Constraining the search space in cross-situational learning: Different models make different predictions

TitleConstraining the search space in cross-situational learning: Different models make different predictions
Publication TypeTalks
AuthorsCassani, G., Grimm R., Gillis S., & Daelemans W.
Place PresentedPsycholinguistics in Flanders 2016, Antwerp
Year of Publication2016
Date Presented27 May
Abstract

In this work we compare the predictions of four different computational models of cross-situational learning to the behavior of both children and adults in a word learning task, using the data from Ramscar, Dye, and Klein (2013).
The computational models we evaluate include two neural networks with no hidden layer: one uses Hebb rule (Hebb, 1949) and the other uses the Rescorla-Wagner rule (Rescorla & Wagner, 1972) to update input-output connections. The second model is also known as Naïve Discriminative Learning (NDL, (Baayen, Milin, Durdević, Hendrix, & Marelli, 2011)). The third model is a probabilistic learner (Fazly, Alishahi, & Stevenson, 2010) that learns a probability distribution over output units for each input node. The last model is a Hypothesis Testing Model (HTM, (Trueswell, Medina, Hafri, & Gleitman, 2013)) that forms one single referent-word hypothesis for every trial and updates the hypothesized word for a referent in the light of new evidence.
Our simulations show that both the Hebbian learner and the HTM fail to match behavioral data from both children and adults: unlike both groups, these two models cannot learn reliable mappings between objects and words in the experimental paradigm used by Ramscar et al. (2013). Their psycholinguistic plausibility as models of cross-situational learning is thus challenged. On the contrary, the NDL model and the probabilistic learner match behavioral data. Therefore, their implementations can provide useful insights into how children discover how to map referents and words during language learning.

Baayen, R. H., Milin, P., Durdević, D. F., Hendrix, P., & Marelli, M. (2011). An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118(3), 438-481. doi:10.1037/a0023851
Fazly, A., Alishahi, A., & Stevenson, S. (2010). A probabilistic computational model of cross-situational word learning. Cognitive Science, 34(6), 1017-1063. doi:10.1111/j.1551-6709.2010.01104.x
Hebb, D. O. (1949). The organization of behavior New York, NY: John Wiley and Sons.
Ramscar, M., Dye, M., & Klein, J. (2013). Children value informativity over logic in word learning. Psychological Science, 24(6), 1017-1023. doi:10.1177/0956797612460691
Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical Conditioning II: Current Research And Theory (pp. 497). New York, NY: Appleton-Century-Crofts.
Trueswell, J. C., Medina, T. N., Hafri, A., & Gleitman, L. R. (2013). Propose but verify: Fast mapping meets cross-situational word learning. Cognitive Psychology, 66(1), 126-156. doi:10.1016/j.cogpsych.2012.10.001

KeywordsCross-situational learning; Discriminative Learning; Word learning; Model comparison; Computational Psycholinguistics
AttachmentSize
pif2016-presentation.pdf1.4 MB
PDF