Accessing spoken words: the importance of word onsets

1. Accessing spoken words: the importance of word onsets William Marslen-Wilson and Pienie Zwitserlood Presented by: Qinghua Tang

2. Introduction • Natural speech differs from written language in its directionality in time. When we hear natural speech, we hear it along a time line. • Does this mean we access words in our mental lexicon along the time line as well?

3. Two different models • Cohort model by Marslen-Wilson, 1984 • Based on the acoustic-phonetic properties, an initial set of candidates are activated. For example, when [t] is heard, all the words that start with [t] will be activated. • Incoming input helps to eliminate candidates until one is chosen. • Emphasis on the beginnings of words • TRACE model by Elman and McClelland, 1986 • Three levels: features, phonemes, and words • Relevant ‘nodes’ are activated as the incoming speech is matched to relevant ‘nodes’ and will keep being activated until the successful access of the word. • A word is accessed when the activation reaches a certain level. • Emphasis on the overall goodness of fit between the complete stimulus and a given lexical representation. • E.g. bleasure can be identified as pleasure because of the overall activation level. • Directionality is not an explicit condition

4. Previous research • Earlier research has shown the effects of word-initial partial matches (e.g. Marslen-Wilson, et al., 1989). • Stimuli: word pairs starting with same segments. The average number of shared segments was 3.3. • For example, ‘kapitein’ (meaning captain in Dutch) and ‘kapitaal’ (meaning capital in Dutch) • Procedure: in the experiment, subjects heard prime words like ‘kapitein’ and ‘kapitaal’ and saw a word that was either related to ‘kapitein’ (e.g ‘BOOT’, meaning ship ) or ‘kapitaal’ (e.g ‘GELD’, meaning money). The subjects were instructed to press a button to make lexical decisions to the word they see. • Results: when the probe word (visual word) was presented during the [t] of ‘kapitein’, both the word related to ‘kapitein’ and the word related to ‘kapitaal’ were facilitated.

5. Previous research shows that partial match can activate lexical representations, but, • Will partial match activate lexical representations if the partial match starts late in the word rather than the beginning of the word?

6. Research questions • Is there a strong temporal directionality in lexical access? • Does the on-line decision process tolerate later entry into the decision space of candidates that mismatch early in the word?

7. Experiment design - subjects 60 native speaker of Dutch. The subjects were randomly assigned to one of the experimental versions, six subjects per version.

8. Experiment design:5 experimental conditions: • Original-word condition: a word and its semantic associate (e.g. honing, meaning honey, and bij, meaning bee): • there was a complete match between acoustic stimulus and lexical element • a baseline • Real-word rhyme condition: the prime word rhymes with the prime word in the original-word condition (e.g. woning, meaning dwelling, and bij, meaning bee): • A partial match between woning and honing • In this case, woning should obviously be activated. Will honing, which rhymes with it, also be activated? If so, the access of bij should be facilitated.

9. Nonword rhyme condition: the prime word was a non-word that rhymed with the the prime word in the original-word condition (e.g foning and bij) • A partial match condition • This condition was included to address one concern from the TRACE model: once a word like woning was heard, it might inhibit the activation of its competitors such as honing. A non-word was therefore needed. • Two control conditions were used: real-word control condition (e.g. pakket and bij) and nonword control condition (e.g. dakket and bij) • In order to match with both real word and nonword conditions

10. Experiment design - materials • 50 sets of rhyming pairs (such as honing / woning) and their associated visual probes. • Each of the 50 sets was assigned to a different version of the experiment. For example, in the new version, woning would become the original prime, a new word HUIS (meaning house) would be used as the probe. Honing would become the real-word prime in this case. • 150 sets of filler pairs were added, balanced for word / nonword types. The lengths of the words were matched to real test targets.

11. Experiment design - procedure • The subjects were instructed to listen carefully to the spoken materials ( the prime words) and decide, as quickly as possible, whether the string of letters that was presented visually after each spoken word was a real word or not.

12. Results:

13. Results: • Only the original-word condition, in which there was a complete match between the input and the word associated with the visual probe, produced significant priming effects (32 ms). • In neither of the two partial match conditions was there any significant facilitation (11 ms for real-word rhyme condition, 4 ms for nonword rhyme condition). • The absolute reaction times for the two partial match conditions were almost identical (547 vs. 548 ms). • However, rhyme primes did have some effect on the lexical representations of items with which they rhyme, even if this produceed only small facilitation.

14. Discussion – amount of segmental overlap • Could the failure of partial match in producing significant facilitation be caused by lack of sufficient overlap between real-word primes and the original words? • Stimuli were divided into four categories, depending on the number of shared segments which ranges from 3 to 6 or more) • The number of items in each overlap group: • 16 items in 3 segment group • 48 items in 4 segment group • 12 items in 5 segment group • 16 items in 6 or more segment group • The average segment overlap for previous experiments was between 2.9 and 3.3.

15. Figure 2: segment overlap effects

16. There was a significant effect of prime type, but no effect of length. • The degree of activation of a lexical representation seemed to be independent of the amount of overlapping segments. • These results make it unlikely that the failure of rhyme primes to produce facilitation was because they did not provide amounts of matching input comparable to the word initial primes in the previous experiment.

17. Discussion – the competitor environment • Does the number of competitors matter? • In current models of lexical access, the number of competitors that a prime word has to contend with plays an important role in determining the response of the system. • the current stimulus sets varied in the number of rhyme competitors from 1 to 23. • 30 sets had only 1 rhyme competitor (I.e. the matched real-word rhyme). The other 70 sets had rhyme competitors ranging from 2 to 23. • the stimuli were divided into 3 groups: • 30 sets with 1 competitor, 30 sets with 2 to 4 competitors, and 30 sets with 6 to 23 competitors

18. Figure 3

19. Variations in the number of competitors did not affect the advantage of full primes over rhyme primes, further confirming the overall advantage of the directionality hypothesis. • On the other hand, the competitor environment did seem to matter. When the competition was low, the overall level of activation was higher for both the complete match and partial match conditions.

20. Conclusion: • The dominant effect in the present results is directionality of mapping. Primes that mismatched at the beginning of the word with the relevant lexical form representation were always less effective than primes that did not mismatch word initially. This suggests that word onsets do have a special status in spoken word recognition. • There was no detectable effect of the lexical status of the rhyme prime. The real-word rhyme and nonword rhyme prime stimuli behaved the same way throughout. • Partial match that starts late in the word rather than the beginning also shows facilitation effects and it interacts with the density of the competitor environment. More research is needed to account for this fact.

21. Thank you!