D8.1 Initial specification of the system
Contents
1. SEVENTH FRAMEWORK PROGRAMME Concept Invention E Theory D8 1 Initial specification of the system Authors Daniel Winterstein UEDIN Ewen Maclean UEDIN Mihai Code scu OVGU Reviewers Lourdes Tavira CSIC Grant agreement no 611553 Project acronym COINVENT Concept Invention Theory Date June 30 2014 Distribution PU Disclaimer The information in this document is subject to change without notice Company or product names mentioned in this document may be trademarks or registered trademarks of their respective companies The project COINVENT acknowledges the financial support of the Future and Emerging Technologies FET programme within the Seventh Framework Programme for Research of the European Commission under FET Open Grant number 611553 Abstract This deliverable presents a specification and architectural plan for the Coinvent system motivated by use cases and a description of a typical user session The architecture described is flexible modular and suitable for distributed computing Analysing target examples has shown that interactive and iterative concept development should be supported The system design uses RESTful http web services to compose building blocks Where possible the building blocks use technology already developed by the Coinvent group HETS HDTP DOL and HR Keyword list specification system architecture Executive Su2. 6 2 Connecting concepts DOL modified o 6 3 Mathematics CASE o 10 604 Music OWL o ss ee a 10 6 5 Other Domains OWL Manchester SyMtaX o 10 6 6 Web service Wrapper JSON o o o e 11 Open Questions and Risks 11 7 1 Software development risks and mitigation o 11 T2 References a ei a ae rr AA a 11 Oo oO A A Appendix 1 Component APIs 12 bb kai Ea eee Go eek me eed Se eee St hae eee 12 A2 Urlstructure ee eaa a i a E E E E E E e E E 12 da o rd eros 13 A ben pad Ge Awe ad 13 ta ea Ens da 13 A E 13 A 7 Cross server requests CORS o e 14 611553 June 30 2014 111 CONTENTS B Component APIs 14 B 1 blend Given a Blend Diagram in Progress compute the Blend Concept 14 B 2 base Given 2 Concepts compute a common base Concept 14 ie OE ack a Gere oe Se 15 B 5 consistency Is a concept consistent o oo 00000 16 B 6 quality How good is aConcept 2 2 0 0000000002 2 16 a Guba bag oa BAe a 17 eis Genin dace ee we See E ede 17 B 8 1 job user_name job_id Show delete job details 18 18 a a e ee Se oe Se eS 19 2 Uta Pane Be ese ee e aloo Geo A 19 EE POR ne tae A Bee ae 19 C 2 Asanon Coinvent developer o o e 19 C 2 1 User A researcher with ontology experience
3. 19 a a a a ee 20 Meh Bete ead toe be eee bee eee 20 C 3 1 Domain Mathematics o 0020000000000 20 C4 Domain Music ee 21 C 4 1 OtherDomains 2 2 000000 000000 22 23 iv June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT 1 Overview This document presents the archictectural design for the integrated Coinvent system Please first read 1 The Glossary D 2 The requirements user stories C which motivate this design 2 A Typical Coinvent Session Coinvent covers several domains and use cases The following description may be taken as typical of planned usage but not normative In a typical session the user browses and selects two concepts from the rich background of domain concepts A blended concept is then developed which may be automatic or interactive Examples of the concept are generated again either automatically or interactively depending on the capability of the automatic software Finally the concept s quality is evaluated based on the examples This is illustrated in the UML sequence diagram shown in Figure 2 Coinvent Typical Session User WebApp file base blend consistency ljob weaken model quality 4 List Select 2 Concepts y Load Compute Common Base gt Base Concept mappings Compute Blend B
4. Cadence sequence progression of chords which concludes a section of music E g fifth to root D G in key of G Colimit Pushout A category theory term which can be applied to blending The least specific specialisation which combines two theories with a common base Concept No one definition but probably a Theory Conservativity conservative extension Safely extending a theory by adding statements about new constants which do not affect the old statements See https en wikipedia org wiki Conservativity_theorem DOL HETS language for describing the relationship between theories Can work with any ontol ogy language So DOL does not itself describe theories Entity Individual GCT Generalised Chord Type e g 0 0 4 7 is C major if in the key of C HETS the Heterogeneous Tool Set A meta logic system which can map between logics and theorem provers 611553 June 30 2014 23 D8 1 Initial specification of the system Idiom a Theory in music Interpretation a symbol mapping between two theories with the following proof obligation the target theory must extend ie logically entail the source theory Most likely the sentences from the source theory are just included with symbol renaming into the target IRI International URI URI which can use unicode A CURIE is a short fragment which with extra context can be resolved into a URI e g like a relative url Mapping signature morphism how to translate s
5. typically modeled in ASP 2 4 5 Model Example Finder Given a Concept find examples models This is a key part of creative blending especially around evaluating a concept It is noticeable that when people learn and evaluate concepts they often do so via examples The meaning of example is domain specific In mathematical theories an example is a model which is itself a theory E g in the complex numbers case study c f nttps github com coinvent coinvent tree master HETS complex_numbers an example is the refined theory where i 1 has been added to provide a constructive formula for the existential axiom V vectors x y 4 vector z x y z note in the dol file this existential axiom is implicit in being a total function In musical theories an example is a piece of music conforming to the theory The initial implementation will be manual Automated implementations will often be domain specific e g we may use a Hidden Markov Model based music generator and may not be possi ble in all domains For relatively simple theories e g small models to finite first order theories model finders such as MACE can be used possibly via HETS 4 6 Consistency Concept Correctness Is it consistent Blending can lead to inconsistent concepts HETS has support for using various theorem provers to check for consistency However detecting inconsistency is not always easy In the complex numbers examp
