ITA White Paper for the 2013 BPP Fact Extraction and Reasoning using Controlled Natural Language (v10) This is a Continuation of Project 4, Task 2, with the addition of new research topics. Research Issue/Technical Approach This section should be up to 2000 words in length, describing the research issues to be addressed and the technical approach to solve them. This should include: Problem background, the associated technical challenges in coalition environments, and why it is important to solve this problem The technical approach used to address the problem Approach/intention to validate/verify the output of the research Key deliverables and planned research outputs Identify the fundamental research issues to be addressed and how the research may advance the state-of-the-art, develop fundamental knowledge, and provide generalizable results. If this is a continuation of an existing project, identify what additional value will be obtained by continuing the project. Problem Statement We propose to extend current P4T2 research on using a Controlled Natural Language, specifically Controlled English (CE) [1,2,3], for fact extraction from unstructured text and reasoning from these facts to provide high-value information [4,5]. By focusing on the challenging problem of Natural Language (NL) processing and reasoning we address key issues to improve our CE capabilities and facilitate greater adoption of CE, as recommended in the peer review. Improving fact extraction is also a requirement in the whitepapers call (TA6-3a), and improving reasoning capabilities will enhance information gathering for decision making. We diagram a “motivating scenario”, a class of coalition problems combining reasoning to obtain high-value information with interpretation of unstructured text, and exchange, explanation and understanding of results with other humans and systems. Problems include collaborative planning, decision making, and intelligence analysis, e.g. by a Battalion in a coalition (shown in dark below): ITA White Paper for the 2013 BPP The context of the UK Battalion [BN] operation is given by a US Brigade [BDE], as plans, intelligence and background information. The Battalion must work and communicate with a US Battalion and an (unspecified) host nation unit. The UK Battalion intelligence officer must provide high-value information and intelligence to the Brigade, based upon information fusion from different sources and information types: from the US Battalion, host nation, Company patrols, specialist reports, sensors, intelligence. Information processing is summarized below, where analysts’ information requirements are determined by decision-making, driving them to construct high-value information and rationale. High-value facts are constructed by extracting basic facts from unstructured and other sources, and reasoning to infer high-value information, patterns or significant events. The need to extract facts may drive the search for other sources. In the coalition, basic information may be provided by a partner, and high-value information may be provided back to other partners. ITA White Paper for the 2013 BPP CE fact extraction and reasoning has benefits: 1. CE as the target for fact extraction can integrate analysis and representation of unstructured and structured information, providing a common human- and machinereadable form of English. This is also useful when the amount of unstructured information is large, and human users need machine assistance in directing their attention to relevant information. 2. CE as a means of configuring NL processing (via human-readable linguistic rules and concepts) can help analysts to understand application of NL processing to their domain specific concepts and ways of expressing those concepts in unstructured text. Analysts can understand the effects of processing on quality of information and can configure the system to process new types of information. 3. CE can provide analysts with a human-friendly unambiguous language and a reasoning system for defining concepts and rules that infer high-value information, patterns or significant events from basic facts extracted and allow users to explore hypotheses about the interpretation of data. 4. CE can facilitate analysis under rapidly changing information needs by supporting dynamic creation and modification of analytic concepts and rules, allowing analysts to explore new ways of analyzing data and incorporate new sources of unstructured information. 5. CE can facilitate assessment of the quality of high-value information and explanation of conclusions via rationale including assumptions and uncertainty factors. 