Efficient Coding and Grice's Objection

This weeks readings were on experiments that investigated processes underlying how we understand implicature. The first experiment (Barner 2011) has children evaluate truth statements involving sets of scalar words (some, all; two, three) in order to understand why children fail to understand some types of scalar implicature but not others. They were interested in testing the hypothesis that children’s had trouble with scalar implicature because they couldn't generate relevant alternatives for specific scales. They tested this by using using situations in which generating the relevant alternatives was easy because it was apparent given the context (i.e. ‘contextual alternatives’) vs. comparatively difficult, more general cases (requiring the generation of words like ‘all’).

The second paper presents three experiments with separate but related aims. Stiller, Goodman and Frank investigate the origins of scalar implicature by using an task in which the participants (3-4 year olds and adults) identify an object based on a description that invokes understanding of either scalar or non-scalar implicature. The first experiment demonstrated that in cases of clear scale, both children and adults a capable of understanding scalar implicature (i.e. when asked to identify the object with ‘some' of the possible features, they mostly chose the object with some but not all features). In the second experiment, where no scale was present, they found that when asked to identify an object with a particular feature, adults did not abide by Grice’s prediction of providing the most informative answer given the situation. This result, however, was captured by the 'linguistic alternatives theory’. In order to test the validity of this theory (as well as evaluate implicature in terms of information transmission), they performed the third experiment. In this experiment, prior to the identification task, the participants were shown a sample space of objects with a skewed distribution of features, and found that during the identification task, participants were more likely to identify an object if the identifier for that object was higher entropy (i.e. occurred less frequently in the training phase).

From both experiments, we can conclude that implicature is very context dependent. In the first experiment, we see that in cases where the context is clearly defined, children are more adept at identifying relevant alternatives. In the second experiment, we learn that the manner in which adults understand implicature is related to maintaining high entropy in communication, which is determined by the underlying distribution of the features evoked in the implicature.

This result, that implicature is understood in a way that maximizes the information transfer rate, is interesting to me for many reasons. First, computational neuroscientists cast much of their research in terms of the 'efficient coding hypothesis’, which states that sensory systems encode sensory information in the most efficient (i.e. fewest bits) way possible. The findings of the third paper seem consistent with this hypothesis, though at level abstracted from neurons/rate coding.However, I feel that the authors are too quick to dismiss the Gricean account for the results in experiments 2 and 3. According to Grice, a cooperative conversationalist will seek to maximize the informativeness of their claim (by the maxim on quantity). However, this is highly dependent on the psychology of both the listener and the speaker. For example, the results of experiment 2 could be explained simply by asserting that the listeners were not attentive enough to the interpretation that would maximize information transfer, and thus were not able identify the most appropriate object. However, in the 3rd experiment, when the prior distribution of features is changed and thus the difference between the informativeness of claims become clearer, they are able to be a more competent Gricean agent. So while the result of experiment 2 shows that people are not ‘perfectly’ Gricean, it doesn’t rule out the possibility of them still being motivated by the Gricean maxims.