After early exposure to research working on expert systems based on temporal logic, rules and frames as a software engineer trainee at IBM and Dassault research centers in France, my research career per-se started at Columbia University, New York where I tackled three key issues in Natural Language Generation (NLG):
(1) the generation of coordinated text and graphics in multimedia tutorials and data summaries,
(2) the generation of concise and semantically dense news summaries, by opportunistically fitting background information as short modifying phrases in carefully chosen locations inside a draft focusing only on the key new facts to report, and
(3) a methodology for knowledge acquisition and quantitative output evaluation of NLG systems based on model text corpus analysis. This work resulted in 12 publications including in one in Artificial Intelligence and one in Computational Linguistics, in addition to my PhD. thesis (cited in 55 publications indexed on Google Scholar).
Once a post-doctoral researcher and then professor at Universidade Federal de Pernambuco (UFPE) in
Over the same period, I also worked on parallel tracks addressing the following issues:
(1) Improving web search by automatically expanding queries with keyword synonyms from a thesaurus,
(2) Engineering methodology and CASE tools for agent teams (e.g., in robotic soccer simulations) that leveraged the complementary strengths of object-oriented modeling and logic programming (using UML, OCL and the based on Transaction Frame Logic     ), and
(3) Inductive Object-Oriented Logic Programming to learn Flora-2 programs from training data. Work on theses tracks resulted in 3 Master's Theses and 4 publications, including a Best Paper Award.
Albeit diverse, all the above decision support systems, web search and multi-agent simulations projects shared two aspects. First, they all attempted to engineer highly innovative, practical IT services by aggregating some form of cutting-edge Automated Reasoning (AR) to technologies from other fields (databases, distributed systems, software engineering). Second, their technology transfer from brittle proof of concepts prototypes to profitable industrial applications all hit one common stumbling block: the AR tools they relied upon were prohibitively hard to learn, use, interoperate, customize, and extend, mainly due to their monolithic software architecture, unfriendly interfaces and non-mainstream execution platforms.
Over the last four years, I thus refocused my research
on removing this stumbling block, by elaborating the long-term project ORC2AS
(Objects, Rules, Constraints and Components for Agents and Simulations).
It studies the low-cost engineering of easy to learn, use, debug, customize, extend and assemble Java or .Net components that encapsulate the most versatile yet scalable AR techniques. It covers the methodological (artifacts, process) technological (languages, components) and theoretical (formal semantics) axes of such study. Its key idea it to integrate in bi-directional synergy:
b. Object-Oriented Rule-Based Constraint Programming.
Due to the ambitious scope and trans-disciplinary nature of this project, I first focused my efforts on creating an international network of collaborations to carry it out. This network currently aggregates a 25 strong team (9 senior researchers, 8 PhD. students and 8 MSc. students) in 7 institutions: UFPE, Universidade de Pernambuco, Université Pierre et Marie Curie, Universität Mannheim, Universität Ulm, Fraunhofer Gesellschaft, and INRIA. Parts of this project have been funded by CAPES, FACEPE, COFECUB, DAAD and INRIA. It has so far resulted in 8 publications.
PhD. Theses (sole supervisor)
MODELOG: Model-Oriented Development with Executable Logic Object Generation.
Centro de Informática, Universidade Federal de Pernambuco (CIn-UFPE),
Hendrik Macedo. A
Software Architecture for Ubiquitous Web Browsing with Application to
Recommendation Systems. CIn-UFPE,
Eloi Favero. Generating
Hypertext Summaries of Quantitative Data in Multi-Dimensional Databases.
PhD. Thesis (co-supervisor)
Gramática do Sintagma Nominal Português em um Formalismo Computacional. (A
Computational Grammar of the Portuguese Noun Phrase), Departamento de Literatura e Linguística,
Universidade Federal de Alagoas,
Master's Theses (sole supervisor):
Klebson Silva. Ambiente
de Análise e Depuração de Sistemas Multi-Agente: Aplicação ao Desenvolvimento
de Time de Futebol de Robô (Analysing and degugging multi-agent systems based
on object-oriented logic with application to RoboCup), CIn-UFPE,
Damasceno. Desenvolver Agentes de Futebol com Objetos e Lógica (Developing
soccer agents with objects and logic). CIn-UFPE,
Um Componente de Mineração de Exceções em Cubos OLAP (A component for outlier
mining in OLAP databases). CIn-UFPE,
Pereira. SKDQL: Uma Linguagem Declarativa de Especificação de Tarefas e
Consultas de Descoberta de Conhecimento em Banco de Dados e sua Implementação
(SKDQL, a declarative language for specifying knowledge discovery queries and
tasks and its implementation). CIn-UFPE,
Franklin Ramalho. Avaliando
o Uso de um Tesaurus no Processo de Expansão Automática de Consultas para
Engenhos de Busca na Web (Evaluation of thesaurus-based query expansion for web
search engines). CIn-UFPE,
Natasha Queiroz. DOODI:
uma API para integracao de bancos de dados multidimensionais e sistemas
dedutivos (DOODI: An API to connect OLAP and object-oriented deductive
JODI: uma API Java para OLAP (JODI: a Java API to access OLAP databases).
