TL;DR
This study investigates how word predictability and morphological structure influence the reduction of schwa sounds at the beginning of Dutch words. Surprisingly, increased predictability with the following word actually led to less schwa reduction, suggesting a complex interplay between predictive processing and articulatory planning.
What Problem Does This Paper Address?
Casual speech is characterized by reduced pronunciation, where sounds can be shortened or even disappear. Understanding the factors that drive this reduction is crucial for building better speech production and comprehension models. While word predictability has been studied, its precise role, especially in relation to word-initial syllables and morphological structure, remains unclear. This paper specifically addresses how the likelihood of a word given its context (predictability) and whether its initial syllable is a prefix (morphological structure) affect the presence and duration of the schwa sound in Dutch word-initial syllables.
Methods
The researchers analyzed spontaneous Dutch speech data, focusing on the initial syllables ge-, be-, and ver-, which typically contain a schwa. They employed mixed-effects regression models to analyze the data, treating speaker and word as crossed random factors.
Key independent variables included:
- Predictability:
- Log word frequency (base 2).
- Mutual Information (MI) with the preceding word.
- Mutual Information (MI) with the following word.
- Morphological Structure:
- A binary factor indicating if the initial syllable was a prefix.
- The ratio of the frequency of the word's stem to the frequency of the word itself.
Control variables included speech rate, utterance finality, and the place of articulation of the consonant following the schwa. The study examined both the presence/absence of schwa and its duration (log-transformed).
Datasets
The study utilized data from the Ernestus Corpus of Spontaneous Dutch, comprising approximately 15 hours of speech from 20 native Dutch speakers. This corpus contains around 153,200 word tokens representing 9035 word types. Frequencies for statistical analysis were drawn from the Corpus Gesproken Nederlands (CGN). No new datasets were introduced.
Key Insights and Ideas
- Predictability Paradox: Contrary to expectations, higher predictability of a word given the following word was associated with less schwa reduction (i.e., schwa was more likely to be present). This suggests that strong predictive links might actually delay the production of the initial part of a word, leading to a clearer pronunciation.
- Morphological Influence on Schwa Duration: The prefix ge- showed longer schwa duration when it functioned as a prefix compared to when it was part of the stem. This aligns with findings in English where affixes can influence segment duration.
- Stem-Word Ratio Matters for Prefixes: For the prefix ge-, a lower stem-word ratio (meaning the word is more frequent relative to its stem) was associated with shorter schwa duration, particularly at lower speech rates. This suggests that highly frequent complex words might be retrieved as single units, leading to more reduction.
- Articulatory Constraints: The place of articulation of the following consonant also played a role. Schwa was less likely to be absent before mid-articulation consonants (alveolar/palatal), and its duration was influenced by the following consonant's place of articulation, especially for the prefix be-.
Key Results
- Schwa Presence: Schwa was more likely to be absent at faster speech rates and in non-final utterance positions. The initial syllable be- had a higher likelihood of schwa presence than ver- and ge-. Schwa was more likely to be absent if the following consonant was mid-articulation (alveolar/palatal), especially if the initial syllable was not a prefix. Crucially, schwa was more likely to be present when the MI with the following word was higher.
- Schwa Duration: Schwa tended to be longer at lower speech rates and in utterance-final words. For the prefix ge-, schwa was longer when it formed a prefix. For the prefix ge-, a lower stem-word ratio was associated with shorter schwa duration, especially at lower speech rates.
Limitations and Future Work
The authors acknowledge that the effect of the stem-word ratio on the prefixes be- and ver- was not definitively established due to limited token counts, suggesting further research is needed. They also note that the observed effects of predictability and morphology are more complex than previously assumed. The authors propose that a clearer understanding of these conditions requires combining corpus studies with controlled production experiments. The study's findings on predictability, particularly the counter-intuitive result regarding MI with the following word, warrant further investigation to confirm and elaborate on the proposed explanation involving delayed production.
Summary updated