the leverhulme trust application form - research project grant · leverhulme research project grant...

Leverhulme Research Project Grant Application Form of 21

The Leverhulme TrustAPPLICATION FORM - Research Project Grant

Applicant: Dr Mateja Jamnik ID/Ref:

Project Title: ARD: Accessible Reasoning with Diagrams

Principal Applicant Details

Submission Date: 30/11/2015Total Requested: 383,728

General DetailsTitle Dr Gender FemaleFirst Name(s) Mateja Date of Birth 05/03/1973Surname Jamnik

Contact DetailsDepartment Computer LaboratoryInstitution Faculty of Computer Science and Technology, University of CambridgeAddress JJ Thomson Avenue

Cambridge

CB3 0FD

Telephone Number

07714765982

Email [email protected]

How is your post currently funded?

University

Date appointed to current post?

01/04/2002

Co-Applicants

Co-Applicant 1Title: DrForename: GemSurname: StapletonGender: FDOB: 01/04/1974Department: School of Computing, Engineering and MathematicsInstitution: University of BrightonAddress: Cockcroft Building

Lewes Road

City BrightonCounty East SussexPostcode BN2 4GJCountry United KingdomEmail [email protected]

Proposal

Long Title ARD: Accessible Reasoning with DiagramsShort Title ARD: Accessible Reasoning with Diagrams

mailto:[email protected]


Main/sub field of study Computing, Science (various)

Start Date 01/05/2016Duration 36

Referees

Referee 1 Professor Alan BundyDepartment : Institution School of Informatics : University of EdinburghPosition: Professor of Automated ReasoningEmail: [email protected]

Referee 2 Professor Peter ChengDepartment : Institution School of Engineering and Informatics : University of SussexPosition: Professor of Cognitive SciencesEmail: [email protected]

Previous and Current Applications

Detailed Research Description

SummaryFile: Detailed Research DescriptionProblem StatementModelling modern high-technology systems is complex and involves multiple stakeholders. As these systems increasingly underpin our everyday lives, and are often safety or security critical, reasoning about correctness is paramount. Thus, modelling and formal reasoning is required in order to convey knowledge unambiguously and correctly. Whilst mathematical modelling adds great rigour, it is opaque to many of the stakeholders. This leads to errors in data handling, delays in product release, and even breaches in consumer privacy. We propose a solution: a new formal diagrammatic approach for developing, debugging, communicating and reasoning rigorously yet accessibly about domain models. Thus, the hypothesis for our project “ARD: Accessible Reasoning with Diagrams”, which challenges the existing symbolic paradigm, is: It is possible to devise an accessible diagrammatic logic for modelling and reasoning in diverse domains. Uniquely, the development of this diagrammatic approach will be guided by extensive empirical studies of what humans understand and find accessible. Whilst pushing forward research into diagrams and logic [1], a major goal is also to bring usable reasoning tools to end-users who need to understand, develop and reason about models of their respective domains.

A Case Study: Privacy Requirements ModellingTo showcase practical relevance, we will apply our research to privacy requirements modelling [2]. This is an important topic from consumer, business and legal perspectives, and involves a wide range of stakeholders [3], including:• software engineers, who must implement associated software systems, • lawyers, who must ensure that data usage is legal, • analysts, who need to understand the data for purposes such as product improvement,


• marketing personnel, who need information for targeted product advertising, and • managers, who need to understand what their teams are doing to ensure privacy is protected.

These stakeholders contribute specialist knowledge in the collaborative act of producing a model of privacy requirements, yet typically have no knowledge of symbolic logics. The diversity of stakeholders and the complexity of the domain make it an ideal area for study, in which we will collaborate with Nokia Networks. Their interest stems from the accessible nature of diagrams, allowing effective and accurate communication between their stakeholders.

Fig. 1 shows a standard symbolically expressed example of a theorem that needs proving to establish that it complies with privacy laws (simplified from Nokia’s privacy model). The same theorem is presented using a concept diagram [4] in Fig. 2. It expresses that when users have consented for their data to be used for both marketing and secondary purposes, the user IDs (UID) from the raw log need to be hashed (i.e., IDs are located and data associated with each is accessed), and their cities of origin must be extracted from their IP addresses to make the data legally ‘clean’. Concept diagrams have an essential hierarchical structure that gives users the ability to abstract, for example, by referring to one diagram from another. This dividing of complexity between diagrams helps to keep each diagram simple and understandable. We have initial