6. would most naturally be developed in javascript perhaps using a test runner Indeed the test scripts we develop as part of software QA will provide templates for scriptable use Where steps involve systems such as the interactive theorem prover Isabelle it is an open question how we script such systems within Coinvent HETS has some support for this call hets I and enter help although the HETS API does not expose it Software Wrapped as a Server Calculation software such as HDTP will be incorporated into this framework using a server which wraps the low level software E g the other components connect to the HDTP server over http and the server manages calling HDTP itself Top level Stateful Low level Stateless Calculation software will be stateless This is simpler and avoids tying the low level components to the bigger system In places it is necessary to page through results E g HDTP can produce multiple outputs for some inputs The first call would return one result then a mechanism is required to access the 2nd or a 3rd result This will be handled via input flags and simply repeating the calculation asking for more outputs i e iterative deepening The overall system will be stateful because being stateful is the most natural way to support several required features e Slow tasks where a job is started runs finishes later e Interactive UI where the user has a fixed reference point to view a
7. A number The 0 1 scale is recommended for most metrics B 7 file Store Concepts and Blend Diagrams Must provide save and load over http Note that most of the other components do not depend on the file component They can work with concepts stored on any server and identified by uri The User Interface Default implementation file system based with git support allowing for Ontohub integ Default end point http coinvent soda sh 8400 file user_name Load Parameters Use the path i e the slug to specify a file using the convention files user_1 E g http coinvent soda sh 8400 files alice stuff alices_boat dol would fetch the file stored at base_dir alice stuff alices_boat dol Note that Ontohub uses the convention repository path_to file The convention he Response the file Save Parameters Use the path i e the slug to specify a file Use http PUT or the parameter value pair action save The post body contains the text to save Response the file A save or a delete triggers a git commit and push Distributed git repositories cany be linked together and in particular OntoHub uses git based storage This allows for connection with OntoHub Delete Parameters Use http DELETE or the parameter value pair action delete Response just the envelope indicating success or failure B S job List open tasks Default implementation Coinvent Integration Ul module 611553
8. June 30 2014 17 D8 1 Initial specification of the system Default end point http coinvent soda sh 8400 job user_name Parameters status string open closed Optional filter cursor url For requesting follow on results Response cargo jobs id string status open closed B 8 1 job user_name job_id Show delete job details Default implementation Coinvent Integration Ul module Default end point http coinvent soda sh 8400 job user_name Use the path i e the slug to specify a file E g http coinvent soda sh 8400 job alice 1234 would fetch details on job id 1234 assigned to Alice If a job id is not specified this endpoint will list jobs see above action delete Use this or the http method DELETE to request cancellation of a jol Response cargo job id string status open closed This may optionally provide more information about the job such as a progress update C User Stories This appendix presents the motivating requirements for the integrated Coinvent system It does so via the agile method of user stories requirements are driven by example use cases of who needs 18 June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT it and why C 1 Asa Coinvent Researcher C 1 1 User Logic researcher Task Explore creating and using blends Needs e Specify co
9. June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT User A reader Wants To be entertained or stimulated Success criteria Producing poems that are rated as interesting either by literature critics or a more general readership Priority Low This is an accessible domain for technology testing and it fits with the research interests of Goldsmiths group However there is considerable extra work required for software to go from a blended concept metaphor to a complete poem Domain Recipes User An amateur cook Wants To create a new recipe or meal plan by blend ing ideas Success criteria Some new and tasty recipes Priority Low This is an accessible domain for proving the technology However there is so much cooking material online that a new resource would not be very valuable D Definitions Glossary Collecting technical terms in use across the Coinvent project Some of the terms marked as equiv alent here do have different meanings in certain domains but are used equivalently within the Coinvent space ABox assertions about an individual describes an instance See TBox ASP Answer Set Programming a class of automated reasoning systems for hard boolean satisi fiability searches Base A common shared starting theory Blend Any theory created by merging two parent theories Might be a colimit but needn t be Blendoid a Theory created by blending
