[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Re: [legalruleml] [Ambiguity] Incomplete representation
Tara, for sure we will include your proposal (integration of the LegalRuleML principles document) in the next TC meeting, and by mailing list we will collect also counter proposals. Passing to the proposal XML fragment. LegalRuleML aims to be agile, semantic, expressive and close to the end user (legal knowledge engineer) mind. That means for instance to express the rule in readable way <Rule idReq="rule1" iri="&ruleml;Implies> <if idReq="imp1" iri="&ruleml;if> <Atom idReq="atm1"></Atom> <Atom idReq="atm2"></Atom> </if> <then idReq="then1" iri="&ruleml;then> <Atom idReq="atm3"></Atom> <Atom idReq="atm4"></Atom> </then> </Rule> mp Il 01/06/2012 15:25, Tara Athan ha scritto:
First, I would like to continue the "ambiguity" thread, which was started in the skype chat after the last telecon, in the email. I have copied the relevant parts of the skype chat at the bottom of this email. Second, I think I have finally understood something about the objectives of LegalRuleML. PROPOSED: to add some text to the principles document, something like the following ... **************************************************** Requirements ... Incomplete Representation. It SHALL be possible for the LegalRuleML representation of a norm to be incomplete, with the missing information supplied by reference(s) to (the most specific) portion(s) of the textual norm. This is useful for several reasons. 1. Sometimes the textual norm is ambiguous, so a complete transformation into a single logical representation, which is by nature unambiguous, is not possible without creating a (legal) interpretation. Rather than being forced by the syntax to create such an interpretation, which the LegalRuleML author may not be authorized to do, they should have the option to NOT represent parts of the norm, such as the internal details of a literal. 2. Even if the textual norm is not ambiguous, complete and accurate representation of it in predicate logic may be beyond the capability of the LegalRuleML author. Representation of the details of a literal may require reference to a legal ontology. A suitable legal ontology may not exist, or the LegalRuleML author may not be sufficiently well trained to carry out the conversion of a natural language text into such a representation. 3. Accurate representation in predicate logic may require rearrangement of the grammatical structure of the sentence, leading to loss of "isomorphism" between the textual representation of the norm and the LegalRuleML representation. An incomplete representation can be used to retain isomorphism. However, an incomplete LegalRuleML representation SHALL contain only entirely correct and unambiguous information. That is, the loss of information is due to some information being missing, rather than the information that is present being approximate or vague. In this way, an incomplete LegalRuleML representation could be made complete by adding the missing knowledge. For example, a particular interpretation of a norm may clarify an ambiguous point. It should be possible to represent this interpretation monotonically - by adding additional knowledge to the knowledge base, rather than taking away or replacing existing knowledge. ****************************************************** I would like to point out that this kind of incomplete representation is already possible within the RuleML syntax, as presented in http://ruleml.org/indoo/indoo.html#Programs-as-Data. In the latest terms, using iri references to refer to the RuleML meta-model ontology that is currently under development, and the new proposed "id-ized" terms: <Atom> <oid> <Ind idref="rule1"/> </oid> <Rel iri="&ruleml;Implies/> <slot> <Ind iri="&ruleml;if"/> <Ind idref="premise1"/> </Expr> </slot> <slot> <Ind iri="&ruleml;then"/> <Ind idref="conclusion1"/> </slot> </Atom> In this construction, an internal identifier "premise1" could be associated (in metadata) either to an IRI denoting a portion of the textual norm for an incomplete representation or an IRI denoting a LegalRuleML representation for complete representation. ********************************************************************************************************* From Skype Chat... [5/31/2012 5:53:07 AM] Tara Athan: @Adam - would you also be willing to help with converting the natural language text, such as Guido's "complaint" example into Controlled English? [5/31/2012 6:02:05 AM] Adam Wyner: @Tara - ah, that is a very good idea. I've done this in the past and have a small paper; there tend to be a large number of issues that arise along the way. I would also like to do this for the regulation sample I made available for analysis to the TC. In the context of our TC discussions, the exercise would be very useful. Also, for me, it is essential to get publications out of such exercises, e.g. for conferences such