6. CE can facilitate information communication between coalition partners by providing formal unambiguous concepts with means for expressing them that help reduce misunderstanding. ITA White Paper for the 2013 BPP P4T2 research has produced results in these areas [4,5,6,7], but significant challenges remain to achieve wider acceptance of CE. The first challenge (C1) is to improve quality of facts extracted from a greater range of unstructured sources: extending capabilities of NL processing configured in CE extending CE-based reasoning capabilities The second challenge (C2) is to extend capabilities of CE itself: extending CE to allow expression of facts in a more readable way providing explanatory capability for conceptual models These challenges are related as fact extraction provides a powerful example of complex reasoning which will drive extensions to CE and reasoning capabilities, benefiting future acceptance of CE. Proposed Approach To meet the challenges, we propose research into interpretation of unstructured text leading to CE facts and reasoning on extracted facts to provide high-value information, ( “Fact Extraction” , “Analyst Reasoning” in the diagram). We build upon the P4T2 framework [4,5] where fact extraction from unstructured text was achieved and configured using CE in the CE store. This included a multilayered conceptual model of linguistics and semantics, based upon scientific principles and permitting users to rapidly configure domain-specific NL processing and reasoning, and providing rationale for conclusions, as demonstrated at ACITA12. We retain CE at the core of this framework as it is a strong enabling factor in the six benefits noted above. We propose to extend this framework to address the challenges: (C1) by extending NL processing techniques in CE to cover more aspects of English and to handle ambiguities, taking advantage of available linguistic resources and theories, to make scientific advances in linguistics more accessible to non-specialists (C1) by extending the reasoning component, allowing users to define assumptions, hypotheses, uncertainties and constraints, and rules for inferring information within different hypotheses (C2) by extending CE syntax and semantics, using insights from our NL processing research, to improve CE expressibility and logical power whilst avoiding ambiguities. (C2) by extending rationale to provide better explanations of conceptual models Enhancing NL processing is beneficial in its own right, but also demonstrates CE in a complex reasoning problem, motivating extensions to the reasoning component and helping the ITA to use CE in other areas. However it offers significantly more; understanding NL processing can provide insights into selection of expressive but non-ambiguous extensions to CE and can inform work on cognition and language. We will apply fact extraction to larger samples of unstructured data (recommended in the peer review) whilst enhancing support to users for constructing and understanding the processing. ITA White Paper for the 2013 BPP We believe that the CE-based approach to NL processing continues to be novel due to the pervasive use of human-readable CE throughout the interpretation and reasoning as well as extending CE to better support extraction of rich information and its exchange across collaborating partners. Enhancing Natural Language Capabilities External Linguistic Resources We propose to leverage external linguistic resources (lexicons and linguistic processing rules) into our CE-based framework, providing greater fact extraction capability whilst making them easier to understand and useable by reasoning components. This requires development of CEbased representations for linguistic concepts defined by these resources, such as lexical categories, word forms and semantic constraints. The nature of these key structures is still a matter of linguistic research, and the development of a unified CE model will be a major task [8,9]. We will also research into how analysts may be involved in tailoring linguistic resources to domain-specific conceptual models. Potential resources include Wordnet [10,11], Verbnet [12] (lexical/semantic relations between words) the English Resource Grammar (ERG) [13] (large English Grammar) and the Grammar Matrix [14] (starter kit for developing grammars and language processing). The latter builds upon the Linguistic Knowledge Base [15], developed by team member Prof Ann Copestake, within the DELPH-IN consortium [16], and thus we will contribute to the NL community. Deeper Semantics Fundamental to NL processing for fact extraction is representation and interpretation of complex relations between words and structures in a sentence and the underlying semantics/meaning of the sentence components, and how these are composed into the propositions expressed in the sentence, so that sentence meaning may be expressed unambiguously, e.g. CE facts based on the conceptual model. More complex sentences require more sophisticated semantics, and in addition to the proposed enhancements provided by leveraging external linguistic resources, such “compositional semantics” remains a fundamental research area [17,33] for capabilities to interpret and represent more complex sentences in CE. We propose in-depth research on how compositional semantic representations used in broadcoverage parsers can both be extended and represented in the context of a CE-based information framework. Focusing on Minimal Recursion Semantics (MRS) [18] used by the ERG parser, we cover: Relating MRS design decisions to CE modeling and reasoning capabilities (e.g. assumptions, constraints) Extending representation of specific linguistic aspects (e.g. tense and aspect) in MRS to take advantage of the rich semantics of a CE model and designing how CE may be extended to represent these aspects ITA White Paper for the 2013 BPP Integrating compositional semantics with constraints deriving from users’ CE conceptual models, drawing on findings from external projects This has the potential to improve fact extraction by incorporating more advanced techniques to map syntax to semantics, allowing a greater range of sentences to be handled correctly; to guide improvements to the semantics of the CE reasoning components, based upon parallel experiences of using MRS in complex NL projects, providing feedback on their robustness and validity; to provide insight into how CE reasoning capabilities of assumptions, constraints, uncertainty can be used in a greater range of situations. Handling Ambiguities We propose research into ambiguities in NL by representing linguistic information (e.g. selectional restrictions) in CE and applying it to disambiguation, using reasoning with assumptions, constraints and uncertainties. [19] Detecting Uncertainty in Sentences We propose to extract uncertainty inherent in unstructured text, allowing these uncertainties to contribute to quality of information of high-value facts generated by reasoning, using: explicit linguistic cues in modal words such as “might”, “believe”, “possibly” context of sentence container (e.g. confidence level from the enclosing report) indirect clues in the language, e.g. hedges, question tags domain knowledge to validate consistency of information Extending Reasoning Capabilities We propose research into extending the CE reasoning component to handle assumptions, hypotheses, uncertainty and constraints based in part on requirements from the Pathfinder UK transition projects. [20] These extensions allow users to express more complex reasoning in CE as well as directly supporting NL processing. Assumption-based reasoning [21] allows exploration of alternative “hypotheses” about facts extracted and inferred high-value information. This will be extended to include numerical uncertainty based upon degree of belief of assumptions, as in Pathfinder, where rationale graphs show assumptions and uncertainties supporting the facts, allowing the user to assess the quality of high-valued information. Generic mechanisms [22] will be used to propagate assumptions and uncertainties, offering a CE-based framework for work on trust and subjective logic where a specific calculus can be provided by the user. Constraint-based reasoning [23] allows users to represent constraints and apply these in the reasoning. Such techniques are used in problem solving e.g. planning [24, 25] and can also guide NL processing in handling ambiguities. Linguistically Motivated CE Enhancements We propose to define CE syntax and semantic extensions to enhance naturalness (e.g. prepositional phrases for expressing relationships between entities), and to increase logical ITA White Paper for the 2013 BPP expressivity (e.g. sentences involving “some”). These have significant benefit, but also a cost: semantic processing becomes more complex and ambiguities may be introduced. Key to our proposal is that studying NL processing gives insight into costs and benefits of CE extensions. Since CE is a form of natural language there are strong parallels in how they may be processed, and understanding NL structures and difficulties in their interpretation is beneficial in determining which language structures should be used to extend CE. We believe our approach is novel: as we extend CE with more complex structures we meet “ambiguity-barriers”, where extension causes potential ambiguities, and we can determine suitable “controls” on the CE specification to avoid them. We will also review expressiveness of other CNLs [31]. CE Model Explanation CE models can become complex, and it is useful to provide users with explanations of the models and rules, so they can more easily extend them to construct new concepts and reasoning. We propose to explore the provision of structured views of the model in a more “tutorial” style by extending rationale mechanisms [26] to analyse logical connections between concepts and rules. Key Deliverables and Planned Research Outputs These are: Contextual scenario Extended fact extraction capabilities from unstructured text based on leverage of external linguistic resources, ambiguity handling and research on deeper semantics Enhanced reasoning techniques based on assumptions, hypotheses, uncertainty and constraints, and CE syntax for expressing this knowledge Model explanation CE extensions Enhanced fact extraction in the Experimental Framework Informal evaluation of benefits of CE for fact extraction Experimentation This section should be up to 300 words in length. Describe how experimentation will be employed as an integral part of the research and how the research will be validated in relevant environments and what experiments will be designed/conducted that can validate the research questions posed by the ITA research. Explain