TL;DR
This study investigates how much of an upcoming word native English speakers (L1) and Mandarin speakers learning English (L2) plan ahead when speaking. They found that L1 speakers plan further ahead, considering phonetic details of the next word, while L2 speakers tend to plan more locally, especially those with lower proficiency. This suggests that the scope of speech planning is influenced by language background and proficiency.
What Problem Does This Paper Address?
Understanding how we plan our speech is a fundamental question in linguistics and cognitive science. Specifically, this paper tackles the "speech planning window" – how far ahead in the speech stream we mentally prepare our utterances. Previous research has suggested that speakers might plan in "phrasal chunks," implying that phonetic details of upcoming words could influence the current word's production. However, there's been little agreement on the exact scope of this planning and how it might differ between native speakers and language learners. This study aims to clarify this by examining how phonetic features of the next word (specifically, its place of articulation and voicing) affect the production of consonant clusters at the boundary between two words.
Methods
The researchers used acoustic analysis of recorded speech to measure three key phonetic cues related to consonant clusters:
- Release Likelihood: Whether the first consonant (C1) in a cluster is fully released (i.e., a distinct burst of sound).
- Closure Duration: The length of time the vocal tract is closed during the production of the consonant cluster.
- Closure Shortening Ratio (Overlap): How much the closure duration of the cluster is reduced compared to when the consonants are produced in isolation, indicating coarticulation or overlap.
These cues were analyzed using generalized linear mixed-effects models, accounting for factors like speaker group (Native English, High-Proficiency Mandarin L2, Low-Proficiency Mandarin L2), phonetic features (place of articulation, voicing), and lexicality (real words vs. nonsense words).
Datasets
The study used recordings of participants producing specially designed stimuli. These stimuli included:
- Real words: Phrases containing stop-stop consonant clusters (e.g., "bad kid").
- Nonce phrases: Similar-sounding nonsense phrases designed to control for lexical familiarity (e.g., "peep pate" as a counterpart to "keep pace").
The recordings were analyzed using the Praat software. No new datasets were introduced; rather, existing speech production data was analyzed.
Key Insights and Ideas
- L1 Planning Extends Across Word Boundaries: Native English speakers showed clear evidence of planning phonetic details of the second consonant (C2) when producing the first consonant (C1) in a cross-word cluster. This was seen in how C1's release likelihood and closure duration were influenced by C2's place of articulation.
- L2 Planning is More Localized: Mandarin-speaking learners of English, especially those with lower proficiency, exhibited less influence from C2's phonetic features on C1 production. Their speech planning appeared to be more focused on the immediate, local segment.
- Place of Articulation (POA) is a Universal Constraint: Both native and non-native speakers showed similar patterns related to the place of articulation of consonants. For instance, front-back clusters tended to have more C1 releases and shorter closure durations than back-front or homorganic clusters. This suggests POA is a fundamental articulatory constraint that transcends language background.
- Voicing is Language-Specific and More Challenging for L2 Learners: The voicing of C2 had a significant impact on C1 release likelihood for native English speakers, but not for L2 speakers. This highlights that language-specific phonetic contrasts, like voicing in English (which is not contrastive in Mandarin stops), are more difficult for learners to integrate into their speech planning.
- External Sandhi is Blind to Lexicality in L1: The observed patterns in native English speakers were consistent across both real words and nonsense words. This suggests that the "external sandhi" (phonetic adjustments at word boundaries) operates automatically and is not dependent on lexical familiarity or cognitive load in L1.
Key Results
- Release Likelihood: The low-proficiency L2 group (NNL) released C1 significantly more often than the native group (NE), indicating a difference in how they handle consonant clusters.
- Closure Shortening Ratio: The NNL group showed significantly less closure overlap (shorter shortening ratio) than the NE group, meaning their consonants were less coarticulated across word boundaries.
- Voicing Effect: Native English speakers significantly increased C1 release likelihood when C2 was voiceless compared to voiced. This effect was not observed in either L2 group.
- Durational Cues: All groups produced longer closure durations when C2 was voiced. However, the high-proficiency L2 group (NNH) showed more pronounced deviations from the native group in terms of closure duration, particularly in nonwords.
Limitations and Future Work
- L2 Proficiency Levels: While the study included two groups of L2 learners (high and low proficiency), further research could explore a more granular spectrum of proficiency to pinpoint specific developmental stages.
- Perception: The study focused on production. Future work could investigate whether these production differences are mirrored in perception, i.e., whether L2 learners also process phonetic details across word boundaries differently.
- Other Phonetic Features: The study focused on place of articulation and voicing. Exploring other phonetic features (e.g., aspiration, manner of articulation) could provide a more comprehensive understanding of speech planning.
- Theoretical Implications: The authors suggest that their findings support a model where phonetic encoding is tightly integrated with prosodic structure and operates with a long-distance planning scope in native speech, contrasting with the more localized planning observed in L2 speakers. This opens avenues for refining models of speech production and second language acquisition.
Summary updated