evidence from Nokia that such diagrams are more accessible to stakeholders than their symbolic counterparts [5], and our proposed diagrammatic logic will develop this much further.We will also collaborate with information analysts at Horizons Regional Council (New Zealand), who need models to ensure consistent country-wide application of local government policies, such as those that define planning requirements. Both these organisations are ideally positioned to collaborate because they already use our diagrams to develop models about which they need to reason.BackgroundCurrent Approaches to Modelling Usually, problem models (e.g. Fig. 1), are expressed symbolically using description, predicate or higher-order logics. Such logics are often highly expressive and always provide proof correctness. They support the specification of complex theories with a range of syntactical tools. Unfortunately, their use requires specialist knowledge, rendering them inaccessible to many potential users. Instead, such groups resort to using informal, ambiguous and often differing terminology and representations. Some stakeholders use description logics and their graphical interfaces (e.g., OWLViz [6], UML, existential graphs [7]), but they report that these are either too complex to understand [8], not expressive enough, or are limiting and lack formalisation. We have previously devised a formal syntax of concept diagrams informed by pilot studies, which model complex privacy properties and are anecdotally reported to be accessible [5]. Our proposal builds on this idea, by formalising an accessible diagrammatic logic of concept diagrams and the reasoning tool that uses them, to express and to prove complex properties across a wide range of domains.

Figure 1: A symbolic theorem about Nokia’s privacy model.

Figure 2: A diagrammatic version of the theorem in Fig. 1.


Existing Diagrammatic Logics Recently, diagrammatic logics have emerged that are defined with sufficient mathematical rigour to allow their use as formal tools. Some have been designed with particular application areas in mind such as number theory [9], real analysis [10,11] and education [12,13], or as formal logics such as Venn-II [14], existential graphs [7,15], and Euler/Venn diagrams [16,17], as well as spider [18,19] and concept diagrams [3,4,5]. Spider diagrams are based on Euler diagrams with addition of spiders to represent the existence of elements, analogous to existentially quantified variables. However, they can only represent monadic predicates, unlike concept diagrams which allow for representing and quantifying over monadic and binary predicates. This gives a significant increase in expressive power, a need identified by Nokia. We have already developed a theorem prover for spider diagrams, Speedith [18], which provides valuable lessons for developing the reasoner for concept diagrams proposed here.Uniting Diagrammatic Logics and Cognition It is a long held assumption that diagrams can support humans with logical reasoning, which for weakly expressive logics, is now supported by a growing body of evidence [20,21,22,23]. Diagrams have been found to aid some students with deductive reasoning tasks as compared to standard symbolic logic [20], and Euler diagrams have been shown to increase people’s accuracy when performing syllogistic reasoning tasks [22]. Recent fMRI studies have demonstrated that in the context of reasoning, diagrams provide cognitive offloading and therefore aid cognition, as compared to stylized natural language [23]. These results provide the appropriate foundations for designing an accessible diagrammatic logic that is suitable for real-world modelling and reasoning, and which is based on a user-centred approach.

ObjectivesThis project marries computer science and cognitive science, and a unique aspect is the use of empirical studies to guide the development of foundational approaches to reasoning. Research contributions fall into six objectives within two main streams:Stream 1: diagrammatic reasoning, led by Jamnik,Stream 2: empirical evaluation and layout, led by Stapleton. The objectives are as follows: O1. develop case studies to identify core modelling and reasoning problems [Streams 1 and 2]O2. qualitatively identify effective diagrammatic representations [Stream 2] O3. design an accessible diagrammatic logic for modelling and visual reasoning [Stream 1] O4. develop empirically informed layout algorithms for diagrams [Stream 2] O5. implement a reasoning system for our diagrammatic representations [Stream 1] O6. test and empirically evaluate the accessibility of our system [Streams 2 and 1]

MethodologyO1: Case Study Development [Streams 1 and 2; Advisory Board (AB): Oliver and Bonnington]Working with Nokia Networks and Horizons Regional Council, we will develop a series of case studies based on real world problems and properties that end users must model and reason about. These will be analysed to identify how concept diagrams could be used to represent them and reason about them, and will be checked with end-users to validate that they indeed represent problems as specified. As is typical with diagrammatic logics, reasoning is formalized using an abstract syntax, which describes diagrams. We must therefore identify a range of descriptional choices that must be made during inference tasks. Moreover, when formulating axioms and theorems, as well as applying inference rules, the diagrams themselves must be chosen; for instance, different diagrams can have the same abstract syntax but their topological and graphical properties can vary. We aim to provide deep insights into what makes an effective representation of information in concept diagrams; no one has previously evaluated the impact of descriptional, topological or graphical choices on diagram effectiveness in the context of logical reasoning.Tasks: 1.1 problem collection and representation choice analysis.


