INFORMATION ACCRETION AND REDUCTION IN TEXT-PROCESSING - INFERENCES

Although it takes more than a well-chosen word to solve a problem, a p oorly chosen one can make it much harder to find a solution. I am sugg esting that the term inference has become such an obstacle and would l ike to propose an alternative framework that might prove more fruitful . The term inference is used in three different domains: as a word in everyday language, as a term in formal logic, and as a term in the cog nitive sciences. The first use is unproblematic, and the second is pre cise enough. After 20 years of research on discourse comprehension, ho wever, it is difficult to avoid the conclusion that something is wrong with the third use of inference. Not much progress has been made in o ur understanding of ''inferencing'' in discourse comprehension. More p recisely, many different analyses of inferences have been proposed, an d a rich set of relevant empirical results is now available, but we ar e as confused as ever. Almost 20 years ago, I set out to study which i nferences are made in comprehension, and when they are made, that is, during comprehension or at the time of testing (Kintsch, 1974). The sa me questions are still being asked today, and the answers that are off ered are highly controversial (e.g., McKoon & Ratcliff, 1992). Why do we see such lack of progress here when in other areas of discourse com prehension so much has been achieved? One possibility is that the orig inal questions were the wrong ones. Inferences are not a unitary categ ory. Hence, we cannot make statements about inferences in general, but rather must distinguish different types of inferences, Second, the qu estion of whether an inference is made during comprehension or at some later time is meaningful only for certain types of inferences. What I am proposing, therefore, is that we need a more differentiated view o f inferential processing than that implied by the global term inferenc e, one that furthermore will allow research on inferencing in discours e processing to develop independently of the logical-philosophical tra dition.