Using Controlled Natural Language and First Order Logic to improve e-consultation discussion forums (DERI reading group talk)

Editor's Notes

• #2: ==== Reading Group ==== *Presenter:* Jodi Schneider *Supervisor:* Alexandre Passant *date:* 7.9.2011 *time:* 11am - 12pm *location:* Conference Room *Title:* Using Controlled Natural Language and First Order Logic to Improve E-Consultation Discussion Forums *Abstract:* Online consultation has become a common way to increase participation in policy-making, yet ensuring that citizens' input is adequately understood remains challenging, since standard discussion forums offer few possibilities for structuring posts based on their meaning. This reading group will discuss efforts to structure the forums while retaining ease-of-use, based on a trio of papers from the IMPACT project [1]. The underlying idea is to construct possible policies by finding the maximal set of consistent viewpoints, since a policy statement must avoid internally contradictions (e.g. be consistent). The first paper is a requirements analysis, providing a justification and overview of the work they propose: adding structure with controlled natural language, argumentation frameworks, and user input of term relationships (e.g. "contradicts, "is premise of", or "is an exception to"). It also introduces a running example: 16 propositional statements, extracted from an online debate "Should people be paid to recycle?" [2], such as "Paying tax for garbage is unfair." or "No householder should pay tax for the garbage which the householder throws away." The statements from the running example are analyzed in the second paper, using a controlled natural language, Attempto Controlled English (ACE) [3], which are then modified in order to ensure that ACE can parse the sentences, and that they yield the intended interpretation. In order to derive possible policies, based on the sentences, the authors use a first-order inference engine to analyze the consistency of sets of sentences. They explore some of the problems they encountered. Extending that work, the third paper envisions using ACEWiki [4] to construct a consistent knowledge base incrementally, based on users' input and using the Pellet inference engine to test for consistency. It presents an argumentation framework for the recycling debate--a graph showing the relationships between the 16 sentences of the running example--in order to find the maximal set of consistent propositions, i.e. a possible policy. ==== based on ==== *Paper title:*A Framework for Enriched, Controlled On-line Discussion Forums for e-Government Policy-making *Authors:* Adam Wyner and Tom van Engers *Published at:* EGOVIS 2010 *Abstract:* The paper motivates and proposes a framework for enriched on-line discussion forums for e-government policy-making, where pro and con statements for positions are structured, recorded, represented, and evaluated. The framework builds on current technologies for multi-threaded discussion lists by integrating modes, natural language processing, ontologies, and formal argumentation frameworks. With modes other than the standard "reply comment", users specify the semantic relationship between a new statement and the previous statement; the result is an argument graph. Natural language processing with a controlled language constrains the domain of discourse, eliminates ambiguity and unclarity, allows a logical representation of statements, and facilitates information extraction. However, the controlled language is highly expressive and natural. Ontologies represent the knowledge of the domain. Argumentation frameworks evaluate the argument graph and generate sets of consistent statements. The output of the system is a rich and articulated representation of a set of policy statements which supports queries, information extraction, and inference. *Link:* http://wyner.info/research/Papers/WynerVanEngersForum2010.pdf *Paper title:* From Policy-making Statements to First-order Logic *Authors:* Adam Wyner, Tom van Enger, and Kiavash Bahreini. *Published at:* Electronic Government and Electronic Participation 2010 *Abstract:*Within a framework for enriched on-line discussion forums for e-government policy-making, pro and con statements for positions are input, structurally related, then logically represented and evaluated. The framework builds on current technologies for multi-threaded discussion, natural language processing, ontologies, and formal argumentation frameworks. This paper focuses on the natural language processing of statements in the framework. A small sample policy discussion is presented. We adopt and apply a controlled natural language (Attempto Controlled English) to constrain the domain of discourse, eliminate ambiguity and unclarity, allow a logical representation of statements which supports inference and consistency checking, and facilitate information extraction. Each of the policy statements is automatically translated