Abstract
This study investigates the relative contributions of segmental (vowels and consonants) and suprasegmental (intonation and speaking rate) cues to the perception of foreign accents in English. By employing acoustic filtering to isolate prosodic features, the researchers examined how Mandarin and Cantonese listeners judge the "native-likeness" of various speaker groups. The findings reveal that removing segmental information actually reduces the perceived accent of L2 learners, suggesting that pronunciation errors in specific sounds are more salient markers of non-nativeness than prosodic deviations.
Introduction
When listeners encounter a foreign accent, they process two distinct layers of acoustic information: segmental cues, which involve the articulation of individual vowels and consonants, and suprasegmental cues (prosody), which include intonation, stress, and speaking rate. While both contribute to the perception of a "foreign accent," their relative weight remains a subject of debate, particularly when the listeners themselves are non-native speakers.
This research also explores the "Interlanguage Speech Intelligibility Benefit," testing whether listeners are more adept at judging accents from their own native language (L1) background. By comparing Mandarin and Cantonese listeners, the study evaluates if linguistic background influences the prioritization of prosodic cues like pitch and speed.
Methodology
The researchers utilized a custom dataset of 40 spontaneous English utterances from four distinct talker groups: Native Speakers of English (NSE), English-Cantonese bilinguals (EC), Cantonese learners (CE), and Mandarin learners (ME). To isolate prosody from segmental content, they employed a "mumble test" using band-pass filtering based on each speaker's pitch range, rendering words unintelligible while preserving intonation and rhythm.
Listeners (15 Mandarin, 15 Cantonese) rated clips across four experimental conditions on a 5-point native-likeness scale:
- F0: Only intonation (filtered "mumble" speech).
- SR: Only speaking rate modified to match native averages.
- F0+SR: Combined intonation filtering and rate modification.
- Original: Unmodified baseline speech.
Results and Discussion
The experimental results established a clear hierarchy of perceived native-likeness: Native speakers received the highest ratings, followed by bilinguals, with Mandarin and Cantonese learners receiving the lowest scores.
The Segmental "Accent Culprit"
A significant finding, illustrated in the figure above, is the "filtering effect." For L2 learners (ME and CE), the perceived accent actually decreased (native-likeness ratings increased) when segmental information was filtered out (the F0 condition). This suggests that "wrong" vowels and consonants are more jarring to listeners than deviations in intonation. Conversely, for native speakers (NSE), filtering out segments made them sound less native-like, indicating that segmental precision is a primary cue for identifying native speech.
Speaking Rate and Listener Background
Modifying the speaking rate of learners to match native speeds significantly improved their perceived native-likeness. However, this modification did not help native speakers sound "more native."
Interestingly, while Cantonese listeners were generally more accurate at identifying native English samples—likely due to greater English exposure in Hong Kong—both listener groups followed remarkably similar rating patterns. This suggests that listeners rely on universal acoustic cues rather than specific L1 familiarity when judging accentedness.
Conclusion
The study demonstrates that segmental errors are the primary drivers of perceived foreign accents for non-native listeners. While prosodic cues like speaking rate do play a role, the clarity and accuracy of individual sounds remain the most critical factors in sounding "native." Future research could expand on these findings by including native English listeners and exploring how these cues affect intelligibility versus perceived accentedness across different proficiency levels.
Summary updated