool for sociophonetic corpora.
• To explore how Y-ACCDIST outputs can objectively assess speaker similarities across linguistic varieties.
• To investigate the corroboration between Y-ACCDIST findings and traditional phonetic analysis.

Main Methods:

• Utilizing Y-ACCDIST to classify speakers based on language background and region.
• Implementing Y-ACCDIST cluster analysis to group speakers and identify potential communities of practice.
• Performing Y-ACCDIST feature selection to identify phonemes crucial for distinguishing speaker groups.

Main Results:

• Y-ACCDIST efficiently and objectively assesses speaker similarities.
• Cluster analysis reveals groupings consistent with localized social networks.
• Feature selection identifies key phonemes for differentiating speaker groups.

Conclusions:

• Y-ACCDIST serves as a valuable tool for preliminary screening and analysis of sociophonetic data.
• The system's outputs can corroborate and enhance traditional sociophonetic analyses.
• Y-ACCDIST facilitates the identification of linguistic variation and social networks within corpora.