10. These tasks will create jobs which run on the server until complete The job component provides an API for fetching information on the state of the job queue and a specific job This allows the web portal or other user interfaces to see the state of a blend and a way to report on any errors 5 Architecture 5 1 Independent components linked via http APIs Requirements 1 The components of the system may be developed in different languages 2 The system should be easy to maintain and extend 3 The system should support both interactive use and repeatable scripted use Requirements 1 and 2 above suggest a loose coupling between components The use of http based APIs is an established way of achieving this In particular http based APIs with a REST like setup i e the url follows a simple readable structure and using JSON to encode data are now becoming the standard for modern software development JSON will provide a wrapper for API level information JSON is not part of the file format for Blend Diagrams Blend Diagrams are written in DOL OWL or CASL and sent within a JSON packet as JSON strings An added benefit of this architecture is that it provides flexibility regarding the hardware in frastructure The components of the Coinvent system may be run either on a single server or across multiple servers Supporting Alternative Component Implementations To provide flexibility especially across varied domains the ar
11. call response would timeout We therefore adopt an actor queue based pattern see https en wikipedia org wiki Actor_model Each top level component is an actor which can send and receive messages to other actors Each API request is either fast or slow 1 Fast requests get an immediate result within the http response 2 Slow requests send an immediate receipt response with a job id using http code 202 ac cepted for processing followed later by a result which may indicate failure This later message is either pushed via a callback or pulled by polling 5 4 System Stack Diagram Default User Interface AJAX jQuery underscore templates Web Browser Java web service wrapper HDTP Files HETS Server SWI Prolog Git Theorem Provers OS Linux Ubuntu 1 The HETS server provides an http API This is lower level than the Coinvent API 2 HDTP might be re written by Martin MAQhrmann to use a different backend 3 Git integration provides OntoHub integration without a hard dependency 8 June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT 6 Languages File Formats Coinvent will focus on human readable text file formats A couple of formats for concepts will be supported with the DOL language used to describe the connections between concepts 6 1 3 Levels of languages software meta logic domain logic 1 Th
12. cation but it does not have to be A Blend Diagram in Progress is simply a Blend Diagram where any part may be missing or incomplete and Concepts do not have to be consistent 4 Components A Coinvent system will be made up of the following components A component can be manual which means a human being provides the result using a web interface 4 1 User Portal Web App An interface through which the user drives the system This will be a web app i e a browser based ajax app A default user interface UD will be built on established toolkits including Bootstrap jQuery and underscore js As the project progresses other domain specific UIs may be developed e g a music viewer and player or a child friendly UI if we cover the fictional beasts domain The architecture setup deliberately keeps the UI separate accessing the processing components via the APIs so that other Uls can be built and by anyone Default implementation To be developed 4 2 Blend Concept Blender Given a Blend Diagram in Progress compute the Blend Concept Default implementation HETS As well as producing blends this component also outputs proof obligations 611553 June 30 2014 3 D8 1 Initial specification of the system The Blender does not have to compute consistency This is the job of the Concept Scorer The HETS API is currently still in development and improvements are needed before it fits all requirements Speci
13. chitecture is organised into separate components This document describes default implementations for each component However any component can be provided by an alternative implementation Any system which fits the API for a Coinvent component slot can fill that slot We envisage certain components will indeed have multiple implementations HETS and HR3 are both candidates for the blend and consistency slots The model slot will benefit from domain specific implementations e g a music generator 6 June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT Component Setup With the architecture supporting different implementations for components the system requires a setup describing which implementations to call Each component has a default implementation which will be specified in a config file The Social Creativity strand of the Coinvent project requires that different setups can be run and interact In order to support this from within one server we allow that a Blend Diagram in Progress can carry meta data about which component implementations to call if different from the defaults This meta data might include implementation specific configuration parameters Scripting the system Requirement 3 relates to research users who need to run repeatable concept development sessions This requirement is met in this architecture by scripts which drive the API Such scripts
