Speech Perception by NonNative Speakers Declines Drastically in
Speech Perception by Non-Native Speakers Declines Drastically in Noisy Conditions Catherine Caldwell-Harris, Inna Ryvkin, Andrei Anghelescu, Loraine K. Obler Catherine Caldwell-Harris Boston University 1
Expert language processing system cleans up noisy input -- Wow! Highly fluent speakers • Fill in from meaning • Are more sensitive to auditory cues (Barbara Shinn-Cunninghams’ talk) Catherine Caldwell-Harris Boston University 2
We already know a lot about factors influencing non-native speech perception ‘Fluency/ability’ matters; and these factors influence fluency (in order of strength): • Proficiency • Number of years of language use • Amount and type of current use • Age of acquisition (often determines above factors) Task and input factors • Noise / environment • Sentence context (predictability of words) Catherine Caldwell-Harris Boston University 3
But much less is known about: How factors that influence skill level interact • Is early acquisition important even controlling for proficiency and current use? Does word frequency matter? How do input factors interact? • Examine word freq in high/low predictable context Similarity between first and second language • Those great Dutch learners of English! Catherine Caldwell-Harris Boston University 4
Speech Perception In Noise Test (Bilger, 1984) Participants were instructed to repeat the last word sentence. 80 items selected from the original SPIN sentences. Three variables, fully crossed: Ø predictability of target (Predictable/Unpredictable) Ø lexical frequency of target (High/Low) Ø noise (Noise/Clear). of each The noise consisted of multi-speaker babble. Ø Signal-to-noise ratio was -2 d. B for sentence-plus-noise items. Ø Each item is a sentence that was prerecorded by an adult male and presented in stereo through headphones. Clear, predictable Noise, predictable Catherine Caldwell-Harris Clear, unpredictable Noise unpredictable Boston University 5
Examples of conditions Predictable Context Target His plan meant taking a big risk. Hi Freq Tom fell down and got a bad bruise. Lo Freq Unpredictable Context Bill might discuss the foam. I was considering the crook. Catherine Caldwell-Harris Boston University Hi Freq Lo Freq 6
Questions in Current Study Similarity between first and second language • With L 1 Spanish, Russian, Mandarin When identifying targets in spoken English • Predict: Spanish > Russian > Mandarin High/low frequency of target words • Predict: frequency effects will depend on proficiency Age of acquisition • Predict: Doesn’t matter, after controlling for proficiency? Catherine Caldwell-Harris Boston University 7
Procedure Ø Questionnaire § Language History § Self-rating of proficiency in L 1 and English § Adult English Language Use Scale (AELU) Ø Hearing Threshold determination § SPIN meant to be administered 50 d. B above hearing threshold. § Threshold tested via a Behringer UB 502 Eurorack 5 Input Mixer Catherine Caldwell-Harris Boston University 8
Learning History Variables Catherine Caldwell-Harris Boston University 9
Catherine Caldwell-Harris Boston University 10
Catherine Caldwell-Harris Boston University 11
Catherine Caldwell-Harris Boston University 12
Overview of Results v Native English-speakers performed nearly at ceiling, Mandarin speakers performed most poorly, and Spanish- and Russian-speakers show in intermediate results. v Noise, predictability, frequency matter for all speakers v BUT: Lexical frequency of target is a small effect of native English speakers Catherine Caldwell-Harris Boston University 13
Effect sizes for main effects and interactions Catherine Caldwell-Harris Boston University 14
Catherine Caldwell-Harris Boston University 15
Predictable Hi Freq Unpredictable Hi Freq Predictable Lo Freq Unpredictable Lo Freq Catherine Caldwell-Harris Boston University 16
Spanish L 1 Catherine Caldwell-Harris Boston University 17
Russian L 1 Catherine Caldwell-Harris Boston University 18
Mandarin L 1 Catherine Caldwell-Harris Boston University 19
Need more participants to determine cline of Spanish > Russian > Mandarin Statistically, group differences greatly diminished when ages-of-arrival are taken into account. Age-of-arrival effects very strong. Why? Age-of-arrival may organize immigrants’ language learning environment (Caldwell-Harris et al, under review). Still, plots for language groups show steeper slope for decline in target word detection with age-of-arrival for Mandarin speakers. But Mandarin group had on average later age-ofarrival. Catherine Caldwell-Harris Boston University 20
Learning History Variables Catherine Caldwell-Harris Boston University 21
Answers in Current Study Similarity between first and second language • Comparing L 1 Spanish, Russian, Mandarin • Confirmed: Spanish > Russian > Mandarin • But: age-of-arrival confound High/low frequency of target words • Confirmed: frequency effects depend on proficiency • Implies ‘subject/experienced’ proficiency is what matters for processing. For high proficiency speakers, need very low freq words? Is log transform sufficient? Catherine Caldwell-Harris Boston University 22
Second language learning ideal for neural net modeling Use modeling to clarify proposals about main effects and interactions of: • The role of frequency (why frequency effects disappear for high proficiency learners) • Predictability -- test hypotheses about when lower proficiency learners can use context to make predictions • Vary L 1 -L 2 similarity in simulations • Vary “age of exposure” and intensity of language contact Catherine Caldwell-Harris Boston University 23
- Slides: 23