1 / 27

Real-Time Dynamics of Language Acquisition in Two-Year-Old Children and Connectionist Models

Real-Time Dynamics of Language Acquisition in Two-Year-Old Children and Connectionist Models. Jessica S. Horst (jessica-horst@uiowa.edu) Larissa K. Samuelson (larissa-samuelson@uiowa.edu) Bob McMurray (bob-mcmurray@uiowa.edu) Dept. of Psychology University of Iowa. Word Learning.

malise
Download Presentation

Real-Time Dynamics of Language Acquisition in Two-Year-Old Children and Connectionist Models

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. Real-Time Dynamics of Language Acquisition in Two-Year-Old Children and Connectionist Models Jessica S. Horst (jessica-horst@uiowa.edu) Larissa K. Samuelson (larissa-samuelson@uiowa.edu) Bob McMurray (bob-mcmurray@uiowa.edu) Dept. of Psychology University of Iowa

  2. Word Learning • Children are amazing word learners: • By 5th birthday know 60,000 words • Estimated to learn 9 words/day from 18mo (Carey, 1978) • Big Question: What it takes to turn a novel name into a known name? “Teddy” “Puppy” “?” “Bunny” “Kitty” “Doll”

  3. Fast Mapping • Linking a novel name to a novel referent with minimal exposure (Carey & Bartlett, 1978) • Literature associates amazing word learning with fast mapping: • Children can fast map several names in a single session (Golinkoff et al., 1992) • Children can determine the referent of a novel name in less than 3 seconds (Halberda, submitted) • Big Picture: Understanding the processes of word learning, by determining exactly what children are learning about name-object mappings taught with minimal exposure

  4. Two Times Scales in Language Acquisition Fast Mapping and Word Learning represent two time scales of learning: Rabbit • Fast Mapping: quick process emerging in the moment • Based on the Principle of Lexical Contrast (Clark, 1987) “Rabbit”? “Rabbit”? “Rabbit”? “Puppy” ≠ “Rabbit” “Kitty” ≠ “Rabbit” This must be “Rabbit”

  5. My Picture Book Rabbit! • Word Learning: gradual process over the course of development • Evidenced by production or identification of the referent after a delay Next week, we’re going to see Sally’s Rabbit. She might let you pet the Rabbit. The Rabbit is very soft, but you must be very gentle with the Rabbit….

  6. Question: Is Lexical Contrast enough to truly learn a word, i.e., to show evidence of learning after a delay?

  7. General Methods • Fast mapping trials: • 2 familiar objects • 1 novel object Cow (familiar) Block (familiar) Yok (novel) Child is asked for both familiar (cow) and novel (fode) objects across trials Five-minute delay period Retention trials: • 1 target • 1 prev. named novel • 1 prev. unnamed novel Tannin (named foil) Yok (target) unnamed foil (previously seen) 3 warm-up trials with 3 familiar objects are presented before the FM trials. 1 additional warm-up trial is presented before the retention trials.

  8. Experiment 1: 1 *** p < .001 *** 0.8 *** 24-month-old children (N = 16) 2familiar & 1 novel objects 8 familiar and 8 novel trials (e.g., “get the cow!” or “get the yok!”) 0.6 Proportion of Correct Choices 0.4 Chance 0.2 0 Familiar Names Novel Names Retention Children were excellent at fast mapping (finding the referent of novel and familiar words in the moment). Children were unable to show evidence of retention* after a five-minute delay. *Note: only tested correctly fast mapped names for retention

  9. Did Children Learn Individual Words? Retention Data 1 0.8 0.6 0.4 Chance Proportion of Correct Choices 0.2 0 Names 3 & 4 Names 5 & 6 Names 7 & 8 Names 1 & 2 • What if children may retain 1, but not as many as 8 names? • names were analyzed by order of presentation during fast mapping • none of the positions above chance levels • Children unable to retain mappings after a 5-minute delay