Deliverables: a set of case studies with associated reasoning problems; a set of representational choices (for each of the descriptional, topological and graphical categories) that could impact on human cognition when interpreting, and reasoning with, concept diagrams.O2: Effective Diagrammatic Representations [Stream 2; AB: Narayanan]

End-users, when formulating axioms, theorems and proofs, will be perceptually sensitive to their diagrams’ topological and graphical properties. In addition, inference rules (to be identified in O3 and implemented in O5) operate on the abstract syntax, determining descriptional choices. We will evaluate choices from all three categories through a series of empirical studies focusing on stand-alone diagrams and proofs. We will take advice from Prof. Narayanan to identify the choices most likely to impact on human cognition, prioritising these in our evaluation. We envisage adopting either a between or within group study design to collect performance data. Tasks: 2.1 initial empirical tests for descriptional choices, followed up with tests after inference rules are designed; 2.2 empirical tests for topological choices; 2.3 empirical tests for graphical choices.Deliverables: cognitive insights into the effects of descriptional, topological and graphical choices on end-user task performance using concept diagrams; a set of guidelines for each category of choice, so that we can create effective diagrammatic representations; a tutorial on effective diagrams.O3: An Accessible Diagrammatic Logic [Stream 1; AB: Benzmueller]

The diagrammatic logic for expressing the problems from O1 needs to include both diagrams representing problem statements, and the inference rules to be used to prove these in an accessible way. This diagrammatic logic will have essential properties, like soundness, and a rigorous characterisation of the problems that it can solve. The problems arising from O1 will be analysed in order to identify the types of inference rules required to solve them. We will then use the insights from O2 about effectiveness of different types of representation to design the inference rules accordingly; these will operate on the abstract syntax of the concept diagrams like the ones in Fig. 2. We will define the notion of proof and the notion of how a failed proof can help revise or repair the model or the set of available axioms. Finally, we will prove important properties (e.g. soundness) of our diagrammatic logic. We will seek input from Prof. Benzmueller as an expert logician about inference rule selection and design.Tasks: 3.1 identify inference rules for problems from O1 and refine through descriptional tests; 3.2 design inference rules w.r.t. O2, task 2.1; 3.3 prove properties of diagrammatic logic.Deliverables: a suite of diagrammatic problem representations and inference rules; formalisation of accessible diagrammatic logic.O4: Empirically Informed Layout Algorithms [Stream 2; AB: Narayanan]

Two approaches to layout will be taken: contextual layout and independent layout. Contextual layout takes the diagram whose abstract syntax is subject to an inference rule and alters its layout accordingly. This approach is entirely novel and has the advantage of maintaining some of the original diagram’s layout. However, for some inference rules a contextual approach is problematic. For instance, when unifying diagrams, topologies could be incompatible and thus not admit contextual layout. In such cases a new diagram must be laid out from scratch, for which we will develop an independent layout algorithm. Both approaches have their challenges. They must ensure that layouts are effective (informed by O2 and [24,25,26]), produced in reasonable run-time when implemented, and be effective. In addition, they must be proved to produce diagrams with the required abstract syntax. Prof. Narayanan will provide advice on the layout methods from the perspective of human cognition.Tasks: 4.1 contextual layout algorithm; 4.2 independent layout algorithm. Both tasks include proof of correctness and empirical evaluation of effectiveness.Deliverables: layout algorithms that permit cognitively effective visualization of diagrammatic proofs.O5: Implement a Reasoning System [Stream 1; AB: Benzmueller and Narayanan]

Using the accessible representations from O2, along with the inference rules and a formal diagrammatic logic from O3, we will implement an interactive theorem prover for reasoning with concept diagrams. Our layout algorithms for displaying diagrammatic problems (O4) will inform the design and implementation


of an accessible user interface to the proof engine. The expertise and experience of Prof. Benzmueller

(implementation of logical systems) and Prof. Narayanan (human-computer interaction) will be pertinent

here.

Tasks: 5.1 implement inference engine using representation from O2, inference rules from O3; 5.2

design the graphical interface for theorem prover from task 5.1 using layout algorithms from O4.

Deliverables: an interactive, accessible diagrammatic theorem prover for concept diagrams, a tutorial to

practitioners on how to use the system.

