1 / 1

Background

Background. Experiments 1 and 2: Title?. Experiment 1.

lynna
Download Presentation

Background

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. Background Experiments 1 and 2: Title? Experiment 1 Several recent studies have reported that adults successfully pair nonce sounds (“words”) and pictures (“referents”) over a small set of referentially ambiguous trials [e.g., 1-3]. Success is argued to be the result of learners keeping track of multiple possible word-referent pairings, using a statistical learning mechanism to gradually converge on a single hypothesized mapping [4-6]. In contrast, we [7] have argued for a learning procedure in which only a single hypothesized word-referent pairing (but no alternative hypotheses) is retained across learning instances, given up only when a subsequent instance fails to confirm the pairing – more like a ‘fast mapping’ procedure [8] than a statistical one [1-5]. Procedure: Participants clicked on one object to indicate which one they thought the word referred to. [N=60] • Appearance of a learning curve is misleading: • Increasing rate of correct responses is due to more and more correct responding occurring by chance alone. • Incorrect guesses are not followed by correct guesses at rates higher than chance, even if the previous trial had been a LI learning instance. • Accurate guesses are simply more likely on HI trials, and correct guesses are then followed by further correct responses. • No learning on HI trials: Accuracy on HI when presented last does not differ from chance, and is no different than when a HI is presented first ( = 0.001527, S.E. = 0.075599, z = 0.02, p = 0.984). • Slight improvement in HI Absent: Accuracy improves across LI trials ( = 0.23036, S.E. = 0.07174, z = 3.211, p = 0.00132). Propose but Verify: Fast Mapping meets Cross-Situational LearningMedina, Tamara N., Hafri, Alon, Trueswell, John, & Gleitman, Lila R. Effect of accuracy on the first learning instance Effect of accuracy on the third learning instance Chance for HI trials Stimuli Chance for LI trials Experiments 1 and 2 Chance for LI trials Chance for LI trials Photographs of objects, body parts, and animals isolated on a white background. Best fitting model includes effect of accuracy on the first learning instance, but not an effect of learning condition: ( = 1.0607, S.E. = 0.1912, z = 5.548, p = 0.0000000289). Best fitting model includes effect of accuracy on the first learning instance, but not an effect of learning condition: ( = 1.2146, S.E. = 0.2166, z = 5.607, p = 0.0000000205). Experiment 2 Procedure: Participants passively viewed stimuli and only began overt responding after the third learning instance. [N=61] Results: Overt responses (clicks) on learning instances 4 and 5 showed the same pattern as that of Exp. 1, indicating that overt responding did not induce the learning strategy observed in Experiment 1. Experiment 3: Title? Discussion + • Conclusion: These findings indicate that the learning mechanism for words involves insightful locking in to a word’s meaning followed by a confirmation procedure, rather than the accrual of multiple alternative hypotheses that are gradually pruned via statistical inference. • Propose but Verify: This learning procedure is similar to ‘fast mapping’ or ‘one trial learning’, with two crucial differences: • A hypothesis must find confirming evidence in the next learning instance; otherwise a new hypothesis will be considered. This is based on results in our previous studies [7]. • Memory is far from perfect. In the present study, the observed drop in performance after being correct cannot be explained as being from a lack of confirmatory evidence. Participants frequently forget hypotheses, especially when learning multiple words simultaneously. In a previous experiment using naturalistic stimuli, we found that accuracy on the HI learning instances was actually lower the later the HI instance was encountered [7] (see blue bars in figure below). “Oh look! A smick!” Procedure: Participants wrote a one-word guess indicated what they thought the nonse word referred to. A two-day delay was imposed between learning instances. [N=32] First DayTwo Days Later HI Middle: LI-LI HI-LI-LI HI Last: LI-LI-LI-LI HI Experiment 3 40-second clips of muted videotaped interactions between parents and 12-15 month old children Design * • 12 nonce words (5 times each) • 12 categories of referents (5 pictures of each) • Referential uncertainty on each trial: • High Informative (HI): 2 objects, or high baseline accuracy • Low Informative (LI): 5 objects, or low baseline accuracy • Four learning conditions (between-Ss): • HI First: HI-LI-LI-LI-LI • HI Middle: LI-LI-HI-LI-LI • HI Last: LI-LI-LI-LI-HI • HI Absent: LI-LI-LI-LI-LI Higher accuracy on HI learning instances following a delay ( = 1.3611, S.E. = 0.4892, z = 2.782, p = 0.005399). References and Acknowledgments Yu, C. & Smith, L.B. 2007. Psychological Science, 18(5), 414-420. Vouloumanos, A. 2008. Cognition, 107, 729-742. Ichinco, D., Frank, M.C., & Saxe, R. 2009. Proceedings of the 31st Annual Meeting of the Cognitive Science Society. Frank, M. C., Goodman, N. D., & Tenenbaum, J. (2009). Psychological Science, 20, 579-585. Yu, C., Smith, L.B., Klein, K.A., & Shiffrin, R.M. 2007. In D.S. McNamara & J.G. Trafton (Eds.), Proceedings of the 29th Annual Conference of the Science Society (pp. 737-742). Austin, TX: Cognitive Science Society. Smith, A.D.M., Smith, K., & Blythe, R.A. 2009. Boston University Conference on Language Development. Medina, T.N., Trueswell, J., Snedeker, J., & Gleitman, L. 2008. Boston University Conference on Language Development. Carey, S. & Barlett, E. 1978. Papers and Reports on Child Language Development, 15, 17-29. Research in our lab is currently supported by the National Institutes of Health (1-RO1-HD 37507; PIs: Trueswell and Gleitman).

More Related