understanding computers and cognition: a new foundation for design

terry winograd and fernando flores

ablex corp, norwood NJ, 1985

winograd is one of the founders of the HCI lab at stanford and has worked on trying to make language understood to computers.

approach through design (the interaction between creating and understanding), and mostly about developing a theory of language

the computer is a device for creating, manipulating and transmitting symbolic (i.e. linguistic) objects.

“every questioning grows out of a tradition” -> what is the tradition of this thesis? (also, meaning is relative to what is understood through the tradition), the analytical tradition

chapter 5 - language, listening and commitment (pp.56-64)

combination of hermeneutics with speech act theory (meaningful acts in situations of shared activity)

mathematical language is, by definition, context-free.

Meaning is created by an active reading, in which the linguistic form triggers interpretation, rather than conveying information.

That which is not obvious is made manifest through language

austin has *five fundamental illocutionary points**:

assertive: says whether that is the case or not
directive: attempt to get the hearer to do sth (questions/commands)
commissive: commit the speaker to some future action
expressive: express a psychological state about X (apologizing, praising, forgiving)
declarative: brings about the correspondence between the propositional content of the speech act and reality

(each of these points can have different forms)

theoretical language (rational) vs. practical (situated)

Language and cognition are fundamentally social (p.61).

structural coupling (skills) with a consensual domain (piece of source code) is how an autopoeitic structure emerges (self-sustained attention? aesthetic experience?) ????

there are three basic kinds of grounding:

experiential (an observation to verify)
formal (a logical language-game to verify)
social (relying on what others said to verify)

chapter 6 - towards a new orientation (pp.70-80)

summarization of concerns

cognition as embeded/dasein (structural coupling)
cognition viewed not as activity in some mental realm, but as a pattern of behaviour that is relevant to the functioning to the person or organism in its world
disagrees with “knowledge is a storehouse of representations, which can be called upon for use in reasoning and which can be translated into language” -> the problem is that “things” do not have independent properties. heidegger: “things” emerge when they breakdown, when they become a matter of concern, when they become distinct (already skill?).
corresponding representation hypothesis: the assumption that cognition rests on the manipulation of symbolic representations that can be understood as referring to objects and properties in the world.

The essence of computation lies in the correspondence between the manipulation of formal tokens and the attributions of a meaning to those tokens as representing elements in worlds of some kind. (p.74)

pre-understanding and background are important for meaning-making (fluency). knowledge is always the result of interpretation, which in turn depends on the entire previous experience of the interpreter and on situadness in a tradition.

their approach to language is not that language is transmission of information, but rather that is a form of human social action, with mutual orientation (but i don’t see those as mutually exclusive)

chapter 7 - computers and representation

focusing on the fundamental issues of language and rationality that are the background for designing and programming computers.

7.1 - programming as representation

The first and most obvious point is that whenever someone writes a program, it is a program about something.

(called the subject domain; interesting how it switched to being called problem-domain in software engineering)

the programmer has a systematic correspondence between the symbols contained in storage cells and how they represent objects and relationships in the subject domain.

Success in programming depends on designing a representation and set of operations that are both veridical [produce results that are correct] and effective [they are usable].

most important parts of a formal system (e.g. a programming language):

there is a structure of formal symbols which can be manipulated according to precisely-defined and well-understood rules.
there is a mapping through which the relevant properties of the domain can be represented by symbol structures
there are operations that manipulate the symbols in such a way as to produce veridical results (derive new structures, which are still veridical)

the problem is that representation is in the mind of the beholder: in no way do programming languages depend on the fact that they represent something (at least, sth too concrete; because they do have pointers, registers, etc., which represent something tangible).

cf. Newell & Simon, 1976, “Computer Science as Empirical Enquiry” -> for the ability of computers to refer to themselves

7.2 levels of representation

computers have levels of representation (abstraction?)

physical machine (components and copper)
logical machine (boole, circuits, gates)
abstract machine (assembly, pointers, instruction sets)
higher-level languages (cobol, fortran, lisp) - this requires compilers or interpreters
a representation scheme for “facts” (which we would call data today), like linked open data, relational, etc.

the subject domain is always the next higher level itself (p. 89)

7.3 can computers do more than told?

some issues, though:

breakdown (some levels might not operate as anticipated)
resource use (some things described are only relevant at a lower level) -> depends on the implementation (about process more than result)
accidental representation: mismatch between machine realization and programmer intention (e.g. glitch, hacks) -> one could say that some structures of the machine do not represent the results it outputs (like hacking a display to render a heart on a TI calculator)