Learning and development in neural networks--the importance of prior experience

Infants can discriminate between familiar and unfamiliar grammatical patterns expressed in a vocabulary that is distinct from that used earlier during familiarization (Cognition 70(2) (1999) 109; Science 283 (1999) 77). Various models have captured the data, although each required that discrimination be distinct, in terms of the computational process, from familiarization. This article describes a simple recurrent network (SRN), equipped only with the assumption that it should predict what comes next, which models the data without distinguishing between familiarization and discrimination. To accomplish this, the SRN requires pre-training on a range of sequences instantiating different structures and different vocabulary items to those used subsequently during familiarization and test. Pre-training enables the network to avoid replacing structure acquired during familiarization with structure experienced at test. An equivalent enabling condition may underpin infants' resistance to catastrophic interference between the different structures and vocabulary items to which they are exposed.     

Forgetting to remember: the functional relationship of decay and interference

Functional decay theory proposes that decay and interference, historically viewed as competing accounts of forgetting, are instead functionally related. The theory posits that (a) when an attribute must be updated frequently in memory, its current value decays to prevent interference with later values, and (b) the decay rate adapts to the rate of memory updates. Behavioral predictions of the theory were tested in a task-switching paradigm in which memory for the current task had to be updated every few seconds, hundreds of times. Reaction times and error rates both increased gradually between updates, reflecting decay of memory for the current task. This performance decline was slower when updates were less frequent, reflecting a decrease in the decay rate following a decrease in the update rate. A candidate mechanism for controlled decay is proposed, the data are reconciled with practice effects, and implications for models of executive control are discussed.     

The old and thee, uh, new: disfluency and reference resolution

Most research on the rapid mental processes of on-line language processing has been limited to the study of idealized, fluent utterances. Yet speakers are often disfluent, for example, saying "thee, uh, candle" instead of "the candle." By monitoring listeners' eye movements to objects in a display, we demonstrated that the fluency of an article ("thee uh" vs. "the") affects how listeners interpret the following noun. With a fluent article, listeners were biased toward an object that had been mentioned previously, but with a disfluent article, they were biased toward an object that had not been mentioned. These biases were apparent as early as lexical information became available, showing that disfluency affects the basic processes of decoding linguistic input.     

Attentional capture of objects referred to by spoken language

Participants saw a small number of objects in a visual display and performed a visual detection or visual-discrimination task in the context of task-irrelevant spoken distractors. In each experiment, a visual cue was presented 400 ms after the onset of a spoken word. In experiments 1 and 2, the cue was an isoluminant color change and participants generated an eye movement to the target object. In experiment 1, responses were slower when the spoken word referred to the distractor object than when it referred to the target object. In experiment 2, responses were slower when the spoken word referred to a distractor object than when it referred to an object not in the display. In experiment 3, the cue was a small shift in location of the target object and participants indicated the direction of the shift. Responses were slowest when the word referred to the distractor object, faster when the word did not have a referent, and fastest when the word referred to the target object. Taken together, the results demonstrate that referents of spoken words capture attention.     

Orthographic directionality and thematic role illustration in English and Arabic

This study tested the hypotheses that people had a bias for drawing agents on the left of a picture when given a verb stimulus targeting an active or passive event (e.g., "kicked" or "is kicked") and that orthographic directionality would influence the way events were illustrated. Monolingual English speakers, who read and write left-to-right, and Arabic speakers, who read and write right-to-left, drew agents and patients in response to verb stimuli. We found no significant orthographic directionality effects and no preference for positioning agents on the left of pictures in either group or sentence type. Instead, participants drew agents on the right regardless of language or sentence type, and this was exaggerated in English speakers illustrating passive verbs. These findings support the existence of a preference for placing agents in the right hemispace that may result from asymmetrical hemispheric (i.e., left>right) activation induced by language processing. Our results are consistent with findings that people prefer pictures in which focus is on the right, a preference strongest in pictures with no implicit directionality of movement. This suggests that the methodology of the current study encouraged a static rather than dynamic interpretation of the verb in most participants.