1 / 22

Segmenting Nonsense

Segmenting Nonsense. Sanders, Newport & Neville (2002). Ricardo Tabone LIN 7912. Background. Behavioural studies  adults segment continuous speech using several segmentation cues Problem: these studies cannot distinguish between fast segmentation and slower linguistic processing

geona
Download Presentation

Segmenting Nonsense

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Segmenting Nonsense Sanders, Newport & Neville (2002) Ricardo Tabone LIN 7912

  2. Background • Behavioural studies  adults segment continuous speech using several segmentation cues • Problem: these studies cannot distinguish between fast segmentation and slower linguistic processing • Speech segmentation has been studied in different groups of speakers (e.g. young infants, bilingual adults, etc) through different tasks • There is a need for an experimental task that can be employed with all groups: Recording ERPs!

  3. Background (cont) • In continuous speech, initial syllables elicit larger negativity (N100) than medial syllables. (Sanders & Neville, in press) • Initial and medial syllables were controlled for loudness, length and other acoustic characteristics. • But…..

  4. Research Question • Do N100 word-onset effects index speech segmentation rather than acoustic characteristics pertaining to word boundaries? • In other words, is speech segmentation affected by lexical processing of speech sounds?

  5. Background (recap) • In continuous speech, initial syllables elicit larger negativity (N100) than medial syllables. • Initial and medial syllables were controlled for loudness, length and other acoustic characteristics. • But….. • Interesting: Behavioural tests  exposure to a continuous stream of nonsense words allows listeners to learn to distinguish between nonsense words and part-word items

  6. Experimental Design • Pre-test  Subjects listened to 36 pairs of 3-syllable Nonsense Words (NWs) and indicate which of the two items seemed more familiar. • Each pair consisted of one of the 6 NWs that would be used later and one part-word items composed of the last syllable of a NW + the first 2 syllables from another word • First Test  Subjects listened to a continuous stream of the 6 NWs (babupu, bupada, dutaba, patubi, pidabu, tutibu), repeated randomly 200 times each • The words were generated by text-to-speech synthesis and were sequenced without pauses: • Babupubupadababupudutabapatubipidabututibu

  7. Experimental Design • Pre-test  Subjects listened to 36 pairs of 3-syllable Nonsense Words (NWs) and indicate which of the two items seemed more familiar. • Each pair consisted of one of the 6 NWs that would be used later and one part-word items composed of the last syllable of a NW + the first 2 syllables from another word • First Test  Subjects listened to a continuous stream of the 6 NWs (babupu, bupada, dutaba, patubi, pidabu, tutibu), repeated randomly 200 times each • The words were generated by text-to-speech synthesis and were sequenced without pauses: • Babupubupadababupudutabapatubipidabututibu

  8. Experimental Design (cont) • ERPs were recorded during the first 14-minute exposure. • Second Test Determined whether or not subjected had learned to recognize words due to exposure alone • Next, trainingSubjects listened to the 6 NWs, separated by 500ms, for 10 minutes, and separated by 100ms for an extra 10 minutes. • Speakers were asked repeated the word and were presented with the text version on the screen. • Third Test  Assessed which words subjects learned • Fouth Test  ERPs were recorded during an extra 14-minute exposure to the same babupubupadababu…. • Firth Test  Assessed which words subjects learned

  9. Experimental Design (cont) • ERPs were recorded during the first 14-minute exposure. • Second Test Determined whether or not subjected had learned to recognize words due to exposure alone • Next, trainingSubjects listened to the 6 NWs, separated by 500ms, for 10 minutes, and separated by 100ms for an extra 10 minutes. • Speakers were asked repeated the word and were presented with the text version on the screen. • Third Test  Assessed which words subjects learned • Fouth Test  ERPs were recorded during an extra 14-minute exposure to the same babupubupadababu…. • Firth Test  Assessed which words subjects learned

  10. Participants • 18 participants • Right-handed • Monolingual English speakers

  11. Results (Behavioural) • Second Test  Subjects performed at ~50% • exposure alone isn’t enough. • Third Test  After training, participants performed at 79.5%. Bingo! • Fifth Test  Performance was measured again after another 14-minute exposure. Nothing changed (79.2%) • No new words learned, no old words forgotten.

  12. Results (Behavioural) • High correlation between individual performance on post-training tests and the difference in N100 amplitude before and after training

  13. ERPs Analysis • Divided the 18 participants into two groups: • 9 High learners (M=55.1%)  (M=90.7%) • 9 Low learners (M=52.2%)  (M=67.9%) • High Learners showed a significant effect of training on N100 amplitude, specially on medial and midline electrodes

  14. High Learners

  15. ERPs Analysis • Divided the 18 participants into two groups: • 9 Low learners (M=55.1%)  (M=90.7%) • 9 High learners (M=52.2%)  (M=67.9%) • High Learners showed a significant effect of training on N100 amplitude, specially on medial and midline electrodes • Low Learners did not show a significant N100 effect

  16. Low Learners

  17. ERPs Analysis • Divided the 18 participants into two groups: • 9 Low learners (M=55.1%)  (M=90.7%) • 9 High learners (M=52.2%)  (M=67.9%) • High Learners showed a significant effect of training on N100 amplitude, specially on medial and midline electrodes • Low Learners did not show a significant N100 effect • All subjects displayed a N400 effect • It is possible that this effect might influence medial and final syllables, since syllables last between 100ms to 300ms

  18. High Learners

  19. Low Learners

  20. Discussion • The N100 effect is similar to the one observed in processing English (Sanders & Neville, in press) • Artificial Language learners (McCandliss, Posner & Givón, 1997) also display the same N400 effects while learning new words • Japanese bilinguals also displayed a N400 effect and no N100 effect while listening to English (Sanders & Neville, in press)

  21. Conclusion • The results indicate than N100 effects cannot be solely explained on the basis of acoustic differences between initial and medial sounds • Listeners who are better at segmenting speech show earlier segmentation effects • Word-onsets effects are similar even when segmentation cues are very different • (e.g. NWs vs Native words) • N100 represents an automatic process of segmentation, whereas N400 indicates a slower, lexically-orientated process of segmentation.

  22. THEEND

More Related