O6: Evaluation of the Reasoning System [Streams 2 and 1; AB: Oliver and Bonnington]Finally, we will run empirical studies involving end-users (partly provided by our collaborators) to test

whether our system is sufficiently expressive and accessible to model diverse problems. The results of

these tests will suggest refinements for the system from O5. We will run tutorials and a workshop to

deploy our system in diverse real-world settings, including specifically targeting our partners’ application

areas (i.e. privacy and policy modelling).

Tasks: 6.1 empirical testing of accessibility of system from O5 and general evaluation of the project’s

success; 6.2 refine the theorem prover on the basis of 6.1.

Deliverables: publicity and tutorial material to deliver academic and real-world impact.

Project Management and PlanningWe will achieve our objectives by teaming two groups (for three years, employing a postdoc at each site)

with the interdisciplinary expertise necessary for delivering world-leading contributions to diagrammatic

reasoning and cognitive science. Recruitment of the postdoc and the associated salary scale point will

be consistent with each institution’s policy; at Brighton the scale point will not rise.

Jamnik, at Cambridge, is pioneering the formalisation of diagrams for reasoning, and implementing

logics that model and prove theorems using both diagrams and symbols [9,10,18,19,27,28]. The

significance of her work was recognised by, for example, an EPSRC Advanced Research Fellowship

(2002-2012). Jamnik will capitalise on her expertise to develop the accessible diagrammatic logic and

the framework for reasoning. She will recruit a postdoc in the area of logical reasoning.

Stapleton, at Brighton, is an expert on the empirical evaluation of diagrams [25,29,30] and in devising

automated layout tools [26,31,32]. The importance of her work was recognised by, for example, a

Leverhulme Trust Early Career Fellowship (2005-2007). Stapleton will utilise her expertise to evaluate

the accessibility of different diagrams and reasoning steps along with devising layout algorithms. She will

recruit a postdoc who has significant expertise in conducting empirical studies.

The team will hold regular meetings to ensure research cohesion between their distinctive contributions.

We will be supported by an Advisory Board comprising: Ian Oliver (Nokia Networks, Finland) and

Adrienne Bonnington (Horizons Regional Council, New Zealand), both experts in real-world modelling

and reasoning problems; Christoph Benzmueller (Freie Universitaet Berlin, Germany, and Stanford

University, USA), expert in reasoning with highly expressive logics; and Hari Narayanan (Auburn

University, USA), expert in cognitive science, diagrammatic reasoning and human-computer interaction.

We will hold regular skype meetings with the Advisory Board members, allowing us to gain advice and

feedback on the research, ensuring it is conducted to the highest standards and fully accounts for

practitioners’ needs.


The tasks’ timings are in Table 1 (tasks led by Cambridge are marked by •, Brighton’s by ♦). In stream 2, the empirical studies, which will be designed by Brighton and are subject to ethical approval, will involve large numbers of participants. The studies will take place in dedicated usability labs, ensuring a ‘gold standard’ environment for data collection. The execution of these studies requires a PC on which to collect data and, due to the scale of these studies, participants will need to be recruited at both sites. It would be inappropriate to collect data on the postdocs PCs, which will contain confidential information. We therefore request a dedicated data-collection PC for each site in addition to the postdocs’ PCs. Risks This research has risks: it departs from the established approach of using symbolic logics; no uniform findings for all types of users may exist; and it may prove impossible to devise an accessible and yet sufficiently expressive diagrammatic logic. However, given the current lack of any accessible logical tool for modelling, even a less general (i.e., directed at specific user groups) yet accessible diagrammatic logic would represent a significant contribution. Furthermore, our experience, supported by Nokia Networks, indicates considerable scope for producing a useful diagrammatic logic for complex domains. Since our research intends to make rigorous modelling and reasoning accessible, without it we risk that associated systems will continue to be built incorrectly and, thus, not be fit-for-purpose.DisseminationThe results will impact on computer science, cognitive science, software engineering, data analytics, biomedical informatics and medicine, with dissemination assumed to be an integral part of each objective. We will publish in all these fields, targeting leading journals and conferences (in some we already have a proven track record, e.g., [4,5,10,18,19,25,26,27,28,29]). Reflecting the close collaboration between sites, publications will be co-authored by all four team members; the level of contribution will determine the lead author. We envisage that each task will lead to at least one publication, so our outputs will cover the spectrum of research covered in the project.We will organise an interdisciplinary workshop and tutorials attached to appropriate conferences to foster cross-fertilisation, thus unlocking potential for further research, and to provide training on how to apply our results. All tutorial material will be freely available via a project specific website, and promoted to our collaborators at Nokia Networks and Horizons Region Council, to ensure immediate impact.Lastly, we will proactively publicise our results, using social media, demos at conferences and meetups that focus on data and visualization. Our software will be available via an open source licence, to encourage long-term development and community participation. These activities will ensure that the results have a life beyond the project and will continue to deliver impact in the long term.