into first-order logic. The result is logical representation of the policy discussion which we can query, draw inferences (given ground statements), test for consistency, and extract detailed information. *Link:* http://wyner.info/research/Papers/WynerVanEngersBahreini2010.pdf *Paper title:* Towards Web-based Mass Argumentation in Natural Language *Authors:* Adam Wyner and Tom van Enger *Published at:* long version submitted to EKAW 2010 *Abstract:* Within the artificial intelligence community, argumentation has been studied for quite some years now. Despite progress, the field has not yet succeeded in creating support tools that members of the public could use to contribute their views to discussions of public policy. One important reason for that is that the input statements of participants in policy-making discussions are put forward in natural language, while translating the statements into the formal models used by argumentation scientists is cumbersome. These formal models can be used to automatically reason with, query, or transmit domain knowledge using web-based technologies. Making this knowledge explicit, formal, and expressed in a language which a machine can process is a labour, time, and knowledge intensive task. To make such translation and it requires expertise that most participants in policy-making debates do not have. In this paper we describe an approach with which we aim at contributing to a solution of this knowledge acquisition bottle-neck. We propose a novel, integrated methodology and framework which adopts and adapts existing technologies. We use semantic wikis which support mass, collaborative, distributive, dynamic knowledge acquisition. In particular, ACEWiki incorporates NLP tools, enabling linguistically competent users to enter their knowledge in natural language, while yielding a logical form that is suitable for automated processing. In the paper we will explain how we can extend the ACEWiki and augment it with argumentation tools which elicit knowledge from users, making implicit information explicit, and generate subsets of consistent knowledge bases from inconsistent knowledge bases. To a set of consistent propositions, we can apply automated reasoners, allowing users to draw inferences and make queries. The methodology and framework take a fragmentary, incremental development approach to knowledge acquisition in complex domains. *Link:* http://wyner.info/research/Papers/WynerVanEngersEKAW2010.pdf [1] http://www.policy-impact.eu/project-summary [2] http://newsforums.bbc.co.uk/nol/thread.jspa?forumID=7269 [3] http://attempto.ifi.uzh.ch/site/ [4] http://attempto.ifi.uzh.ch/acewiki/ ====Why you should attend?==== This is a very practical approach to overcoming the knowledge acquisition bottleneck while retaining ease-of-use, allowing information to be added by a wide range of non-experts. Controlled natural languages, first order logic reasoners, semantic wikis, and argumentation frameworks will be briefly introduced.
• #3: Context: IMPACT Project, an FP7 program, “ Integrated Method for Policy Making Using Argument Modelling and Computer Assisted Text Analysis" http://www.policy-impact.eu/ “ The research goals of the project aim to further the state-of-the-art of computational models of argumentation about policy issues; contribute to computational linguistics by developing methods for mining arguments in natural language texts; discover ways to increase the inclusiveness and quality of public participation in consultation processes, in ways which cut across language barriers; and, finally, discover or invent user-interfaces and visualizations for computational models of policy argumentation which make these models accessible and usable to a broad public.” - http://www.policy-impact.eu/project-summary They’re developing a toolbox of 4 tools: an argument reconstruction tool, policy modelling and analysis tool, a structured consultation tool, an argument analysis, tracking and visualization tool.
• #6: 700+ comments! http://newsforums.bbc.co.uk/nol/thread.jspa?forumID=7269&edition=2&ttl=20110906115945
• #7: http://newsforums.bbc.co.uk/nol/thread.jspa?sortBy=2&forumID=7269&edition=2&ttl=20110906115945&#paginator Limitations of existing systems * Participant must understand (and specify) the relationships between statements * Statements cannot be context-dependent (e.g. where the conclusion of one argument is the premise of another argument) * The architecture is not modular (e.g. to allow an administrator to add different relationships or debate components) * "The linguistic content of the statements is unanalysed and unconstrained; that is, the statements are not parsed, or given a semantic interpretation, or required to be relevant and novel to the current discussion, or constrained in terms of terminology and length" * There is no formally specified argumentation semantics, which would allow determination of sets of consistent statements.
• #8: Manual reading?!