as JURIX. How about if I take lead on the legislation -> ACE exercise and associated paper write up (background, surrounding discussions), and add others as co-authors since we will discuss the analyses and related issues of rule representation? Would that suit you and others do you think? [5/31/2012 6:05:50 AM] Tara Athan: That would suit me very well. Once the text is in ACE, I can easily see how to convert to RuleML, but getting it into ACE is outside of my experience. [5/31/2012 6:10:23 AM] Adam Wyner: Good. This sort of task is really very much needed in the community. Timewise, I'm off to a conference for a week tomorrow, but during the week I'll get myself set up for this, make a start, and let you know. There is an ACE -> RuleML conversion tool, so I'll see how that works. Also, there is ACE in Protege, for ontology expression.... My experience is that the translation can go slower than expected since there are interpretation issues and work arounds; I'll record these issues as they are relevant to better understand the issues and the task. [5/31/2012 6:36:24 AM | Edited 6:56:46 AM] Tara Athan: Unfortunately the webservice translator (http://ruleml.org/translator/translator.jnlp) from ACE to RuleML is not working at present. It relied on the APE webservices that has been moved. The API of the webservice has also been changed. It appears to be a simple matter of changing the URL and a few query string parameters, so it may be easy to get the translator working again. The result is in RuleML 0.9, but there are XSLTs available to convert to RuleML 1.0. Following the conversion by APE from ACE->DRS, there is an "isomorhpic" conversion into RuleML, as demonstrated in http://ruleml.org/translator/exa/own-ACE.ruleml , for example. The code to perform the DRS -> RuleML conversion is available as a java package, in binary or source, from http://ruleml.org/translator/ I have not yet tried out the APE package that is available for download (https://github.com/Attempto/APE/downloads) from the ACE website (http://attempto.ifi.uzh.ch/site/description/) for ACE -> DRS conversion. Are you familiar with that? [5/31/2012 7:02:24 AM | Edited 7:02:38 AM] mp: Take care to use ACE in legal domain, because the risk is to manipulate too much the norm and to affect the normative meaning per se. [5/31/2012 7:03:09 AM] mp: this is actually one of the scope of LegalRuleML: don't manipulate too much the NORM for permitting multiple interpretation [5/31/2012 7:03:25 AM | Edited 7:04:07 AM] mp: LKIF-rule also is based on this principle (Tom Gordon) because the state of the art in legal domain usually manipulate too much the norm according with other paradigma [5/31/2012 7:06:17 AM] Tara Athan: Ideally the legal text would be written in a controlled natural language to begin with. [5/31/2012 7:06:27 AM] mp: It is not [5/31/2012 7:06:59 AM] Tara Athan: That's why I said "ideally" [5/31/2012 7:10:35 AM] mp: This is a matter of "legal drafting" techniques. Since 30 years have tried to provide best practices and methodology, but the political level needs to provide different messages, sometime ambiguous and vague. Not for an error, but because ambigous and vagueness guarantee constitutional principles, flexibility in the interpretation of the legal values, respect of the fundamental rights over the time (e.g. regime). [5/31/2012 7:12:42 AM] mp: the idea is to not provide an ideal formalization of the law, but to use the law with its codification and its paradigma. This is the only way to provide concrete information system usable by the legal operators. [5/31/2012 7:24:41 AM] mp: http://www.semantic-web-journal.net/blog/semantic-web-journal-special-issue-semantic-web-legal-domain [5/31/2012 7:27:24 AM] Adam Wyner: Very interesting discussion and points I understand. I've used ACE locally in the past and will have to reload it to my current machine, test it out, etc. Tara's problems noted, so I'll check that out. [5/31/2012 7:27:54 AM] Adam Wyner: I understand Monica's point about 'normalising norms'. Interesting to emphasise it like this. [5/31/2012 7:28:33 AM] mp: There is a strong debate in the legal theory on this topic: the textualism theory is dead [Easterbrook, Frank H.,Textualism and the Dead Hand, 66 Geo. Wash. L. Rev. 1119 (1997-1998)] [5/31/2012 7:29:50 AM] mp: and a lot of constitutional and comparative law experts are now moving on a different direction called "implicit law" [Implicit law, Gerald J. Postema (Ed.), Law and Philosophy, Volume 13, Number 3, 361-387, 1994.] [5/31/2012 7:30:29 AM] Adam Wyner: Almost *all* work, so far as I'm aware, in NLP is about *disambiguation*, whereas Monica's point is about maintaining the ambiguity. How best to do this? That said, it is also not the case that *all* ambiguities are intended to be maintained for the reasons that Monica gives. I would say that there is much in legislation that suits unambiguious representation, perhaps the majority