14. d by HETS We will adopt it as the language Coinvent should use for mathematical concepts 6 4 Music OWL The music team are investigating the use of OWL by producing a worked example of cadence blending It is anticipated that musical idioms will comprise both rules e g constraints on a cadence and statistical parts e g hidden markov models 6 5 Other Domains OWL Manchester Syntax The established OWL description logic via the Manchester Syntax should be used for other domains 10 June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT 6 6 Web service Wrapper JSON The web service components see below will use JSON as a black box wrapper for web services to talk to each other Le JSON is used to encode commands such as blend these concepts and describe results such as success or failure due to timeout JSON is the best choice here because it s a universal standard Low level components e g HDTP will not take in JSON That would be done by a web service wrapper which then calls HDTP 7 Open Questions and Risks Many development questions remain open at this stage in the project Notable open questions are 1 How to generate examples in the different domains 2 How to use 3rd party systems such as the Isabelle interactive theorem prover within Coin vent 3 How to weaken theories to resolve inconsistencies 4 What f
15. developing blend The state would be handled at the top level The components HDTP HETS as a colimit calculator etc are used in a stateless manner 52 Manual Interactive Mode The goal of the Coinvent system is to be semi automatic Certain components such as evaluating musical quality are best handled manually at least in this project There are also components 611553 June 30 2014 7 D8 1 Initial specification of the system such as consistency checking where a fully automatic solution is desirable but for mathematically difficult theories this may not be possible We therefore support a manual interactive mode for each component This works as part of the actor model Certain actors correspond to systems e g the hets user connects to the HETS system Other actors correspond to people For these actors the UI will show a list of open job requests sent to them and an interface for responding to a job The user is then free to call on any outside tools they wish e g working via another theorem prover in order to complete the job These responses will be stored so that sessions can be re played 5 3 Actor Queue Pattern Requirements 1 The components of the system call on each other to perform tasks 2 but these tasks may be slow to complete Since tasks may require human input eg a proof in an interactive theorem prover like Isabelle a task could even take days This means a standard
16. e of a uri by a json or dol file ending sentence type A specific sentence within a Concept Either the sentence itself or a DOL annotation labelling that sentence All inputs are of course sent URL encoded A 4 Common Outputs With the exception of files all outputs are in JSON and have a common envelope A 5 Envelope Each response has the same top level structure success boolean usually true cargo the meat of the response or null if it is a slow asynchronous request jobid string The job id for slow requests messages string an array of notifications for the user usually null cursor uri the next page in a series usually null A 6 Authentication none No authentication is required at this stage in the project 611553 June 30 2014 13 D8 1 Initial specification of the system A 7 Cross server requests CORS Cross server calls where a webpage hosted on one server makes an ajax request on another server are supported via CORS This means they just work although CORS may not work with older browsers B Component APIs B 1 blend Given a Blend Diagram in Progress compute the Blend Concept Default implementation HETS Default end point http coinvent soda sh 8400 blend hets Parameters lang owl casl inputi concept input2 concept base concept base_input1 mapping from base to inputl base_input2 mapping from base to
17. e outer language son is used to connect software components E g specifying software inputs and callbacks reporting on job status etc 2 The meta logic language DOL is used to describe the mappings between concepts 3 The domain logics OWL CASL are used to describe actual concepts 6 2 Connecting concepts DOL modified DOL Distributed Ontology Modeling and Specification Language is a language for describing how ontologies connect DOL is currently in development by the OntolOp group It is used by the HETS system This integration with HETS makes it a good choice for Coinvent DOL is both larger than we need and does not provide a couple of features we require So we specify a modified version of DOL We will work with the DOL team aiming to converge on a true subset of DOL DOL subset Supported Language Terms Coinvent does not need all the features of DOL and so we specify a reduced subset of DOL which should be used Sticking to a smaller set reduces the learning curve for new Coinvent users Briefly the supported symbols are logic Specify the Concept language e g logic OWL ontology Start defining a Concept end Finish defining a Concept view Specify a mapping between Concepts e g view MyView A to B with Introduces a symbol mapping to rename symbols from a theory A typical use in Coinvent would be as part of a combine statement combine Compu
18. ect Success criteria Developers from outside the project produce projects which may be just proof of concept scratch projects using Coinvent Priority For Year 3 C 3 As an end user C 3 1 Domain Mathematics User Coinvent researcher Task Explore the blending of mathematical concepts Use case Define concepts Pick 2 concepts as input and get back a blended concept Further develop the blend to resolve quality issues such as Further develop the blended concept to create interesting outputs possibly models examples Requirements 1 A format which can define mathematical concepts 2 Specify two theories and blend them to produce a new concept 3 Weaken input concepts to remove inconsistency and strengthen output blends to reach interesting results 4 Given a concept check it for consistency 5 Given a concept produce examples a model 6 Given a concept identify the unresolved questions in the blend for complex numbers the missing formula x y 7 Solve the unresolved questions in the blend adding axioms and resolving any proof obliga tions Note that some steps may have to be manual or semi automated and so it is a requirement that the system supports interactive reasoning Success criteria It works Priority High This is a project deliverable 20 June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT User No
