TL;DR
This paper investigates how the way we group words together with pauses and emphasis (prosody) helps us understand ambiguous sentences in Mandarin. They found that prosodic grouping, particularly the length of certain phrases and the placement of pauses and stress, significantly influences how listeners interpret sentences with ambiguous relative clauses, supporting the Implicit Prosody Hypothesis.
What Problem Does This Paper Address?
Spoken language is full of ambiguity. One common type is "relative clause attachment ambiguity," where a descriptive phrase (the relative clause) could modify one of two nouns in a sentence. For example, in "the servant of the actress who was on the balcony," was it the servant or the actress who was on the balcony? This ambiguity is often resolved by subtle cues in how the sentence is spoken – its prosody. The Implicit Prosody Hypothesis (IPH) suggests that speakers unconsciously project a preferred prosody onto sentences, which then guides their interpretation of ambiguous structures. However, how this works in Mandarin, a tonal language with a different word order for relative clauses compared to English, is less understood. This paper aims to explore how Mandarin speakers use prosodic grouping to resolve relative clause attachment ambiguity, testing the IPH across different aspects of language processing.
Methods
The researchers employed a multi-faceted approach to investigate prosodic grouping and ambiguity resolution in Mandarin:
- Preference Survey: Native Mandarin speakers were presented with ambiguous sentences and asked to choose which noun the relative clause modified. This was done in silent reading to gauge implicit preferences.
- Production Experiments:
- Default Reading: Participants read target sentences aloud to capture their natural, "out-of-the-blue" prosody.
- Contrast Reading: Participants were explicitly asked to produce sentences with either a high or low attachment meaning, often using a "clarification" frame ("I mean X, but not Y") to force contrastive prosody. This allowed researchers to see how speakers consciously manipulate prosody.
- Perception Experiment: Synthetically manipulated audio stimuli were created, varying the placement of pauses and the prominence (stress) of specific words. Listeners then heard these stimuli and were asked to resolve the ambiguity, revealing how prosodic cues influence perception.
The researchers analyzed prosodic features like pause location and prominence using the M_ToBI (ToBI for Mandarin) convention for labeling prosodic events.
Datasets
The study utilized a set of 8 target sentences designed to create relative clause attachment ambiguity in the structure "RC de NP1 de NP2" (Relative Clause of Noun1 of Noun2). These sentences were manipulated in terms of the phonological length (short vs. long) of the relative clause (RC), NP1, and NP2, resulting in 8 different sentence types. Additionally, 16 "filler" sentences were created with unambiguous interpretations to serve as control conditions and balance processing biases. The stimuli for the perception experiment were generated by a trained phonetician based on these target sentences.
Key Insights and Ideas
- RC Length Matters: The length of the relative clause significantly impacts how listeners resolve ambiguity in silent reading. Shorter RCs lead to a stronger preference for low attachment (the RC modifying the closer noun), while longer RCs result in a more balanced preference.
- Prosody Reflects Meaning: Prosodic grouping, particularly the placement of pauses and prominence, can effectively signal different attachment readings. This suggests that prosody isn't just a byproduct of speech but actively contributes to meaning.
- Conscious Control for Contrast: While default prosody might not always perfectly align with a specific attachment preference, speakers can consciously manipulate prosodic cues (like pauses and stress) to clearly signal desired meanings when forced to contrast them.
- Pause and Prominence as Key Cues: Both pause placement (early vs. late boundaries) and prominence (which noun is stressed) are crucial prosodic cues for disambiguation. An early pause or NP1 prominence tends to signal high attachment, while a late pause or NP2 prominence signals low attachment.
- Cue Conflict Leads to Confusion: When prosodic cues conflict (e.g., an early pause but NP2 prominence), listeners can become confused, highlighting the importance of cue consistency.
- Prominence: An Underappreciated Cue in Mandarin: For Mandarin, where tonal marking for small prosodic phrases is limited, prominence emerges as a particularly important prosodic grouping cue, comparable to pauses, for disambiguating structures.
Key Results
- Preference Survey: Native Mandarin speakers showed a general preference for low attachment (around 70% favoring NP1 in the RC NP1 NP2 structure). However, this preference was significantly modulated by RC length: short RCs strongly favored low attachment, while long RCs led to a more balanced preference.
- Production Experiments:
- Default Reading: While not all sentences produced in default reading showed clear sub-grouping, those that did demonstrated a correlation between prosodic structure and attachment resolution (late boundaries for low attachment, early boundaries for high attachment).
- Contrast Reading: Speakers consistently used distinct prosodic groupings (defined by pause and prominence) to signal high vs. low attachment when explicitly asked to contrast meanings.
- Perception Experiment: The experiment showed that manipulated prosodic cues could effectively guide listeners' interpretations. For instance, an early pause with NP1 prominence led to a high attachment preference, while a late pause with NP2 prominence led to a low attachment preference. When cues were compatible, agreement on preference was high; when they conflicted, listeners were confused.
Limitations and Future Work
- Implicit Prosody in Default Reading: The study found that the alignment between default prosody and attachment preference wasn't always perfect. Sub-grouping was optional, suggesting that while speakers might be aware of prosodic grouping, it's not always obligatory in spontaneous speech.
- Prominence Definition: The authors acknowledge that "prominence" is a broad term, defined here as loudness, often cued by pitch range and intensity. Further research could explore more nuanced definitions and the interplay of different prominence cues.
- Future Directions: The findings suggest that prominence is a vital prosodic cue in Mandarin, especially given the language's tonal system. Future work could delve deeper into the universal aspects of prosodic grouping and its specific realization in Mandarin, potentially benefiting spoken language processing models. The study also opens avenues for exploring how these prosodic cues interact with semantic and syntactic information in real-time language processing.
Summary updated