how the ITA Experimental Framework will be used. Clearly link (a) experimentation to the key deliverables described in the “Research Issue/Technical Approach” section; and (b) contributor(s) responsible for leading the experimental work in the “Contributors and their Roles” section. We will contribute to the Experimental Framework CE store [27] with agents, models and reasoning from our fundamental fact extraction and reasoning research, together with research tools e.g. Analysts Helper [28] and review stations [5]. We will broaden our unstructured text ITA White Paper for the 2013 BPP sources from SYNCOIN [32] to include “Hursley reports” from UK Pathfinder projects, and potentially open sources e.g. newsfeed, social media. We will develop a scenario from the problems described above, including intelligence analysis and humanitarian emergency response, incorporated into the ITA-wide scenario. We will use the CollaborativePlanningModel [25,29] and associated tools [29,30] to model military context (e.g. commander’s intent), supporting coalition decision making, and the “Decision-Making to Support Mission Command” theme. IBM UK will lead experiments to assess how CE assists users in the scenario, creating an informal evaluation using personnel representing analysts and coalition partners. Analysts will have information requirements and operational context, e.g. higher-level plan. Unstructured text will be provided from which analysts attempt to extract relevant information. Possible stages include testing the construction of simple reasoning, dynamic creation of NL processing for new source types, explanation of conclusions to partners, utility of structured explanations of the CE model. This will inform our research into how CE might be extended, how more advanced reasoning processes can be understood and how the model can be best presented to users. We will measure success: pragmatically, via experimentation, as the quality of CE facts extracted from a range of sources, the ability for users to understand, judge and communicate results and the facility by which users can construct new domain specific concepts and reasoning. theoretically, via the researching of sound linguistic and semantic models and reasoning, as measured by the acceptance of the work in the linguistic community by readiness of the technology for transition Military and BPP13 Relevance This section should be up to 300 words in length. Describe the military relevance of the research, how it will tackle research challenges associated with Coalition Operations, and how it is relevant to the coalition needs/research challenges for the ITA Technical Areas. We aim to provide theoretical and technical underpinnings to solve the class of military problems described above, applying human knowledge to interpretation of unstructured text (TA6 3a1), reasoning to obtain high-value information and intelligence; exchange, explanation and understanding of the results with other humans and systems within the coalition context. Our research will support intelligence analysis, planning, logistics, resource allocation and has a key role in supporting “Multi-Int” fusion. The MIPS UK ITA transition project applied CE fact extraction techniques to such problems. Addressing a wider range of unstructured sources provides opportunity to apply these techniques in new areas. 1 These (and others like it in this section) are references to the specifically relevant items in the BPP13 call for whitepapers document. ITA White Paper for the 2013 BPP Using a Controlled Natural Language (e.g. CE) to configure NL processing and reasoning facilitates users in applying specialist knowledge to extraction of high-value information thus allowing them to adapt more readily to changing circumstances. This, when coupled with an explanation capability, offers potential for greater understanding between coalition partners about the meaning of the information and how that information is obtained, especially when CE is extended with greater linguistic expressivity and logical power. By situating our work within a contextual CPM-based scenario, we can integrate to BPP13 research involving decision making, as well as external projects such as that involving the NATO tool TOPFAS. There is additional relevance to BPP13: the work provides a comprehensive example of CE in a complex task, motivating and guiding extensions to CE, encouraging wider uptake; the work supports: shared understanding within the coalition, via shared language and models more complex reasoning capabilities via hypotheses (TA6 1d) uncertainty, constraints, rationale, explanation, thus assisting decision making (TA6 1c) research into argumentation using CE to express arguments research into subjective logic and trust using CE to express support for conclusions research into collective cognition using CE to communicate between agents Transition Opportunities This section should be up to 300 words in length describing potential transition opportunities, exploitation possibilities and potential routes to market that may be created from the output of the research. The current