19. f computing colimits itself therefore it requires a feedback loop with HETS In case the colimit is inconsistent the TW analyzes the input theories and the inconsistencies in the blend and weakens the theories based on this information by removing sentences from the input theories This removal will be based on heuristics known from amalgam reasoning 4 The whole procedure is repeated until a consistent colimit is found For the implementation of the TW we are currently investigating using an Answer Set Pro gramming ASP approach calling on online ASP solvers This allows us to i rapidly implement prototypes ii make use of the highly optimized search problem algorithms that drive modern ASP solvers 2 and iii use the advantages of online ASP solvers like oclingo 1 that will re use partial solutions of earlier weakenings Towards this a weakening operation is modeled as sequence of theory transitions based on atomic operations that each affect the different types of sentences eg SubClassOf EquivalentTo ObjectProperty etc for OWL within theories Per ceiving the theory weakening as a sequence of theory transitions also allows us to exploit the iter 4 June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT ative problem solving capabilities of modern ASP solvers and to re use coding strategies known from other kinds of state transition based problems that are
20. fic open issues are logged here https github com coinvent coinvent issues labels HETSkpage 1kstate open 4 3 Base Given 2 Concepts compute a common base Concept This component finds a common base theory and mappings to the input concepts That is it fills in the bottom half of the blend diagram The common base is not always unique there may be a choice of several A common base and mappings defines an analogy between the concepts This component does not itself calculate the blend However if a push out based blend component is used as we envisage then the base and mappings do uniquely determine the blend concept Default implementation HDTP which works via anti unification Required work HDTP currently uses a custom Prolog based format To plug into Coinvent or HETS without Coinvent a translation layer will be developed so that HDTP can take in OWL or CASL Note HDTP will remain stateless It will not itself manage sessions or run a web server 4 4 Weaken Amalgam Finder Given an inconsistent Blend Diagram weaken the inputs Initial implementation manual Since the colimit operation and blending in general can generate inconsistent blends it may be necessary to weaken the input theories until a consistent blend is found The Amalgams team will lead on this exploring the generalization operation for amalgams described in 4 and developing a stand alone Theory Weakener TW The TW will not be capable o
21. inputl Response cargo blend concept which is a blend of input1 and input2 input1_blend mapping from inputi to blend input2_blend mapping from input2 to blend B 2 base Given 2 Concepts compute a common base Concept Default implementation HDTP Default end point http coinvent soda sh 8400 base hdtp Parameters lang owl casl inputi concept input2 concept base concept base_input1 mapping from base to inputl base_input2 mapping from base to inputl cursor url For requesting follow on results 14 June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT Response cargo base concept which is a common base for inputi and input2 base_input1 mapping from base to inputl base_input2 mapping from base to input2 B 3 weaken Given an inconsistent blend diagram in progress weaken the con cepts This is the key method for Amalgams Default implementation Manual Amalgams later in the project Default end point http coinvent soda sh 8400 weaken user_name Parameters blend_diagram A Blend Diagram in Progress cursor url For requesting follow on results Response cargo A weakened blend diagram Open question What information should be passed to weaken to guide it For example it might take in line numbers or DOL labels marking sets of inconsistent B 4 model Give
22. le it has been found to require manually guided proofs Any proof of an inconsistency of the blend can give hints about which axioms to look at namely the axioms used in the inconsistency proof This component should therefore provide feedback where possible to enable a workflow enriched with vision and debugging techniques 4 7 Quality Concept Scorer How good is a Concept A key part of creativity is judging the quality of the outputs discarding low value concepts and selecting valuable ones How to do this is one of the research questions this project will explore HR has automatic scores for detecting interesting concepts which it would be interesting to incor porate Default implementation Manual 4 8 File Concept Store Store Concepts and Blend Diagrams Must provide save and load over http Note that most of the other components do not depend on a dedicated storage component They can work with concepts stored on any server as long as the format is correct and the concept 611553 June 30 2014 5 D8 1 Initial specification of the system can be identified by url The User Interface does require a file storage component to store blend diagrams as they re modified Default implementation File system backed By adding git to the file system we can provide integration with Onto hub 4 9 Job Provide meta data on slow tasks As discussed in 95 some of the system tasks are not suitable for a fast blocking API
