F2F3 9 Jun 2006 Session 4

9 Jun 2006


Chris Welty


Chris: This discussion is to review the current RIFRAF discriminators proposed by Harold and make sure we understand them.

Semantic discriminators

terminating vs. non-terminating rule-bases, discriminator on language?

Harold: part of language (or given KB in language)

csma: is it possible to write rule-bases such that some queries do not terminate?

Harold: prohibit from language non-terminating queries, or prohibit recursion

csma: Feature of rule-base, not language?

Piero: Feature of semantics, i.e. semantic discrimintor

Hassan: discriminiator on rule set

Piero: Object to consider termination as discriminator of language, more aspects

Harold: examples where termination needs to be guaranteed
... discriminator needs rewording
... Does the language allow to express non-terminating queries (as annotations)

Chris: But that's not a discriminator of the language

chris: remove rulebases from semantic discriminator 1

Piero: property of termination is a collective combination of language, rules base, inference engine, ...

<sandro> gary: a rule engine that didn't terminate for a terminating language is a bug

Piero: termination of a language depends on evaluation (e.g. bottom-up vs. top-down)

<sandro> Piero: top-down might be the right thing.

Piero: objects to 1 as discriminator on language (even if "rulebase" omitted)

<sandro> Jos: "The problem of ground entailment is Decidable" is a good phrasing.

josB: add "real" semantic discriminators

e.g. non-decidable

Turing complete is a property of expressiveness

chris: decidable vs non-decidable, Turing vs. Non-Turing complete as semantic discriminators
... 1st Turing ...

<scribe> New number 2: decidable ...

josb: have discriminator "expressiveness" (that can be Turing complete etc.)

PaulV: Why are these discriminators useful?

chris: apply discriminators to different systems, different dialects

<scribe> New: 1. Turing-complete vs. non ...

<scribe> New 2: Decidability

<josb> 3. finite-model property

Harold/chris: delete old "4" (modality ...)

chris: keep modality, skip intentionality
... add operational, declarative semantics as discriminator?

sandro: not property of language, but of language specification

Piero: is property of program

Agreement: operational/declarative semantics NOT added

pragmatic discriminators

Axel: How to account for describing under which circumstances a system terminates?

Piero: inference control not a discriminator of the language

Human annotation of rule sets or inherent discriminator of rule sets?

<sandro> Chris: If you have a language which HAS inference control, then you want to be able to convey that.

Piero: distinguishe rule systems from just languages when assigning discriminators

Axel: separate discrimantors on languages vs. systems

Chris: these are discriminators on SYSTEMS

hassan: pragmatic discriminators = How systems/languages are used

dave: e.g. side-effects are part of language

Piero: Notion of "language" needs to be clarified
... Language = syn sem or syn sem pragm?

Axel: Keep "inference control", but in longer run be more precise as to language vs. system discrimantors

chris: label "pragmatic" not important

sandro: what are actions following from point "inference control"?

chris: do topics belong to RIFRAF is important, not headings

next discriminator: computational complexity

chris: granularity of complexity discriminator?

add explicit complexity categories?

Hassan: complexity preserving translations are to be guaranteed

Piero: Which operations have which complexity consequences: needs to be taken into account
... potential impact on RIF: certain languages might not be embeddable if the have certain complexities

Harold: Most rulesets can be classified according to complexity classes (see "deciabilty" topic on WIKI)

csma: question for translator/implementor not for RIF?

Harold/Piero: useful tool to classify classes of languages

chris: copy text from "above" to topic comp. complexity

next discriminator: interoperabilty

<sandro> 3.1. Annotations

Harold: Annotations means comments, e.g. in controlled English

3.1 annotations accepted as discrim

3.2 test cases not a discriminator?

test case is kind of annotation

<sandro> 3.2. Test Cases -- does the language allow you to say "this is a test case?"

<sandro> 3. Annotations for Interoperability

<sandro> (a kind of metadata)

3.3 mappings

chris: does language allow to attach interoperabilty information?
... syntactic sugar in one language not supported in other language, use clusters

csma: decide what to do wrt to 2nd WD

Paul: need enumeration on certain topics, e.g. complexity classes

chris: Put refined/elaborated version of discriminators in next WD

csma: need to have sth about RIFRAF in next WD
... make clear that current/refined version is work in progress
... agreement that the list makes sense "in principle"
... list should be edited/commented/detailed, e.g. by Gary?

chris: reconcile list with requirements needed

Who is taking action?

chris: identify/refine new discriminators

sandro: stick to phase I discriminators

csma: two questions: what to do wrt to next WD AND what to do in the longer run

annotate existing discriminators wrt to phase 1 and phase 2

Hassan: put all discr. in next WD, but annotate according to phase

concrete proposal after break

Summary of Action Items

[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.127 (CVS log)
$Date: 2006/06/17 20:13:49 $