Cognitive Models Material from Authors of Human Computer Interaction Alan Dix, et al

Overview Cognitive models represent users of interactive systems hierarchical - user’s task and goal structure linguistic – user-system grammar physical and device – human motor skills architectural – underlie all of above

Cognitive models They model aspects of user as they interact: understanding knowledge intentions processing Common categorization: Competence – represent kinds of behavior expected of user Performance – allow analysis of routine behavior in limited applications

Goal and task hierarchies Solve goals by solving subgoals - Mental processing as “divide-and-conquer” produce report gather data. find book names.. do keywords search of names database …further sub-goals.. sift through names and abstracts by hand …further sub-goals. search sales database..further sub-goals layout tables and histograms..further sub-goals write description..further sub-goals

Issues for goal hierarchies Granularity Where do we start? Where do we stop – how far to subdivide? Get down to a routine learned behavior, not problem solving - the unit task Conflict More than one way to achieve a goal Treatment of error

Techniques Goals, Operators, Methods and Selection (GOMS) Cognitive Complexity Theory (CCT) can represent error behavior

GOMS Goals - what the user wants to achieve Operators- basic actions user performs (granularity) Methods - decomposition of a goal into sub goals/operators may be more than one way or method to do that Selection - means of choosing between competing methods (GOMS attempts to predict)

GOMS example GOAL: ICONIZE-WINDOW [select GOAL: USE-CLOSE-METHOD MOVE-MOUSE-TO-WINDOW-HEADER POP-UP-MENU CLICK-OVER-CLOSE-OPTION GOAL: USE-L7-METHOD PRESS-L7-KEY] For a particular user Sam: Rule 1: Select USE-CLOSE-METHOD unless another rule applies. Rule 2: If the application is GAME, select L7-METHOD.

GOMS as a measure of performance selection rules can be tested for accuracy against user traces stacking depth of goal structure can estimate STM requirements good for describing how experts perform routine tasks not for comparing across tasks not for predicting training time

Cognitive Complexity Theory - CCT - basic premises of goal decomposition - provides more predictive power Two parallel descriptions: User - production rules of the form: if condition then action Device - generalized transition networks covered under dialogue models

Example: editing with vi Production rules are in long-term memory - 4 rules in the text on page 425 User sees a mistake - Model contents of working memory as attribute-value mapping (GOAL perform unit task (TEXT task is insert space) (TEXT task is at 5 23) (CURSOR 8 7)

Example: editing with vi Rules are pattern-matched to working memory, e.g., LOOK-TEXT task is at %LINE %COLUMN is true, with LINE = 5 COLUMN = 23. Four rules model inserting a space –1 st one only one that can fire: SELECT-INSERT-SPACE //bind to location INSERT-SPACE-DONE //finished - unbind INSERT-SPACE-1 //move cursor INSERT-SPACE-2 //hit insert key and space

Example: editing with vi When fired, binds the LINE and COL to 5 and 23 respectively and adds to working memory (GOAL insert space) (NOTE executing insert space) (LINE 5) (COLUMN 23) Now INSERT-SPACE-1 will fire

Notes on CCT Rules don’t fire in order written, may repeat Parallel model – rules can fire simultaneously Novice versus expert style rules Error behavior can be represented Measures Depth of goal structure Number of rules (more means interface more difficult to learn) Comparison with device description

Problems with goal hierarchies description can be enormous a post hoc technique – risk is that it is defined by the computer dialog and not user expert versus novice Simple extensions possible goal closure (makes sure subgoal satisfied) eg. ATM example

Linguistic notations User’s interaction with a computer is often viewed in terms of a language. Backus-Naur Form (BNF) Task-Action Grammar (TAG)

BNF Very common notation from computer science A purely syntactic view of the dialogue Basic syntax: nonterminal ::= expression An expression contains terminals and nonterminals combined in sequence (+) or as alternatives (|). Terminals lowest level of user behavior CLICK-MOUSE, MOVE-MOUSE Nonterminals ordering of terminals; higher level of abstraction select-menu, position-mouse

draw line ::= select line + choose points + last point select line ::= pos mouse + CLICK MOUSE choose points ::= choose one | choose one + choose points choose one ::= pos mouse + CLICK MOUSE last point ::= pos mouse + DBL CLICK MOUSE pos mouse ::= NULL | MOVE MOUSE + pos mouse

Measurements with BNF Number of rules or number of + and | operators Complications same syntax for different semantics reflects user’s actions, not user's perception of system responses enforcement of consistency in rules Extensions include “information-seeking actions” in grammar parameterized grammar rules

Task-Action Grammar - TAG Making consistency in language more explicit than in BNF Encoding user's world knowledge (eg. up is opposite of down) Accomplished by Parameterized grammar rules Nonterminals are modified to include additional semantic features

Consistency in TAG In BNF, three UNIX commands would be described as copy ::= cp + filename + filename | cp + filenames + directory move ::= mv + filename + filename | mv + filenames + directory link ::= ln + filename + filename | ln + filenames + directory

Consistency in TAG In TAG, this consistency of argument order can be made explicit using a parameter, or semantic feature for file operations.

file op[Op] ::= command[Op]+ filename + filename | command[Op]+ filenames + directory command[Op = copy] ::= cp command[Op = move] ::= mv command[Op = link] ::= ln

Notes Ignore system output (there are extensions to BNF and TAG) Hierarchical and grammar-based techniques initially developed when systems were mostly command-line or keyboard and cursor based.

Physical and device models Based on empirical knowledge of human motor system User's task: acquisition, then execution. These models only address execution Models are complementary with goal hierarchies Models The Keystroke Level Model (KLM) Buxton's 3-state model

Keystroke Level Model - KLM Six execution phase operators Physical motor K keystroking P pointing H homing D drawing Mental M mental preparation System R response Times are empirically determined. T execute = TK + TP + TH + TD + TM + TR

Example GOAL: ICONISE-WINDOW [select GOAL: USE-CLOSE-METHOD MOVE-MOUSE-TO-WINDOW-HEADER POP-UP-MENU CLICK-OVER-CLOSE-OPTION GOAL: USE-L7-METHOD PRESS-L7-KEY]

Models so far GOMS – cognitive processing involved in deriving subgoals to carry out a task to achieve a goal CCT – distinction between LTM (rules) and STM (working memeory) Linguistic (BNF and TAG) – focus on syntactic KLM – motor and mental operators

Architectural models All of cognitive models make assumptions about the architecture of the human mind. Problem spaces – behavior viewed as sequence of agent/environment states (can predict erroneous behavior) Interacting Cognitive Subsystems provides model of perception, cognition, and action 9 subsystems (5 physical, 4 mental) view of user as information processing machine concerned with determining how easy particular procedures of action sequences become

Last notes Cognitive models attempt to represent users as they interact with the system Three categories – what were they? Most cognitive models do not deal with user observation and perception. Some techniques have been extended to handle system output, but problems persist. Issues: Level of granularity Exploratory interaction versus planning