P4T2 work, upon which we are building, has already been the basis for successful transition. The execution of these transition projects has shown us many challenges that remain and these have informed our proposals for further research. We anticipate the proposed research will be capable of transition, following a similar path, to more areas characterized in the class of problems described above, including: reasoning and uncertainty handling for intelligence analysis, e.g. Pathfinder transition projects, fact extraction for intelligence analysis, e.g. MIPS transition project; collaborative planning, e.g. NATO CPM-TOPFAS project; intergroup sharing of information on socio-cultural infrastructure, e.g. the CWP proposal New opportunities arise from extend our NL processing capabilities: using CE for NL processing can feed into the research community such as the DELPH-IN consortium, as well as being incorporated into existing NL systems such as LanguageWare and the Watson DeepQA project. ITA White Paper for the 2013 BPP There is potential collaboration with NS-CTA in areas such as text mining and ontologies. We will continue to brief external agencies such as DRDC, Canada. Further transition opportunities include: support for multiple large scale intelligence analysis programs (Boeing) support for knowledge based construction and as an advanced user interface front end for the Integration Visualization Tool (Boeing) support for Knowledge-Base Engineering for knowledge base construction and as an advanced user interface front end (Boeing) transitioning the CE store, incrementing its capabilities and hardening it into a more production-strength asset. (IBM) Transition could lead to new assistive applications and services that support collaborating users in tasks such as intelligence analysis, Multi-Int fusion, collaborative planning, information exchange, in military and non-military coalitions, based upon knowledge and data communicated in structured and unstructured formats, and providing a user interface that more closely aligns the users’ and the systems’ conceptual processing. Dependencies (to be reviewed) This section should be up to 100 words in length identifying the Technical Area (TA) that the proposed research applies and its dependencies on previous and proposed research efforts. Highlight whether this research is a continuation of current ITA research (identify the project/task) or a new effort. Describe the dependencies that this research has with: 1) Current ITA research projects, 2) Other proposed research tasks responding to this Call for White Papers, 3) Any other technical efforts or data collection not included in this white paper. This should not only highlight the linkages but also summarize how this program leverages these other efforts or how it would be affected if other research tasks were not selected. The research is in TA 6, based upon P4T2 research, including the CEStore, with links into the following known proposals, though not critically dependent upon them: “Agile decision-making through diversity models and semantic enrichment” “In-time Assembly and Presentation of Relevant Information in Support of Decision Making” “Decision-Aware Assistance for Analysts at the Network Edge” Inter-agent communication using CE “Agile (Re)Configuration of Services at the Edge with Humans-in-the-Loop” A Framework for Coalition Collective Expressing trust and uncertainty in CE Presentation of arguments using CE Using CE and NL processing for conversational queries Using CE to express trust and to provide ITA White Paper for the 2013 BPP Intelligence rationale for conclusions There is further opportunity to collaborate in areas of distributed reasoning, language and cognition. Contributors and their Roles This section should include a list of the key ITA researchers contributing to this task and should describe their roles and contributions in the proposed research. Clearly show the synergy to be gained from existing or proposed collaborations, including US/UK, academia/industry/government, cross-project, and cross-TA collaborations. You must make sure you have some cross UK/US collaboration, i.e., UK industry and US academia or vice versa in the proposal. Each task should not exceed $400k a year. We don’t release the rates of the ITA partners, but once you’ve submitted the proposal, we will cost it for you and provide feedback on how much the proposal will cost. As a guide for when you are initially putting the proposal together, it should be of the order of 1 full time student, plus not more than 1 person’s effort per year from industry; alternatively a student plus a combination of effort from industry, a post graduate researcher or a professor totalling not more than one person’s effort for a year. Other combinations are permitted of course but tasks that are too big will need to be reworked to come in under the $400k cost. The research team comprises a mix of US and UK, academic and industrial collaborators. The work proposes to extend the current P4T2 research, and so the core IBM UK and Boeing collaborators will continue to be involved in the same roles. IBM