Task Year 1 Year 2 Year 31.1: problem collection •♦2.1: descriptional tests ♦ ♦ ♦ ♦

2.2: topological tests ♦ ♦

2.3: graphical tests ♦ ♦

3.1: identify rules • • •3.2: design rules • •3.3: prove properties • • •4.1: contextual layout ♦ ♦

4.2: independent layout ♦ ♦

5.1: implement inference engine • • • • •5.2: graphical interface • • • •6.1: system evaluation ♦ ♦ ♦

6.2: refine theorem prover • •Workshop (w) and Tutorials (t) •♦t •♦t •♦wTable 1: Approximate timings, indicating the proportion of effort devoted to each task.


SignificanceModelling and formal reasoning is required in many real-world domains in order to convey knowledge unambiguously and correctly, yet the practitioners in these domains are typically not logic experts. Thus, currently available symbolic logic approaches are inadequate and often lead to errors in handling data, delays in product release, and even breaches in consumer privacy. We provide a solution: a new formal diagrammatic logic for developing, debugging, communicating and reasoning rigorously yet accessibly about domain models. Thus, the project has the potential to revolutionise approaches to modelling and reasoning across a wide range of sectors.Reducing miscommunication between disparate groups and improving model accuracy removes key risks relating to the processing of data, including safety or security implications. One example is in the area of personal privacy, which is of high importance to commerce and society because of the significant risks and costs associated with data misuse. The scale of potential losses can be extremely large, exceeding 10s of millions of Euros, in addition to the immeasurable financial implications caused by reputational damage. However, at present, companies such as Nokia cannot devote the resources required to deliver (possibly risky) blue skies research, as it requires specialist expertise and is distant from commercialisation. As a result of this project, Nokia will be able to make rapid use of concept diagrams, potentially avoiding costly delays in production. Nokia and all of the end-users of their services will directly and immediately benefit through better specified privacy protection mechanisms. The fact that there are millions of Nokia end-users demonstrates the level and scale of impact that will be achieved by this research on diagrammatic logics. By extension, these types of benefits will arise for all end-users of our research.Our results will also impact on basic research in computer science and in cognitive science. For the first time, a logic will be designed and implemented with human understanding and accessibility at the fore. Our results will give significant insight into human reasoning and which aspects of logic people find intuitive.

References

[1] E. Hammer. Reasoning with Diagrams and Sentences. Notre Dame J. Formal Logic, 35(1):73-97, 1994. [2] H. F. Nissenbaum. Privacy in Context: Technology, Policy, and the Integrity of Social Life. Stanford University Press, 2009. [3] I. Oliver, J. Howse, G. Stapleton, E. Nuutila, S. Torma. Visualizing and Specifying Ontologies using Diagrammatic Logics. 5th Australasian Ontologies Workshop, vol. 112, pages 87-104. CRPIT, 2009. [4] J. Howse, G. Stapleton, K. Taylor, P. Chapman. Visualizing Ontologies: A Case Study. International Semantic Web Conference 2011, Springer, pages 257-272, 2011. [5] I. Oliver, J. Howse, G. Stapleton. Protecting Privacy: Towards a Visual Framework for Handling End-User Data. IEEE Symposium on Visual Languages and Human-Centric Computing, IEEE, pages 67-74, 2013. [6] M. Horridge, OWLViz, https://code.google.com/p/co-ode-owl-plugins/wiki/OWLViz, accessed Nov. 2015. [7] D. Roberts. The Existential Graphs of Charles S. Peirce. Mouton, 1973. [8] P. Warren, P. Mulholland, T. Collins, E. Motta. The Usability of Description Logics: Understanding the Cognitive Difficulties Presented by Description Logics. European Semantic Web Conference 2014, Springer, pages 550-564, 2014. [9] M. Jamnik. Mathematical Reasoning with Diagrams. CSLI, 2001. [10] M. Jamnik, A. Bundy, and I. Green. On Automating Diagrammatic Proofs of Arithmetic Arguments. Journal of Logic, Language and Information, 8(3):297–321, 1999. [11] D. Winterstein, A. Bundy, C. Gurr. Dr Doodle: A Diagrammatic Theorem Prover. International Joint Conference


