... his^1.1

Rather than bloat this book with the disjunction ``he or she'', I will uniformly refer to the generic speaker as ``she'' and the generic listener, or hearer, as ``he''. This should be easy to remember since ``hearer'' starts with ``he'' and ``speaker'' and ``she'' both start with ``s''. The occasional other generic characters in this book will be referred to as ``he or she''.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... defended.^1.2

Perhaps the best statement of the theoretical stance of cognitive science is that of Haugland (1981). Other good accounts can be found in Newell and Simon (1976), Dennett (1978), Pylyshyn (1981), and Winograd (1977).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... covered.^1.3

``Data'' is a singular English mass noun derived from a plural Latin count noun.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... data.^1.4

Or what Lakatos (1970) might call ```data''', conventionally agreed upon.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... interacts.^1.5

This is a straw man Putnam (1981) does battle with.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... level.^1.6

Variations on this view dispense with the symbolic or with the connectionist level.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... sometimes^1.7

Formerly more than today.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... index.^1.8

A more elaborate treatment of some of these interpretation reports and others can be found in Mann et al (1975).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... John.^1.9

Note however that we do not assume such reports are reliable across informants.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... power.^1.10

I think most AI practitioners share my bias. From the point of view of AI as technology, there is something perverse about seeking to imitate people's shortcomings, except where, as Black (1980) has urged, we view people's shortcomings as indicative of greater strengths. For example, a lack of facility with center embeddings may be a side-effect of the ability to parse in real time, and thus a clue as to how the latter is done.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... theory.^1.11

A further move that is available to us somewhere in this sequence is to challenge, or even dismiss, the interpretation report, thus assaulting what Lakatos calls the ``interpretive theory'', rather than the explanatory theory. But this is an action directed not at one's theory of the corpus, but at one's colleagues, who decide what the `data' is.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... on.^1.12

The confusion between inference as part of a conscious problem-solving effort and inference as a theoretical construct is common enough. It infects, for example, Zajonc's (1980) criticism of cognitive explanations in social psychology, and it even occurs, mystifyingly, in the latter parts of the otherwise excellent paper by Haugeland (1981).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... consideration.^1.13

Pylyshyn (1980) has written an excellent analysis of the role of reaction time data in cognitive science, with which this paragraph is not, I believe, inconsistent.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... knowledge.^1.14

This is usually referred to in AI as the ``representation of knowledge''. The downgrading implicit in my use of ``logical notation'' is deliberate.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.