UK will continue to lead the project and perform a significant portion of the research on fact extraction and reasoning using CE, the extension of CE syntax and semantics, the building of experimental software and its integration to the ITA Experimental Framework, and the design of the informal evaluation. The Boeing team will continue to provide input on the use of linguistic theory and reasoning and conceptual models for fact extraction as well as focusing on the tasks of detecting uncertainty in sentences and developing the lexicons. We propose to introduce a new partner, Prof Ann Copestake, a leading authority on linguistics from Cambridge University, who researched and developed the language system on which the enhanced linguistic resources from ERG and the Grammar Matrix (as described above) are based. She will provide consultancy and guidance on the use of these resources and will perform fundamental research into the deeper links between syntax and semantics. She is also a key member of the DELPH-IN NL processing consortium, giving us the opportunity to feed back our work on CE to the wider NL community. Prof Copestake will also offer projects based on this work for two MPhil post graduate students per year, to run from November to June, supervised by herself and informally advised (or “uncled”) by members of the team. This will augment the research with relatively short term and focused projects on aspects of CE for fact extraction relevant to the milestones of the proposal, as well as helping to provide experience to the academic community in the techniques being researched. ITA White Paper for the 2013 BPP Resources Please complete the following resources table together with the name and email address of the task leader. The table will be used to calculate the cost of your proposal. You should identify the name and affiliated organisation of each contributor to the task together with the number of days per year they will devote to the research. In the type column, please identify their role in the organisation. Where possible, use the following abbreviations: Academic Principal Investigator PI Academic Post-Doctoral Researcher PDR Academic Post-Graduate Researcher PGR Industrial Principal Researcher PR Industrial Senior Researcher SR Industrial Researcher R Task Leader: Email address: Dr David Mott [email protected] Assume 220 days = 1 Person Year Organisation IBM Boeing University of Cambridge Name David Mott, Simon Laws Stephen Poteet, Ping Xue, Anne Kao Prof Ann Copestake Type PR PR PI Days per Year 177 102 40 Plan for two MPhil postgraduate projects per year supervised by Prof Ann Copestake (expenses coming from the University of Cambridge budget allocation) In the following table, list the contact details for other ITA partners that have expressed interest in collaborating with this proposal but are not to be funded. Note that any Government collaborators (including TALs) MUST be identified. Organisation IBM University of Southampton ARL CMU University of Name Dave Braines, Stephen Pipes, Saritha Arunkumar Paul Smart Cheryl Giammanco YuQing Tang Alun Preece ITA White Paper for the 2013 BPP Cardiff Milestones and Deliverables Please complete the following table with your initial set of research milestones and deliverables for each quarter of the BPP. Quarter BPP13- Q1 BPP13- Q2 BPP13- Q3 BPP13- Q4 BPP13- Q5 Milestone Define motivating scenario, including data sources, overall flow of extraction and reasoning, initial specification of evaluation Establish principles for mapping between the Grammar Matrix (and other resources) and CE Incorporate initial set of linguistic resources (IBM) and lexicon (Boeing) into CE framework Understand initial relations between semantics and the reasoning component Establish MPhil project proposals Define mechanisms for using assumptions for ambiguities, hypotheses and uncertainties Establish mechanisms for extracting uncertainty from sentences Define set of enhancements to CE syntax and semantics Initial update of Experimental Framework BPP13- Q6 Walkthrough of informal evaluation Incorporate advanced set of linguistic resources (IBM) and lexicon (Boeing) into CE framework Incorporate advanced semantic processing into CE framework Deliverable Report: Motivating Scenario (IBM, Boeing) Report: Initial principles for mapping external resources into CE framework (IBM, Boeing) Demonstration: ACITA13, using initial enhancements of linguistic resources and lexicon (All, led by IBM) Report: “Initial review of MRS and its relationship to CE-based reasoning”, (University of Cambridge) Paper: Extracting uncertainty from English sentences (Boeing) Report: CE-based mechanisms for handling ambiguity in NL (IBM) Paper: “Specific representation topics in MRS and CE”. (University of Cambridge) Paper: Extensions to CE (IBM, Boeing) Report: initial feedback on walkthrough of evaluation (IBM) Paper: representing linguistic resources in the CE framework (Boeing, IBM) Demonstration: ACITA14 using rich explanation, more advanced ITA White Paper for the 2013 BPP BPP13- Q7 BPP13- Q8 Techniques for rich explanation of conceptual model Establish