of regulation is this way. RuleBurst/Oracle have a large business in providing such representations as web-based executable law. [5/31/2012 7:31:17 AM] mp: in common law not in civil law domain [5/31/2012 7:31:39 AM] Adam Wyner: It is also the case that legal drafting guidelines try to specify *out* ambiguity - I just reviewed a paper for the controlled language workshop about this in a comparative study. [5/31/2012 7:35:00 AM] Tara Athan: How much of the ambiguity is in the vocabulary as opposed to the grammatical structure? [5/31/2012 7:36:24 AM] Adam Wyner: I understand Monica's point about the purposes of vagueness and ambiguity, but there is a further side. The point is that the legislation is then further specified and articulated in application - the interpretation of the article. Here legal cases may arise about the interpretation, where parties are in conflict about the interpretations. This may then be recycled back up to the legislative drafters for revision of the legislation.... So, at the level of interpretation, clarity is imposed on the initial legislation for the purposes set by the context. [5/31/2012 7:37:49 AM] mp: Ok legalruleml would like to permit ambiguities using defeasibility logic and not to force the language to a unique interpretation or an excessive simplification just for formalizing the legal language. If the legislator prouces ambiguities we maintain that because in some cases, like the human rights, it is normal to have vagueness and ambiguities for not blocking the norm to a particular historical period (e.g. regime, not democratic government, etc.) There are 10 years of legal theory on that, and the Sartor, Rotolo, Governatori etc. school (bologna school) follows this. [5/31/2012 7:38:01 AM] Adam Wyner: This discussion has already headed into the long Anglo-Analytic v. Continental-Interpretive debate. Oh, boy, that's fun! However, at some point, we shall want to scope the discussion and the analysis. [5/31/2012 7:38:10 AM] mp: Nice to have this discussion, in order to clarify the point. [5/31/2012 7:38:42 AM] mp: Thanks for the elicitation. [5/31/2012 7:40:04 AM] Tara Athan: [Thursday, May 31, 2012 7:35 AM] Tara Athan: <<< How much of the ambiguity is in the vocabulary as opposed to the grammatical structure? [5/31/2012 7:40:33 AM] Adam Wyner: And thanks to you for elicitation. I'm very pro interpretation (as any linguist must be), but I also wear my syntax-semantics translation into logic hat. Formalisation is not always the end in itself, but a great tool to understanding; that is, the effort to formalise makes clearer what we do/don't understand and precisely why. [5/31/2012 7:41:06 AM] Adam Wyner: Tara, I think there is going to be ambiguity in both the lexicon and grammatical structure. [5/31/2012 7:41:48 AM] Adam Wyner: That said, I think there are going to be many many many cases where a disambiguated lexicon and grammar are acceptable. [5/31/2012 7:43:58 AM] mp: Adam: yes we need a balance. I like the approach of Adam, but for sure to reduce the NORMS to ACE is not a good exercise for our purpouses in LegalRuleML that intends to preserve legal rules defeasible and sometime ambigous [5/31/2012 7:44:04 AM] Tara Athan: Ambiguity in the lexicon would be easy to handle- just create a user-defined lexicon that is "open to interpretation". The sentence would be one step away from controlled language, with substitution of unambiguous words for the ambiguous ones. I don't see any possible hope of representing an ambiguous grammatical structure in RuleML, nor would I want to. [5/31/2012 7:44:27 AM] Guido Governatori: There should be no ambiguity. What I mean is that what are there should be no ambiguity about what are the elements of a norms. The same syntactic structure can be intrepreted in different way, and the representation should allow for the specification of the semantics. [5/31/2012 7:44:31 AM] mp: So ACE is nice for starting, but it is not a good approach also because in Italian, in Spanish, in France, etc. doesn't exist (civil law) [5/31/2012 7:45:20 AM | Edited 7:53:44 AM] Adam Wyner: About this point: ambiguities using defeasibility logic. Much can and will have to be said. I think we really ought to be clear amongst ourselves about 'levels' of argumentation. Where we have an unambiguous lexicon and grammar, we can still formulate arguments and counterarguments. Where we have an ambiguous lexicon and grammar, we can have another level **of arguing about the lexicon and grammar/structure/interpretations**. Larry Solan, a US linguist and law scholar, claims that at least in the US, ambiguity and vagueness issues are **not** frequent or usually problematic. But, I know that the European legal tradition is very different, so maybe the 'meaning' of 'bank' is different in Greece and Spain than in the UK :-) Smiley is necessary here so you don't take my 'bank' point to heart. [5/31/2012 7:46:37 AM] mp: Adam: great arguemnts