  10. Experiments 2 and 3 • Initial findings replicated with simpler tasks: • effect of number of names or trials? • Children’s difficulty in retaining newly fast-mapped names is not related to the number of names or trials Replication #1 (E2) (N = 16) Replication #2 (E3) (N = 16) • 1 Novel Name • 8 Familiar Names • 7 Preference Trials • 1 Novel Name • 2 Familiar Names Expected by chance: 5.33 Expected by chance: 3.33 Expected by chance: 5.33 Expected by chance: 2.67 * Binomial, p < .05, † Binomial, p = .12

  11. Intermediate State During Learning The Model Decision Units • 15 Auditory & 15 Visual units: Activate according to what child sees and hears • 90 Decision units • Names presented singly with a variable number of objects Visual Units Auditory Units Decision Units • Name-Decision & Object-Decision associations strengthened via learning • After 4000 training trials network forms localist representations • Learns name-object links and to ignore visual competitors Visual Units Auditory Units End State Post Learning

  12. Processing In The Model “fork” “spoon” “cup” “plate” “knife” “napkin” • Activation feeds from input layers to decision layers. • Decision units compete via inhibition. • Activation feeds back to input layers. • Cycle continues until system settles. Auditory Inputs Decision Units (Hidden) Layer Visual Inputs (McMurray & Spivey, 2000) • Unsupervised Hebbian learning occurs on every cycle.

  13. 1 0.9 0.8 0.7 0.6 Activation 0.5 0.4 0.3 0.2 0.1 0 0 2 4 6 8 10 12 14 16 18 Cycles • Online decision dynamics reflect auditory and visual competitors.

  14. Decision Units Decision Units 10 10 20 20 30 30 40 40 50 50 60 60 70 70 80 80 90 90 10 10 9 9 8 8 7 7 6 6 Auditory Input Auditory Input 5 5 0.2 4 4 3 3 2 2 0.15 1 1 0.1 0.05 9 16 26 30 32 39 41 49 65 67 Connection Strength

  15. 20 networks initialized with random weights 15 word lexicon (names & objects): Familiarization with Initial Vocabulary: Familiarized with 5 familiar items for 5000 epochs Items presented in random order Fast Mapping Experiment: familiar novel held out cup ball “Yok” yok “Fode” fode fode ??? • 5 familiar names • 5 novel names • 5 held out • 5 retention trials • 10 fast mapping trials

  16. Fast Mapping In The Model *** 1 *** 0.8 0.6 Proportion of Correct Choices 0.4 Chance 0.2 0 Familiar Name Novel Name • Model succeeded on both types of fast-mapping trials • Model behavior patterned with empirical results • Learning was not turned of during fast mapping

  17. Retention In The Model 1 *** *** 0.8 0.6 Proportion of Correct Choices 0.4 Chance 0.2 0 Familiar Name Novel Name Retention • The model fails to “retain” the newly learned words after a “delay” • Learning was not turned of during retention

  18. Why Didn’t The Simulations Retain? 0.000005 0.000004 0.000003 Squared Deviations 0.000002 0.000001 0 Familiar Words Novel Words Control Words After Test 1 1 0.8 0.8 0.6 0.6 Activation Activation 0.4 0.4 0.2 0.2 0 0 0 5 10 15 20 0 5 10 15 20 Cycles (familiar words) Cycles (novel words) • Analyses of weight matrices revealed that relatively little learning occurred during fast mapping trials. Change (RMS) in portions of weight matrix 2 1.6 1.2 Squared Deviations 0.8 0.4 0 Familiar Familiar Novel Control Words Words Words Words After After Test End End Learning Temporal dynamics of processing

  19. 10 10 20 20 30 30 40 40 50 50 60 60 70 70 80 80 90 90 Prior to Experiment 0.2 0.15 Connection Strength 0.1 14 14 12 12 0.05 10 10 After Experiment 8 8 6 6 4 4 2 2 1 4 66 80 86

  20. Implications • Making the name-object mapping in the moment is not enough to form a long-term memory representation of the novel name • Lexical Contrast provided in 1 fast mapping trial not sufficient to turn a novel name into a known name • Goal for Experiment 4: • Add more support to the task to facilitate word learning: • Increase the number of naming instances (see Merriman & Marazita,1995) • Provide reinforcement • Provide ostensive naming (see Mervis & Bertrand, 1994)