MPhil project proposals Further Update of Experimental Framework Preparation for evaluation Undertake informal evaluation reasoning and NL capabilities (All, led by IBM) Report: rich explanation of conceptual models(IBM) Paper: Fact Extraction and reasoning using CE (IBM, Boeing) Paper: “Integration of compositional semantics and constraints deriving from conceptual models (University of Cambridge) Paper: Evaluation results (All, led by IBM) Report: Final report (All) References [1] Mott, D. (2010). Summary of Controlled English, ITACS, https://www.usukita.org/papers/5658/details.html. [2] Sowa, J., Common Logic Controlled English, http://www.jfsowa.com/clce/clce07.htm [3] Mott, d., Braines, D, Poteet, S., Kao, A., Ping, X., Slidesets of Controlled English, June 2012, ITACS, https://www.usukitacs.com/node/2071 [4] Mott, D., Braines, D., Poteet, S., Kao, A., Controlled Natural Language to facilitate information extraction, ACITA 2012, https://www.usukitacs.com/node/2226 [5] Xue, P., Mott, D., Braines, D., Poteet, S., Kao, A., Giammanco, C., Pham, T., McGowan, R. (2012). Information Extraction using Controlled English to support Knowledge-Sharing and Decision-Making. In 17th ICCRTS “Operationalizing C2 Agility.”, Fairfax VA, USA, June 2012 [6] Mott, D., Laws, S., Poteet, S., Demonstration of Fact Extraction using Controlled English at ACITA12, Sept 2012 https://www.usukitacs.com/node/2231 [7] Mott, D., Summary of Communications processing, 08/14/2012, https://www.usukitacs.com/node/2121 [8] Mott, D., Braines, D., Laws, S., Xue, P. Exploring Controlled English for representing knowledge in the Linguistic Knowledge Builder, Sept 2012, https://www.usukitacs.com/node/2231 ITA White Paper for the 2013 BPP [9] Mott, D., Representing Typed feature Structures in Controlled English, 06/01/2012, https://www.usukitacs.com/node/2066 [10] Wordnet, a lexical database for English, http://wordnet.princeton.edu/ [11] George A. Miller (1995). WordNet: A Lexical Database for English. Communications of the ACM Vol. 38, No. 11: 39-41. [12] Palmer, M., http://verbs.colorado.edu/~mpalmer/projects/verbnet.html [13] Ann Copestake and Dan Flickinger (2000) An open-source grammar development environment and broad-coverage English grammar using HPSG In Proceedings of the Second conference on Language Resources and Evaluation (LREC-2000), Athens, Greece. [14] LinGO Grammar Matrix http://www.delph-in.net/matrix/ [15] Ann Copestake, Implementing Typed Feature Structure Grammars, CSLI Publications, 2002 [16] DELPH-IN, http://www.delph-in.net/ [17] Cann, R., Formal Semantics: An Introduction, Cambridge University Press, Feb 6, 1993 [18] Ann Copestake, Dan Flickinger, Ivan A. Sag and Carl J. Pollard Minimal Recursion Semantics: An Introduction, Research on Language and Computation (2005) 3:281-332, Springer 2006 [19] Mott, D., Poteet, S, Ping, X, Hypothesis Support in Information Extraction, Nov 2012, https://www.usukitacs.com/node/2235 [20] SIG day - CNL and Pathfinder https://www.usukitacs.com/sites/default/files/SIG%20day%20%20CNL%20and%20Pathfinder%20v1.1.pdf [21] de Kleer, J. (1986). An assumption-based TMS. Artificial Intelligence, 28:127-162. [22] Gabbay, Labelled Deductive Systems: A Position Paper, http://projecteuclid.org/DPubS/Repository/1.0/Disseminate?view=body&id=pdf_1&handle=eucl id.lnl/1235423708 [23] Constraint-based Reasoning, Freuder, E., Mackworth, A, MIT Press 1994. [24] Tate, A. Representing Plans as a set of constraints – the <I-N-O-V-A> model. In the proceedings of the 3rd International Conference on Artificial Intelligence Planning Systems, May 1996 ITA White Paper for the 2013 BPP [25] Jitu Patel, Michael Dorneich, David Mott, Ali Bahrami, Cheryl Giammanco, Improving Coalition Planning by Making Plans Alive, IEEE Intelligent Systems, 08/14/2012, https://www.usukitacs.com/node/2126. [26] Mott, D., Giammanco, C., Braines, D., Dorneich, M., and Patel, D., . Hybrid Rationale and Controlled Natural Language for Shared Understanding. Proc 4th ACITA London, UK, Sep 2010. [27] Braines, D., CE Store - Alpha Version 2 (software), ITA, Feb 2012, https://www.usukitacs.com/node/1670. [28] Mott, D., The Prolog Analyst's Helper, ITA, Feb 2012, https://www.usukitacs.com/node/1909. [29] Dorneich, M.C., Mott, D., Bahrami, A., Patel, J., & Giammanco, C.,. (2011). Evaluation of a Shared Representation to Support Collaborative, Distributed, Coalition, Multilevel Planning, ACITA11 [30] Mott, D., The representation of logic within semantic web languages, ITACS https://www.usukitacs.com/?q=node/4986 August 2009 [31] P. R. Smart, “Controlled natural languages and the semantic web,” School of Electronics and Computer Science, University of Southampton, Technical Report ITA/P12/SemWebCNL, 2008 [32] Graham, Rimland, & Hall, A COIN-inspired Synthetic Dataset for Qualitative Evaluation of Hard and Soft Fusion Systems: Proc, 14th international conference on information fusion. Chicago, IL, 2011 [33] Sag, Ivan A., Thomas Wasow, and Emily M. Bender.. Syntactic Theory: A formal introduction, Second Edition. Stanford: CSLI Publications [distributed by University of Chicago Press], 2003
© Copyright 2026 Paperzz