on Automated Reasoning, Springer, pages 331-335, 2004. [12] J. Barwise. J. Etchemendy. Hyperproof. CSLI, 1994. [13] R. Cox, R. Dale, J. Etchemendy, D. Barker-Plummer. Graphical revelations: Comparing Students’ Translation Errors in Graphics and Logic. International Conference on the Theory and Application of Diagrams, Springer, pages 257–265, 2008. [14] S.-J. Shin. The Logical Status of Diagrams. CUP, 1994. [15] S.-J. Shin. The Iconic Logic of Peirce’s Graphs. Bradford Book, 2002. [16] N. Swoboda, G. Allwein. Using DAG Transformations to Verify Euler/Venn Homogeneous and Euler/Venn FOL Heterogeneous Rules of Inference. Journal of Software and System Modeling, 3(2):136-149, 2004. [17] R. Takemura. Proof Theory for Reasoning with Euler Diagrams: A Logic Translation and Normalization. Studia Logica, 101(1):157–191, 2013. [18] M. Urbas, M. Jamnik, G. Stapleton. Speedith: A Reasoner for Spider Diagrams. Journal of Logic, Language and Information, 24(4):487-540, 2015. [19] G. Stapleton, M. Jamnik, M. Urbas. Designing Inference Rules for Spider Diagrams. IEEE Symposium on Visual Languages and Human-Centric Computing, IEEE, pages 19–26, 2013. [20] K. Stenning, R. Cox, J. Oberlander. Contrasting the Cognitive Effects of Graphical and Sentential Logic Teaching: Reasoning, Representation and Individual Differences. Language and Cognitive Processes. 10:333-354, 1995. [21] K. Mineshima, Y. Sato, R. Takemura, M. Okada. Towards Explaining the Cognitive Efficacy of Euler diagrams in Syllogistic Reasoning: A Relational Perspective. Journal of Visual Languages and Computing, 25(3):156–169, 2014. [22] Y. Sato, K. Mineshima. How Diagrams can Support Syllogistic Reasoning: An Empirical Study. Journal of Logic, Language and Information, 24(4):409-456, 2015. [23] Y. Sato, S. Masuda, Y. Someya, T. Tsujii, S. Watanabe. An fMRI Analysis of the Efficacy of Euler Diagrams in Logical Reasoning. IEEE Symposium on Visual Languages and Human-Centric Computing, IEEE, pages 143-151, 2015. [24] H. Purchase, C. Pilcher, B. Plimmer. Graph Drawing Aesthetics - Created by Users, not Algorithms. IEEE Transactions on Visualization and Computer Graphics, 18(1):81–92, 2012. [25] A. Blake, G. Stapleton, P. Rodgers, J. Howse. The Impact of Topological and Graphical Choices on the Perception of Euler Diagrams. Information Sciences, 2015. Accepted; doi:10.1016/j.ins.2015.05.020. [26] G. Stapleton, P. Rodgers, J. Howse, L. Zhang. Inductively Generating Euler diagrams. IEEE Transactions on Visualization and Computer Graphics, 17(1):88–100, 2011. [27] M. Urbas,M. Jamnik. Diabelli: A Heterogeneous Proof System. International Joint Conference on Automated Reasoning, Springer, pages 559–566, 2012. [28] M. Urbas, M. Jamnik. A framework for heterogeneous reasoning in formal and informal domains. International Conference on the Theory and Application of Diagrams, Springer, pages 277–292, 2014. [29] P. Rodgers, G. Stapleton, P. Chapman. Visualizing Sets with Linear Diagrams. ACM Transactions on Human-Computer Interaction, 22(6), article 27, 39 pages, 2015. [30] P. Chapman, G. Stapleton, P. Rodgers, L. Micallef, A. Blake. Visualizing Sets: An empirical Comparison of Diagram Types. International Conference on the Theory and Application of Diagrams, Springer, pages 146–160, 2014. [31] G. Stapleton, L. Zhang, J. Howse, P. Rodgers. Drawing Euler Diagrams with Circles: The Theory of Piercings,


IEEE Transactions on Visualization and Computer Graphics, 17(7):1020-1032, 2011. [32] G. Stapleton, J. Flower, P. Rodgers, J. Howse. Automatically Drawing Euler Diagrams with Circles. Journal of Visual Languages and Computing, 23(3):163–193, 2012.