  21. Methods Correct choice: Incorrect choice: • Two conditions: reinforced and ostensive definition • All children heard the names 5 times before each trial: • “Can you get the yok? Help me find the yok! Are you ready to find the yok? Can you help me get the yok? Let’s get the yok!” Reinforced Condition: “Yes, that is the yok” (Child holding) Ostensive Naming: “Look, this is the yok” (Exptr holding Target & pointing) Reinforced Condition: “That is the cow.” (Child holding) Ostensive Naming: “Look, this is the yok” (Exptr holding Target & pointing)

  22. Experiment 4: *** p < .001 1 *** Retention *** *** Ostensive Definition *** 0.8 0.6 Proportion of Correct Choices 0.4 chance 0.2 0 Familiar Names Novel Names Retention • Again, children were excellent at fast mapping • Overall, however, they were very poor at retaining* • No effects of Condition were found for fast mapping • * Note: only tested correctly fast mapped names for retention

  23. 1 Retention * Ostensive Definition 0.8 * 0.6 Proportion of Correct Choices 0.4 0.2 0 Block 1 Block 2 Block 3 Block 4 * p < .05 • When analyzed by block, it is clear that children retained names in the Ostensive Definition Condition • Data suggests children can learn up to 4 names in this task • Analysis of looking indicated that children attended to novel object twice as much in OD condition

  24. Conclusions • Overall, children were excellent at finding the referent in the moment, but unable to retain the names over a five-minute delay (E1) • Follow-up experiments indicate poor retention not due to • The number of names introduced (E2) or • The number of trials in the session (E3) • The Connectionist Network captured the data and • Showed the same pattern of results: • excellent Fast Mapping, poor Retention • Learning was occurring during fast mapping, but not enough learning to support later evidence of retention

  25. Together, Experiments 1 - 3 and Simulations suggest that Lexical Contrast alone is not enough to allow children to form a strong enough representation of a novel name to show evidence of word learning after five minutes • Children areable to retain words taught in a Fast Mapping Task if: • Provided with multiple naming instances • And ostensive definitions (E4) • But only able to learn up to 4 words • Future research will explore the role of attention in helping children turn novel names into known names

  26. Take Home Message: Fast Mapping is a quick, online mechanism that can produce smart online behavior but not actual word learning.

  27. References Carey, S. (1978). The child as word learner. In M. Halle, J. Bresnan & A. Miller (Eds.), Linguistic Theory and Psychological Reality (pp. 264-293). Cambridge, MA: MIT Press. Carey, S., & Bartlett, E. (1978). Acquiring a single new word. Proceedings of the Stanford Child Language Conference, 15(17-29). Clark, E. (1987). The Principle of Contrast: A Constraint on Language Acquisition. In B. McWhinney (Ed.), Mechanisms of Language Acquisition (pp. 1-33). Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Golinkoff, R. M., Hirshpasek, K., Bailey, L. M., & Wenger, N. R. (1992). Young-Children and Adults Use Lexical Principles to Learn New Nouns. Developmental Psychology, 28(1), 99-108. Halberda, J. & Goldman, J. (submitted). One Trial Learning in 2-Year-Olds: Children Learn New Nouns in 3 Seconds Flat. Mervis, C. B., & Bertrand, J. (1994). Acquisition of the Novel Name Nameless Category (N3c) Principle. Child Development, 65(6), 1646-1662 Merriman, W. E. & Marazita, J. M. (1995). The Effect of Hearing Similar Sounding Words on Young 2-Year-Olds’ Disambiguation of Novel Reference. Developmental Psychology, 31(6), 973-984. McMurray, B., & Spivey, M. (2000). The Categorical Perception of Consonants: The Interaction of Learning and Processing, The Proceedings of the Chicago Linguistics Society, 34(2), 205-220. Acknowledgements The authors would like to thank Joseph Toscano for programming assistance and support. This work was supported by NICHD Grant R01-HD045713 to LKS.

More Related