23. lend Concept Evaluate consistency Status nconsistent Solve inconsistency 2 Weaker Concepts Create example gt T Examples of Concept T Concept and Examples gt T Quality score 4 Save gt User WebApp file base blend consistency Ijob weaken model quality www websequencediagrams com Figure 1 UML sequence diagram 611553 June 30 2014 1 D8 1 Initial specification of the system 3 Core Object The Blend Diagram in Progress The object type at the heart of Coinvent is the Blend Diagram The specification below is in tentionally broad to support several use cases The Blend Diagram is illustrated in the flowchart diagram shown in Figure 2 Metadata gt gt x weakening gt gt t Symbol mappings Figure 2 Flowchart diagram for blending Crucially we must support blend diagrams which are works in progress The high level process of developing a new blended concept is 1 Pick two concepts to blend 2 Compute a common base concept 3 Compute a blend 4 Evaluate the blend for consistency 5 Which may lead to changes in either the input concepts such as logical weakenings to remove the conflict This process of evaluate modify could be iterated several times 6 Evaluate the blend for value perhaps by ge
24. mmary The Coinvent software will produce novel ideas through unfamiliar combinations of familiar ideas a difficult computational task We aim to develop a computationally feasible cognitively inspired formal model of concept creation drawing on Fauconnier and Turner s theory of conceptual blend ing and grounding it on a sound mathematical theory of concepts The software must support different domains and different types of user The key target do mains are mathematics and music but it should be applicable to a wide range of domains beyond those We have mapped out different use cases and analysed the requirements that stem from them By exploring examples in different domains we have concluded that a high level of flexibility and user control will be required This design allows for that accordingly A mathematical case study where the complex numbers are invented was particularly informative in developing the requirements this case study has been submitted to the C3GI workshop for publication The use cases have emphasised the importance of the model generation and evaluation com ponents as key parts of the system alongside the concept blending engine We propose a modular system design which supports the flexibility necessary to cover the varying domains The modular design is also a good basis for coordinating the work of the dis tributed teams involved in the project Recent trends in software development have seen the grow
25. n a Concept find examples Default implementation Manual Default end point http coinvent soda sh 8400 model user_name Parameters lang owl casl concept concept cursor url For requesting follow on results Response cargo models concept 611553 June 30 2014 15 D8 1 Initial specification of the system B 5 consistency Is a concept consistent Default implementations HETS for automated consistency checks Manual where HETS fails Can we include semi automated consistency proofs e g using Isabelle Default end point http coinvent soda sh 8400 consistency hets Parameters lang owl casl concept concept goal sentence For focused testing Response cargo result boolean inconsistent sentence unknown sentence B 6 quality How good is a Concept Default implementation Manual HR has automatic scores for detecting interesting concepts which it would be interest to explore Quality scoring will most likely work on examples rather than on the concepts directly Default end point http coinvent soda sh 8400 quality user_name Parameters lang owl casl concept concept models concept metric string Optional name of a metric to score against Response cargo 16 June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT score number
26. n mathematical Analyst e g someone who would normally use Excel Task Quantitative analysis of data Wants To meaningfully analyse some data using bayesian math ematical models Specific use case Interpret Twitter conversational data on a topic in terms of types of person and types of conversation Needs e A friendly UI e To create a model by blending pre defined models e To understand what they ve made without understanding the source code i e a clear natural language summary of the model To apply the model to the data e And get back results they can understand e g pie charts and time series charts with meaningful labels Success criteria It works and is adopted for use by a researcher or analyst outside the project Priority Medium This is an achievable valuable goal within the important mathematics thread C 4 Domain Music User Coinvent Researcher Task Explore the blending of musical idioms Use case Define idioms Pick 2 musical idioms as input and get back a blended idiom Given an idiom evaluate 1t by creating examples and subjectively assessing them Requirements 1 A format which can define musical idioms Open question What levels of structure does this have to capture 2 Optional The ability to statistically analyse a body of example music as part of fleshing out a musical idiom e g creating an HMM for use in generation 3 Blend idioms to produce new idioms This requires support for
27. ncepts e Create blends e Evaluate blends this is domain specific see specific domains below e Reliably replay a session of concept blending plus associated concept development such as theorem proving or example music generation Success criteria The software supports research leading to papers on blending which extend the state of research Priority High C 1 2 User Social creativity researcher Task Explore multiple users agents creating and using blends Needs e As for Logic Researcher plus e Agent based system to model and explore social interactions e Setup different starting models for agents Priority This is a key strand of the project plan For Year 2 3 C 2 Asa non Coinvent developer Later in the project we aim to run Coinvent hackathons where other developers use Coinvent to build things C 2 1 User A researcher with ontology experience Task Explore concept blending Needs Same as for a Coinvent Logic Researcher plus a user friendly Ul and documentation 611553 June 30 2014 19 D8 1 Initial specification of the system C 2 2 User A developer Task Making a creativity tool for a domain of their choice e g one of the domains considered below Needs e The system must be flexible enough to support e Modified back end components e Other user interfaces built on top of the API e Enough documentation to support 3rd party developers including a Hello World example proj