Proposal Summary

AbstractAdvances in scientific research are increasingly dependent on the analysis of ever larger datasets. However, the traditional mathematical logics used to represent, model and reason about information are inaccessible to most people. By combining approaches from computer science and cognitive science, we aim to develop a novel and accessible diagram-based logic that is suitable for information representation and reasoning across a wide range of subject areas. The tools we develop will enable better communication and understanding between those who produce the models, and those who use them, and ultimately lead to more robust and effective models to underpin scientific research.

ContextScientific advances today increasingly depend on understanding, manipulating and querying data. In addition, businesses that can capitalise on the value of information will have a competitive edge. This is especially true when businesses collect personal data: their ability to gain insight from it is dependent on compliance with complex, and often international, privacy laws. Typically, data has structure which can be described by formal mathematical statements called axioms. These axioms are typically written in symbolic, mathematical notations, which are inaccessible to most people. This is a serious shortcoming because ensuring the axioms are correct is particularly hard. The correctness of the axioms is important to ensure the data is stored and analysed appropriately. One way of exploring the correctness of axioms is to write down statements that must be true given the axioms. The task then is to reason that the statement is indeed true, or to identify why it is not. Such a reasoned argument is called a proof. Writing proofs using symbolic notations is even harder, if not impossible, for non-mathematically trained people. Thus, accessible techniques are needed that help those users who are manipulating and querying data to create and reason about axioms. Our proposed project addresses the problem of how to produce accessible models (i.e., collections of axioms) and accessible ways of reasoning about them. Our proposed solution represents a paradigm shift: we hypothesize that diagrams can be used instead of mathematical symbols to yield an accessible reasoning system. These diagrams are just as formal as the traditional mathematical approach. A particularly exciting aspect of our project is that it draws on both computer science and cognitive science, to address a long-held assumption that using diagrams makes modelling and reasoning accessible. To do this, we will produce a formal and accessible diagrammatic reasoning system. In doing so, we aim to bring the full communicative benefits of diagrams to the field of knowledge management. This will enable non-specialist users to access and understand data, a process which is vital to scientific advances in the 21st century. In addition to the rapid rise in quantity and availability of data, and the benefits this stands to bring to society if suitably understood, recent research has demonstrated that diagrams bring cognitive benefits over symbolic and textual notations. This cognitive offloading, identified using neuroscience approaches, shows that people find reasoning tasks significantly easier when using diagrams. These results mean that the time is right to design an accessible diagrammatic logic that is suitable for real-world modelling and reasoning.

ObjectivesThis project marries two different fields, computer science and cognitive science. A unique aspect is the use of empirical studies to guide the development of foundational approaches to reasoning. Major research contributions fall within two main streams: Stream 1: diagrammatic reasoning, led by Jamnik, Stream 2: empirical evaluation, led by Stapleton. The objectives are:

• O1. develop case studies to identify core modelling and reasoning problems [Streams 1 and 2]

• O2. qualitatively identify effective diagrammatic representations [Stream 2]

• O3. design an accessible diagrammatic logic for modelling and visual reasoning [Stream 1]

• O4. develop empirically informed layout algorithms for diagrams [Stream 2]


• O5. implement a reasoning system for our diagrammatic representations [Stream 1]

• O6. test and empirically evaluate the accessibility of our system [Streams 2 and 1]

These objectives will be achieved by teaming two groups (for three years, employing a postdoc at each site) with

the interdisciplinary expertise necessary for delivering world-leading contributions to diagrammatic reasoning and

cognitive science.

Significance and originalityThe results will impact on basic research in computer science and cognitive science. The representations and

inference steps used in formal reasoning will be, for the first time, chosen and guided by empirical evaluation of

what people find accessible. At the same time, our work will give us insight into how humans reason and what

representations they find helpful.

Our basic research will enable further scientific advances in areas that need formal reasoning and modelling but

whose practitioners are not logic experts, for example, software engineering, data analytics, biology and medicine.

Current symbolic approaches are inadequate, leading to delays in product release or drug development; errors in

data handling; an increased risk of misdiagnosing – all of which can be costly for society. We will provide end-

users with a new, more accessible way to develop, debug, communicate and reason rigorously about their domain

models. This will improve the model development process and reduce risks and costs.