and counterarguments means to permits to have both and to delegate to the interpretation and to the concrete facts. [5/31/2012 7:47:37 AM] Adam Wyner: Guido's point has some assumptions about natural language syntax and semantics in it. There are several different approaches, and there is nothing for us to commit to. One approach says that 'underlying' syntactic structure correlates to one semantic interpretation; other approaches are not 1-1. [5/31/2012 7:48:28 AM] Adam Wyner: I agree with Monica's point about multiple languages. If Grammatical Framework were to have a semantic translation, then that would be better.... Perhaps that is in the pipeline. [5/31/2012 7:49:47 AM] Adam Wyner: I agree with Monica about arguments about interpretation and concrete facts. Trevor and I also have discussions and some work on these issues, so I understand the point. Just that I find it helpful to separate out 'spheres' of what is argued about and how. We can argue about values, about facts, about proposals for actions, about expertise.... [5/31/2012 7:50:33 AM] Tara Athan: The use of ACE in an example should not preclude the application of LegalRuleML to any other means of representing the legal text in RuleML. But each example must have some means to do so. [5/31/2012 7:51:58 AM] Adam Wyner: I think the Norms to ACE exercise should be an opportunity to make explicit and clear what does/doesn't work as well as how and why. I'm not claiming that it is correct or all that needs to be done. [5/31/2012 7:55:54 AM | Edited 7:56:33 AM] mp: So the only recommandation is: use ACE for now just for make some examples, but it is not a methodology adopted in LegalRuleML beacuse we would like to have also the possibility to manage ambiguities and contradictory norms. The normalization of the language of the law is very reductive. [5/31/2012 7:56:41 AM] Guido Governatori: to establish different levels of burden of proof one can use one and the same rule, but the semantics used (or the proof theory) could differentiate the conclusions. We do not have multiple rules, but different ways in which rules are used. [5/31/2012 7:57:42 AM] Tara Athan: Handling contradictory norms through defeasible logic is completely compatible with ACE. [5/31/2012 7:59:43 AM] mp: if you have subordinate clause, dependent clause and exception of exception, you are forced with ACE to break the rule/norm in several sub-rules and the effect is different. [5/31/2012 7:59:56 AM] mp: in Italian law is quite frequent. [5/31/2012 8:00:09 AM] mp: and also in the European Directive (UE Parliament) [5/31/2012 8:00:44 AM] mp: one of the limitation of LKIF-rule was exactly this point. Now is able to manage exception, but not exception of exception. [5/31/2012 8:03:48 AM] Adam Wyner: I agree with this: So the only recommandation is: use ACE for now just for make some examples. Not sure what Tara means by this: Handling contradictory norms through defeasible logic is completely compatible with ACE. ACE has nothing to say about handling contradition or defeasible logic; on the contrary (last I looked) it is really only for classical strict reasoning. That said, we can use ACE to translate multiple sentences which may be contradictory and then process them outside of ACE. Monica's point could use with some explicit worked examples with ACE: if you have subordinate clause, dependent clause and exception of exception, you are forced with ACE to break the rule/norm in several sub-rules and the effect is different. Not saying I disagree, just that it should be seen. @Adam - yes that is what I meant: the defeasibility is "on top of" the ACE representation, so the expression of defeasibility does not need to be stated in ACE, just the clauses. I would also be interested to see the example Monica mentions. [5/31/2012 10:30:41 AM | Edited 10:30:59 AM] mp: You mean <logop> example? [5/31/2012 10:32:06 AM] Tara Athan: I mean the example of exception of exception. [5/31/2012 10:32:42 AM] mp: yes modification of modification it also a case, repeal of repeal [5/31/2012 10:32:47 AM] mp: ok --------------------------------------------------------------------- To unsubscribe, e-mail: legalruleml-unsubscribe@lists.oasis-open.org For additional commands, e-mail: legalruleml-help@lists.oasis-open.org .
-- =================================== Associate professor of Legal Informatics School of Law Alma Mater Studiorum Università di Bologna C.I.R.S.F.I.D. http://www.cirsfid.unibo.it/ Palazzo Dal Monte Gaudenzi - Via Galliera, 3 I - 40121 BOLOGNA (ITALY) Tel +39 051 277217 Fax +39 051 260782 E-mail monica.palmirani@unibo.it ==================================== LA RICERCA C’È E SI VEDE: 5 per mille all'Università di Bologna - C.F.: 80007010376 http://www.unibo.it/5permille Questa informativa è inserita in automatico dal sistema al fine esclusivo della realizzazione dei fini istituzionali dell’ente.
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]