Towards a ContextSensitive Structure for Behavioural Rules Context

Towards a Context-Sensitive Structure for Behavioural Rules (Context, Scope, Antecedents, and Results) Bruce Edmonds Centre for Policy Modelling, Manchester Metropolitan University Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 1

Summary of Talk: a view from Cognitive Science Suggest dividing behavioural rules into 4 bits: – Context – Scope – Antecedents – Results • Since this, I argue, seems to align with human cognitive structure • Which are each dealt with and updated in different ways (making their use feasible) • And thus might be a more “natural” structure for human behavioural rules Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 2

Different Aspects I Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 3

Different Aspects II Universe of Knowledge indicated by current cognitive context Knowledge that is possible to apply given circumstances Cause 1 & Cause 2… Result 1 & Result 2… Cause 3 & Cause 2… Result 5 & Result 2… Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 4

Bit 1: Context Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 5

A (simplistic) illustration of context from the point of view of an actor Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 6

Situational Context • The situation in which an event takes place • This is indefinitely extensive, it could include anything relevant or coincident • The time and place specify it, but relevant details might not be retrievable from this • It is almost universal to abstract to what is relevant about these to a recognised type when communicating about this • Thus the question “What was the context? ” often effectively means “What about the situation do I need to know to understand? Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 7

Cognitive Context (CC) • Many aspects of human cognition are contextdependent, including: memory, visual perception, choice making, reasoning, emotion, and language • The brain somehow deals with situational context effectively, abstracting kinds of situations so relevant information can be easily and preferentially accessed • The relevant correlate of the situational context will be called the cognitive context • It is not known how the brain does this, and probably does this in a rich and complex way that might prevent easy labeling/reification of contexts Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 8

The Context Heuristic • The kind of situation is recognised in a rich, fuzzy, complex and unconscious manner • Knowledge, habits, norms etc. are learnt for that kind of situation and are retrieved for it • Reasoning, learning, interaction happens with respect to the recognised kind of situation • Context allows for the world to be dealt with by type of situation, and hence makes reasoning/learning etc. feasible • It is a fallible heuristic with social roots in terms of the coordination of action, norms, habits Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 9

Some Possible Examples of Cognitive Context? • • Greeting someone you do not know A lecture An interview Being Lost Being Socially Embarrassed Travelling on a train/bus Leaving home to go somewhere Accidently bumping into someone you do not know on the pavement/in the corridor Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 10

Some Research Responses to Context-Dependency A number of responses: • Only do research within a single context, resisting any generalisation • Only use discursive, natural language approaches where context is implicitly dealt with (but also mostly hidden) • Try to see what (inevitably weaker knowledge) is general across the various contexts in what is being studied Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 11

Context-Dependency and Randomness Lots of information lost if randomness used to “model” contextual variation Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 12

However • Although Cognitive Context in General might be hard to identify • Socially Entrenched Contexts are often rather obvious • But one needs to drop the imperative of looking (only) for abstract and general theories for behaviour • Being satisfied with more “mundane” and context-dependent accounts Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 13

Choice and Update of Cognitive Context • CC is largely learnt from experienced situations in a rich and unconscious way • Occasionally one can realise one has the wrong context if a lot of the detailed knowledge it indicates is simultaneously wrong but which is the right CC is a matter of recognition from past positive learning • Once CC is learnt it is very difficult to change, but new CC can still be learnt Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 14

Identifying Context from Narrative Evidence • Apart from socially entrenched contexts (lectures, parties, interviews etc. )… • …the relevant CC is hard to identify from narrative evidence because: – To a large extent, we recognise the right CC for any text unconsciously and easily – The CC are learnt in a rich, “fuzzy” manner over a long period of time by inhabiting them which resists reification • This is one of the prime needs: how to “mark up” the CC behind narrative evidence? Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 15

Bit 2: Scope Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 16

About Scope • By “scope” I mean the reasoning as to which knowledge is possible given the circumstances • For example, if all the seats are taken in a lecture, then the norms, habits and patterns as to where one sits do not apply • Reasoning about scope can be complex and is done consciously • However once judgements about scope are made then they tend to be assumed, unless the situation changes critically Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 17

Scope vs. Cognitive Context • Both scope and cognitive context determine which knowledge is useful for any particular situation that is encountered • However, they play very different roles: – CC is learnt using pattern recognition over a long time, but then is largely a ‘given’, is almost impossible to change when learnt, is quick and automatic and is socially rooted – Scope is largely reasoned afresh each time, taking effort to do so, is possible to re-evaluate but only if needed, and is more individually oriented Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 18

Identifying and modelling scope • Compared to CC, scope is relatively well studied using formal models of reasoning – e. g. Updating Markoff/state representations of causation, non-monotonic logics, causation in Baysian networks etc. • Scope plays a relatively explicit part in human language, sometimes being explicitly stated, at other times using relatively well understood rules – e. g. conversational implicature • It is often possible to infer participant’s judgements as to scope and possibility, when not explicitly mentioned Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 19

Bits 3&4: (local) Narrative Steps Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 20

Encoding Narrative Steps • *If* CC and scope is identified then, I hypothesize, the local narrative structure will be easier to understand, because changing CC and/or scope can do a lot of the “work” in expressing/encoding knowledge • Within CC & scope I suggest a simple basic structure of sets of statements of the form: (on the whole) Z follows/followed from A, B… • A very special case of this is when we say that: A, B… implies Z or that: A, B… causes Z • (I will write A, B… Z), where A, B are the “Antecedents” and Z is the Results Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 21

About Narrative Steps • These might not be crisp but of the nature More A and B tends to result in more Z • These are often chained in forwards, branching or backwards manner to make an inference or a narrative • (even quite classical) formal logics and annotation systems capture these • Most AI/expert systems encode these, but rarely touch on scope (that is advanced AI) and never on Context Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 22

Conclusion Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 23

CSAR as a bridging structure between narrative text and behavioural rules *IF* this structure turns out to be a useful and “natural” encoding of human narrative knowledge/expression then two steps are needed: 1. Techniques to capture/approximate/guess appropriate Cognitive Contexts and Scope judgments from Narrative Text 2. AI/Computer science architectures that make the encoding and use of CSAR structured knowledge Towards a Context-Sensitive Structure for Behavioural Rules, Bruce Edmonds, Informal Workshop on Qual. Evidence & Rules, MMU, Sept. 2012, slide 24