In that field there is a famous problem, "Gavagai problem." When we meet a novel word and estimate its meaning, there is an infinite of logically possible meanings for the word. It can mean a basic level category of rabbit , the superordinate category of animal, a part of rabbit body, its color, the rabbit's individual name, a running rabbit, and so on. Nevertheless, children as well as adults can estimate the meaning of the words relatively well. Although children have little knowledge of the world, their learning is very fast, actually by only one presentation. Such fast mapping can't be explained by existing machine learning algorithms based on trials and errors. In addition, when children see a novel object named by a novel noun and estimate the noun's meaning, they have tendencies to generalize the noun based on similarities in particular features (e.g. shape similarity) to the named object. Why they can realize such a fast and biased mapping, even though there is an infinite number of logically possible feature to attend to?
To explain these phenomena, developmental psychologists have suggested "word learning biases." They explained that the biases narrow the infinite number of logically possible correspondences between a word and features, and that the biases enable children to estimate a word's meaning more accurately. They revealed existence of such biases by some psychological experiments. But their problem is that these biases are just phenomenological explanations and don't explain why the biases exist or how they are realized in human brain. So I aim to explain more detailed computational mechanisms of the biases.
To closely explain them, we use computational modeling and computer simulation. We model them in mathematically rigorous and detailed form, and verify by computer simulation whether the model behaves in the same manner as human children. In psychological study, researchers can quite clearly examine the phenomena that are derived from the faculties concerning word learning by psychological experiments. But it's difficult to verify more detailed mechanism of the faculty. Meanwhile, in computational modeling study, we don't have the skill of examining children's real phenomena. So, I think that it is our work to model the human cognitive faculty more concretely based on findings of other research areas, and to suggest unobserved behaviors of human by the model.