• #10: They don’t say how they extracted these – but they say Someone makes statement (1) Someone else gives (4) as a reason/premise for (1) Someone else gives (3) as an additional reason for (1) (2) Is a counterproposal with a range of supporting reasons
• #11: They don’t say how they extracted these – but they say Someone makes statement (1) Someone else gives (4) as a reason/premise for (1) Someone else gives (3) as an additional reason for (1) (2) Is a counterproposal with a range of supporting reasons === Icons: http://findicons.com/icon/27954/girl_5?id=27964# http://findicons.com/icon/27930/boy_8?id=27939# http://findicons.com/icon/27955/girl_4?id=27965#
• #12: However (4) is criticised by the claim that the Professor is not objective (11) (implying that his statement (9) does not hold) (16) attacks (15) – this is one of the premises of (2) [hence negates a premise]
• #13: However (4) is criticised by the claim that the Professor is not objective (11) (implying that his statement (9) does not hold) (16) attacks (15) – this is one of the premises of (2) [hence negates a premise]
• #14: However (4) is criticised by the claim that the Professor is not objective (11) (implying that his statement (9) does not hold) (16) attacks (15) – this is one of the premises of (2) [hence negates a premise]
• #16: CNL & FOL details is where we get into the 2 nd paper -- From Policy-making Statements to First-order Logic.
• #17: http://www.cafepress.com/+man_eating_shark_sticker_rectangle,410247547 http://www.movieposterdb.com/poster/730f3df5
• #18: Note that this “intricate expression” simplifies coreference resolution: “A man… the man”
• #19: Simplify lexicon & syntax. Avoid gerunds, participles, complex nouns. Make implicit knowledge explicit, state all relevant participants. Use determiners on nouns (some, a, every). Use possessive nouns, not pronouns. Full list of translations to ACE (1) Every household should pay some tax for the household’s garbage. (2) No household should pay some tax for the household’s garbage. (3) Every household which pays some tax for the household’s garbage increases an amount of the household’s garbage which the household recycles. (4) If a household increases an amount of the household’s garbage which the household recycles then the household benefits the household’s society. (5) If a household pays a tax for the household’s garbage then the tax is unfair to the household. (6) Every household should pay an equal portion of the sum of the tax for the household’s garbage. (7) No household which receives a benefit which is paid by a council recycles the household’s garbage. (8) Every household which does not receive a benefit which is paid by a council supports a household which receives a benefit which is paid by a council. (9) Tom says that every household which recycles the household’s garbage re- duces a need of a new dump which is for the garbage. (10) Every household which reduces a need of a new dump benefits the house- hold’s society. (11) Tom is not objective. (12) Tom owns a company that recycles some garbage. (13) Every person who owns a company that recycles some garbage earns some money from the garbage which is recycled. (14) Every supermarket creates some garbage. (15) Every supermarket should pay a tax for the garbage that the supermarket creates. (16) Every tax which is for some garbage which the supermarket creates is passed by the supermarket onto a household.
• #20: Getting rid of modal verbs
• #24: (Side note)
• #27: This is where we get to the 3 rd paper: Towards Web-based Mass Argumentation in Natural Language
• #28: They don’t say how they extracted these – but they say Someone makes statement (1) Someone else gives (4) as a reason/premise for (1) Someone else gives (3) as an additional reason for (1) (2) Is a counterproposal with a range of supporting reasons === Icons: http://findicons.com/icon/27954/girl_5?id=27964# http://findicons.com/icon/27930/boy_8?id=27939# http://findicons.com/icon/27955/girl_4?id=27965#
• #31: a1 = 1,3,4,9,10 a2 = 11,12,13 a3 = 2,5,6,7,8,14,15 a4 = 16
• #32: From http://attempto.ifi.uzh.ch/site/pubs/papers/semwiki2009_kuhn.pdf http://attempto.ifi.uzh.ch/acewiki/
• #33: From http://attempto.ifi.uzh.ch/site/pubs/papers/semwiki2009_kuhn.pdf
• #35: ACE was the main manual work
• #44: Integrated Method for Policy Making Using Argument Modelling and Computer Assisted Text Analysis