28. nerating examples models and evaluating them 7 This evaluation may lead to modifications and so the whole process is iterative Hence we talk of a Blend Diagram in Progress to allow that blends may be developed step by step and iteratively Concepts are defined in CASL or OWL Mappings between concepts are defined in DOL Metadata is also defined in DOL The Blend Diagram may be a single file or several linked files using DOL s support for urls A Blend Diagram consists of 1 A Blended Concept 2 June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT 2 A set of Input Concepts typically 2 3 Optionally weakenings of the Input Concepts E g a mapping Concept WeakerCon ceptl by dropping symbols or sentences 4 A Base Concept which is a common base for all the Input Concepts If there are weaken ings then the Base is a base for the weakened concepts not the originals 5 Mappings from the Base Concept to each of the weakened Input Concepts 6 Mappings from each of the weakened Input Concepts to the Blended Concept 7 Optional metadata about each of the Concepts such as evaluation scores proof obligations or relationship to other Concepts 8 Optional metadata about the system setup such as preferred component setup The Blended Concept may be a colimit but it does not have to be The Base Concept may be found via anti unifi
29. ormat is best for musical idioms 7 11 Software development risks and mitigation Much of the Coinvent system builds upon existing components HETS HDTP DOL or can use standard software and approaches the UI These components therefore have a relatively low risk Though note that all the base components require some development to meet the needs of this project Higher risk components are model and weaken How these functions can be implemented is a research topic The problems are deep and no readily adaptable solution exists The team do have relevant expertise in techniques to tackle these tasks but as with any open ended research task there is a high level of technical risk To mitigate these risks the system design allows for components to be fulfilled by manual interactive input This means work on automated solutions to model and weaken is not a blocker for use of the system 72 References CASL user manual http www informatik uni bremen de cofi wiki index php CASL_user_manual and reference manual http www informatik uni bremen de cofi wiki index php CASL_reference_manual DOL https github com tillmo DOL 611553 June 30 2014 11 D8 1 Initial specification of the system HDTP http link springer com chapter 10 1007 2F978 3 642 54516 0_7 HETS http www informatik uni bremen de agbkb forschung formal_methods CoFI hets OWL Manchester Syntax http www w3 org TR owl2 manches
30. t The default file store also fits into this pattern with component files E g the user Alice s houseboat file could be http coinvent soda sh 8400 file alice houseboat omn 12 June 30 2014 611553 D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT A 3 Common Inputs Concept type Concepts can be provided as the source text itself or as a uri for a file which contains the source text A DOL file may contain more than one concept To reference a concept within a file we follow the DOL MMT Ontohub convention and write file_uri ontology_name Mapping type Mappings are provided either as 2 o9 99 99 1 JSON maps e g sun nucleus planet electron 9999 2 DOL fragments using only the inner part of the DOL mapping e g sun gt nucleus planet gt electron BlendDiagram type A packet of data comprising the Concepts and Mappings for a blend di agram in progress The component Concepts of a BlendDiagram use the names base blend inputl input2 and the Mappings base_inputl base_input2 inputl_blend input2_blend If weakenings are used then these Concepts are names weakinputl weakinput2 and weak base with corresponding Mappings Can be provided as the source text itself or as a uri for a file containing the BlendDiagram Can be in JSON or in DOL identified in the cas
31. te the colimit e g ontology B combine I1 I2 hide Used in a mapping to drop a symbol ee 6 6 List separator 611553 June 30 2014 9 D8 1 Initial specification of the system Part of a mapping e g zero gt 0 predicate value An annotation on a sentence label A line label Starts a comment implied Marks a proof obligation i e a statement which is required to be true but has not been proved DOL Extra Conventions and Language Terms There are some features Coinvent needs which DOL does not yet provide see the project s GitHub issue tracker https github com coinvent coinvent issues labels DOL 2F file formats Where possible we fit these within the current DOL specification by specifying non standard annotations 1 Dropping symbols from a Concept hide can be used when defining a mapping to drop sentences as well as symbols 2 Marking inconsistent sentences by the annotation inconsistent 3 Marking relative importance of sentences by the annotation importance X where X is a number in the range 0 1 The is interpreted by the weaken component How this is interpreted is not fixed but it may be interpreted as a probability that the sentence holds 4 Evaluating Concepts That is adding metadata which describes how good a concept is Format TBD 6 3 Mathematics CASL CASL is the first order logic language use
