Linguistic rhythm in foreign accent

Jiahong YuanInterspeech 2010

TL;DR

This paper investigates how a speaker's native language (L1) influences the rhythm of their second language (L2) English. The study finds that L1 rhythm characteristics, specifically related to vowel duration and pitch, transfer to L2 English, offering a potential way to identify foreign accents.

What Problem Does This Paper Address?

Spoken languages are known to have distinct rhythmic patterns, often categorized as stress-timed, syllable-timed, or mora-timed. While these categories are impressionistic, research suggests they impact how languages are perceived and understood. A key question in second language acquisition is the extent to which these native language rhythmic features transfer to a learner's L2. This paper specifically addresses how L1 rhythm influences the prosodic salience of L2 English, particularly focusing on the duration and pitch of stressed and reduced vowels, which are crucial elements of rhythm. Understanding this transfer is important for improving intelligibility and developing better accent reduction tools.

Methods

The researchers utilized the GMU Speech Accent Archive, which contains recordings of native and non-native English speakers reading a standardized paragraph. To ensure the quality of the speech samples, they automatically determined speech boundaries and then filtered out "incompetent" speakers by analyzing two acoustic measures: the number of pauses longer than 0.5 seconds and the percentage of silence in the speech sample. Samples with fewer than six long pauses were selected.

For acoustic analysis, phone durations were calculated from forced alignments. Pitch (F0) was extracted using esps_get_f0. Both duration and F0 were normalized. Durations were normalized by dividing by the total speech duration (excluding pauses) and multiplying by 100. Pitch was converted to semitones using a base frequency defined as the 10th percentile of all F0 values in a given speech paragraph. The analysis focused on the mean normalized duration of stressed and reduced vowels, and the normalized pitch contours of stressed vowels, segmented into four equal parts.

Datasets

The study used data from the GMU Speech Accent Archive. This archive contains speech samples from native English speakers and non-native speakers of English from various L1 backgrounds. The specific paragraph read by speakers was: "Please call Stella. Ask her to bring these things with her from the store: Six spoons of fresh snow peas, five thick slabs of blue cheese, and maybe a snack for her brother Bob. We also need a small plastic snake and a big toy frog for the kids. She can scoop these things into three red bags, and we will go meet her Wednesday at the train station."

The paper does not introduce a new benchmark dataset. Instead, it selects a subset of existing data based on fluency criteria. The number of selected speech samples for each language group were:

  • Stress-timed: English (native): 203, German: 14, Russian: 8
  • Syllable-timed: French: 9, Italian: 11, Spanish: 26

Key Insights and Ideas

  • L1 Rhythm is a Strong Predictor of L2 Rhythm: The core finding is that a speaker's native language rhythm significantly influences their L2 English rhythm. This suggests that prosodic patterns are not easily shed when learning a new language.
  • Duration Ratios as Accent Indicators: The study highlights that the ratio between the duration of stressed and reduced vowels in L2 English can serve as an indicator of foreign accent. This is a more nuanced approach than simply looking at absolute durations.
  • Pitch Contour Differences: Stressed vowels in L2 English spoken by speakers of syllable-timed languages exhibit different pitch contours compared to those from stress-timed language backgrounds. This difference in pitch movement contributes to the perception of rhythm and accent.
  • Clustering of L1 Groups: Speakers from stress-timed L1s (German, Russian) showed L2 English rhythms that were more similar to native English speakers than those from syllable-timed L1s (French, Italian, Spanish). This supports the idea that languages with similar rhythmic foundations will lead to more similar L2 transfer.

Key Results

  • Vowel Durations: Compared to native English (L1 English), L2 English speakers generally had shorter stressed vowels and longer reduced vowels.
  • Pitch Contours: Stressed vowels in L2 English from speakers of stress-timed languages (German, Russian) had higher normalized pitch contours than those from syllable-timed languages (French, Italian, Spanish).
  • Fluency Filtering: The acoustic measures for fluency (pause duration and silence percentage) were highly correlated (r = 0.90), indicating a robust method for selecting high-quality speech samples.

Limitations and Future Work

Limitations: * The study focuses on a limited set of L1 languages, primarily European ones. The transfer of rhythm from non-European languages might differ. * The analysis of pitch contours is simplified into four segments. More detailed temporal analysis might reveal finer distinctions. * The study relies on impressionistic rhythm categories (stress-timed vs. syllable-timed) which, as the authors note, lack strict phonetic evidence.

Future Work: * Investigate the influence of L1 on L2 rhythm in a wider range of L1 languages, including those from different language families. * Explore the perceptual consequences of these rhythmic differences on L2 intelligibility. * Examine other acoustic correlates of rhythm, such as the duration of consonants or the timing between stressed syllables. * Develop more sophisticated computational models for accent identification that incorporate these rhythmic features.

Summary updated

BibTeX
@inproceedings{yuan2010linguistic,
  author = {Jiahong Yuan},
  title = {{Linguistic rhythm in foreign accent}},
  booktitle = {Interspeech},
  year = {2010},
}
Figure