Bachelor's Degree Graduation Projects (sole supervisor):
Marcos Aurelio da
Silva. CHORD: Constraint Handling Object-oriented Rules with Disjunctive
Luiz Lacerda. Interface
Visual de Depuração de Time de Futebol de Robô (A GUI for Simulated Robotic
Soccer Team Debugging). CIn-UFPE,
André Novaes. Estudo
de Viabilidade de uma Implementação Oracle9i de um SGBD Dedutivo Orientado a
Objeto no Formalismo F-Logic (Feasibility of implementing a Frame Logic
deductive object-oriented databases in Oracle9i). CIn-UFPE,
From 1990 to 1996, I co-developed with Michael Elhadad the open-source, reusable SURGE (Systemic Unification Realization Grammar of English) syntactic realization component for Natural Language Generation (NLG). Written in the high-level, declarative, sorted feature logic language FUF (Functional Unification Facility), SURGE encapsulates a wide coverage of the English grammar that computationally formalizes and integrates the linguistic theories and grammars of Halliday, Quirk at al. and others. It consists of over 7500 lines of FUF code. After the release of version 2.0, it quickly became to most widely adopted syntactic realization tool in the worldwide NLG research community, being cited in about half of all NLG publications over the period 1995-2001. It is now cited in 233 research publications and 10,300 web pages. It was also successfully reused in several industrial systems (cf. technology transfer section). The availability of FUF compilers to Lisp, C and Java make SURGE efficient and available on many platforms.
As a research consultant at Bellcore, New Jersey, in 1993-1995, I participated to the successful transfer of the NLG technology to automate production of required business documentation: (a) textual summaries of telecommunication equipment installation plans that justify pricing decisions to regulatory agencies or judges during consumer lawsuits, and (b) textual summaries justifying business reengineering plans to management of consultancy client firms. In the first project, the PLANDoc NLG component was integrated to an expert system generating a base plan and an alternative scenario simulator allowing an engineer to test base plan variations. It automatically generated the textual summary from traces of these two other components. In the second project, the FLOWDoc NLG component was integrated to a workflow modeling tool to acquire the client's current business process and an optimization tool suite to discover better business processes. It automatically generated the textual summary from the workflows of the current and proposed reengineered business processes. These two systems were put in profitable day to day use at Bellcore. PLANDoc generated documents that were used in court as evidence in a million dollar range lawsuit won by Bellcore. FLOWDoc generated documents that convinced several Bellcore clients to adopt the proposed reengineering of their business process.
In 2003, I was hired by CESAR -- a financially self-sustained IT consulting and start-up farm NGO and the largest organization in downtown Recife's IT organization cluster (Porto Digital) -- to head a consulting team which mission was to define the strategic mission, legal status, organizational structure, funding strategies and initial projects of I3 (Instituto de Inovação em Informática), a CESAR spin-off dedicated to connect local universities, companies, government agencies and NGO to carry out economically promising IT innovation projects. I3 is now playing a key role in Porto Digital to foster technology transfer and create profitable synergies in the regional IT ecosystem.
 Bonner, A. and Kifer, M. Transaction Logic programming (or, a logic of procedural and
declarative knowledge). Technical Report. CSRI-323, Computer Systems
 Bonner, A. and Kifer, M. Concurrency and
Communication in Transaction Logic. Proceedings of the Joint International Conference and Symposium on
 Kifer, M. Deductive and Object Data Languages: A Quest for Integration. In Lecture Notes in Computer Science 1013. Springer. 1995.
 Kifer, M., Lausen, G. and Wu, J . Logical Foundations of Object-Oriented and Frame-Based Languages. Journal of the ACM 42. 1995.
 Yang, G. A
Model Theory for Nonmonotonic Multiple Value and Code Inheritance in
Object-Oriented Knowledge Bases. PhD. Thesis, Computer Science Department,
Stony Brook University of
© Jacques Robin 2007. Last Updated: 06/03/2007