32. ter syntax ASP 1 M Gebser et al Engineering an Incremental ASP Solver In International Converence on Logic Programming 2008 no 1 2 M Gebser et al Answer Set Solving in Practice Morgan and Claypool 2012 3 M Gelfond and V Lifschitz The Stable Model Semantics for Logic Programming In Proceedings of the International Conference on Logic Programming ICLP 1988 4 S Ontan and E Plaza Amalgams A formal approach for combining multiple case solutions In ICCBR 2010 A Appendix 1 Component APIs A l Common This document uses JSDoc to describe input and output types e g string would mark an op tional string For describing urls variable is used to mark a variable within a url A 2 Url structure The default server is coinvent soda sh and the default port is 8400 For each component we provide a default implementation and these follow a common url pattern Other implementations are possible and may not follow the pattern http server port component actor This means we can have multiple different instances of a component e g http coinvent soda sh 8400 blender hetsandhttp coinvent soda sh 8400 blender hr3 The actor may refer to a piece of software e g hets or to a user which allows that any component function can be fulfilled manually by a human being E g This structure anticipates multi agent setups which will be wanted for the investigation of social aspects later in the projec
33. th of a RESTful variant of the model view controller pattern where a json over http API provides a clear separation between data operations and user interface We adopt that pattern here adding an agent queue model to support potentially slow computations We are fortunate to be building on existing tools which will provide building blocks for several components D8 1 Initial specification of the system FP7 ICT 2013 10 Collaborative Project 611553 COINVENT Contents p Overview N A Typical Coinvent Session 1 Core Object The Blend Diagram in Progress 2 EN Components 3 4 1 User Portal Web App 2 20 0 0 200 2 eee 3 CONCEP oi aka EPS SG Pek ae a a he Rae ee oe N 4 3 Base Given 2 Concepts compute a common base Concept 4 4 Weaken Amalgam Finder Given an inconsistent Blend Diagram weaken the inputs 4 5 Model Example Finder Given a Concept find examples models 4 6 Consistency Concept Correctness Is it consistent 4 7 Quality Concept Scorer How good is a Concept o o o o oo 4 8 File Concept Store Store Concepts and Blend Diagrams 4 9 Job Provide meta data on slow tasks Di UU Ur UU HB E Y UN Architecture ee eee Raps DS a hea eee Boer ee EEA a ar GG Gade Meade oe as ar Oe ale Ad a oh GP dh diate at Gs AN GS a o ee at os a ae Be wOoOornan N Languages File Formats ie abd aa Seba
34. them Needs e A very friendly UI e An ontology of animals behind it e To specify two animals to blend e To get a pictorial view of the blend e To see some behaviour e g robot cat chases mutant mice stops to drink some milk short circuits Success criteria Adopted for use by a school or distributed publicly as an app Priority Medium For Year 2 This is an achievable goal for this project Domain Mechanical engineering User A mechnical engineer Task Create a new product and streamline the 3D modelling requirements checking and safety testing by leaning on an existing database Wants To specify a new component e g a car ejector seat which has aspects of existing components Then get a CAD model which fits the spec Priority Low This lacks a sponsor within the project Domain Poetry User A poet Task Create poetry fragments to use as inspiration source material Use case 1 Pick 2 concepts as input and get back a newly created metaphor and sentences illustrating it 2 Pick a concept and a property and get back a 2nd concept which when blended with the first provides a metaphor for the desired property Success criteria A poet collaborates with the project and uses Coinvent to create a piece of poetry they then publish Priority Medium This is an achievable goal for this project This is an accessible domain for technology testing and it fits with the research interests of Goldsmiths group 22
35. weakening or otherwise resolving clashes as fully fledged idioms are likely to be logically inconsistent 4 Given an idiom produce example music 5 Optional Support for evaluating music idioms Success criteria It works Priority High This is the basis for all music domain work User A Listener Task Create music they like Use case Pick 2 musical idioms as input and get back a newly created high quality piece of music Success criteria A group is identified who adopt Coinvent based software as part of their music listening software Or a Coinvent created piece of music gains small scale public success Priority Low This is a very hard task and may not be achieved within this project 611553 June 30 2014 21 D8 1 Initial specification of the system User An avant garde composer Task Create music to use as inspiration source material Use case Pick 2 musical idioms as input and get back a newly created low quality piece of music Success criteria A composer collaborates with the project and uses Coinvent to create a piece of music they then perform Priority For Year 2 3 This is an achievable goal for this project C 4 1 Other Domains Domain Fictional beasts User A child Task Create a fictional beast Use case Pick two input beasts using a simple user friendly GUD and get back a fictional blended beast with a mini story about how it behaves Wants To play around creating fun new animals and interacting with
36. ymbols between two domains E g bank here maps to river_bank there Model An example of a Theory which may be a more concrete theory or a domain specific example such as a a piece of music Ontology Domain Space Theory Idiom OWL a description language for writing ontologies most commonly OWL2 using Manchester syntax Sentence A statement in a Theory E g an axiom Signature the set of symbols used in a theory Sort Type Class E g Person or Boat Symbol Term Anything used in a Sentence E g x or or Chord TBox assertions about a domain T for terminology Theory a set of symbols the signature and sentences 24 June 30 2014 611553
Download Pdf Manuals
Related Search