MethodThe project starts and finishes with both teams working together to collect, analyse and understand case studies

from practitioners, and to evaluate the outcomes of the project. The core of the project for each stream will be

carried out at the respective sites, utilising their specialist and differing expertise. However, close collaboration will

be essential for the success of this interdisciplinary project to ensure that the design of the diagrammatic logic and

the implementation of the reasoning system really do reflect what humans find accessible. Similarly, we must

ensure that the empirical evaluations focus on the features of communication and problem solving that are

important for the practitioners. At the same time, these features must be instrumental in designing the logic and in

implementing the system. The multidisciplinary inter-dependency is very important, as it is a core of the novelty of

this project (it is also the reason the proposed work does not fit the scope of other funding bodies).

Each objective forms a work package that has its own tasks and deliverables as presented in the detailed

proposal. With the steering from the Advisory Board and regular group meetings, we are passionate to deliver the

promised results, as they are unique and important for society.

Why the Leverhulme Trust?Contrary to the usual use of mathematical logics for reasoning, we propose to develop a diagram-based system.

Our original approach attempts to bridge the growing knowledge gap between those who develop scientific

models, and those who use them. This project is based on the widespread, yet untested, assumption that

diagrams are an accessible way to represent complex concepts. The outcomes will be important for the range of

disciplines where scientific advances depend on reasoning with models, such as biology, data analytics, software

engineering and medicine.

Our methodology brings together computer science and cognitive science in a truly novel way, by incorporating

user studies into both the design and testing of our diagrammatic logic. In doing so, we will provide

evidence either for or against the accessibility assumptions around diagrammatic representations, deliver major

advances in the diagrams field, and ultimately produce the first empirically informed diagrammatic logic with real

practical relevance.

There are inherent risks in such an ambitious cross-disciplinary project. The assumption that diagrams are

accessible may be false. Also, end-users’ skills may limit the extent to which they can take advantage of any

logical representation. However, Jamnik and Stapleton have the appropriate complementary skills to mitigate

these. Jamnik is renowned for contributions to diagrams, logic and automated reasoning. Stapleton has expertise

in diagrams and empirical evaluations, and a track record of generating related practical impact. This unique blend of expertise will be essential in realising our vision, and in producing end results of the widest relevance.

Staff Schedule – Principal Applicant

Leverhulme Research Project Grant Application Form Page 19 of 21

Steering Committee/Advisors

Steering Committee/Advisors

An Advisory Board comprising leading researchers and practitioners with expertise relevant to the project will support the project team. Dr Ian Oliver, a privacy expert at Nokia Networks in Finland, will provide access to real-world case studies in the privacy area. Nokia’s modelling and reasoning needs will inform the research from a practical perspective, and is particularly relevant because of their diverse stakeholders, including lawyers, managers, software engineers and information analysts. Adrienne Bonnington, an information analyst at Horizons Regional Council in New Zealand, needs models to ensure consistent application of local government policies, such as enforcing planning requirements, across New Zealand. She is ideally positioned to collaborate on ARD because Horizons Regional Council are already using concept diagrams to develop models that will enable the exchange of cemetery data between councils. Prof. Christoph Benzmueller, from Freie Universitaet Berlin in Germany and also the Heisenberg Fellow of the German Research Foundation, is a world-leading scientist on logic, automated reasoning and artificial intelligence. He has developed a number of automated reasoning systems, most notably a higher-order theorem prover LEO3 and its ontology relative ONTOLEO. He is ideally placed to advise our team on the design of the diagrammatic logic as well as the implementation of the reasoning system. Prof. Hari Narayanan, from Auburn University in the USA and also a former Program Director at the National Science Foundation (NSF), is an internationally leading expert on cognitive science, human-computer interaction, artificial intelligence, and learning science and technology. His advice will be particularly important to the design of empirical tests of candidate representations and layout algorithms with respect to factors that affect human cognition. We will hold conference calls and undertake email communication with the relevant Advisory Board experts depending on the stage and tasks of the project. They will help us ensure that our design, experimentation and implementation are of the highest standard, fully informed by the state-of-the-art, practically relevant in the real world, and that we meet the project’s objectives.

Current Grant Report

Current Grant Report

None. N/A

Finance

Salary Budget Staff Type:Research assistantPercentage Of Time Spent On The Project:100

Year 1 Year 2 Year 3Scale Point Grade 7

point 48Grade 7 point 49

Grade 7 point 50

Basic Salary £37,394 £38,511 £39,685Local Allowance £0 £0 £0National Insurance £2,987 £3,103 £3,226Superannuation £5,983 £6,162 £6,350Overall Total £46,364 £47,776 £49,261

the leverhulme trust application form - research project grant · leverhulme research project grant...

Documents