Brian Plüss

We describe ongoing research towards modelling dialogue management for conversational agents that can exhibit and cope with non-cooperative behaviour. Empirical studies of conventional dialogue behaviour in the domain of political interviews and a coarse-grained notion of conversational games are used to characterise non-cooperation. We propose an agent architecture that combines conversational games and discourse obligations, and suggest an implementation.

@InProceedings{pluss:SemDial-2011:b,
  author= "Brian Pl{\"u}ss and Paul Piwek and Richard Power",
  title = "Modelling Non-Cooperative Dialogue: the Role of  Conversational Games and
      Discourse Obligations",
  booktitle = "Proceedings of SemDial 2011, the 15th Workshop on the Semantics and
      Pragmatics of Dialogue",
  location = "Los Angeles, California",
  numpages = "2",
  year = "2011",
  month = sep
}

This paper reports on an ongoing effort to enable the rapid development of multi-party virtual human negotiation scenarios. We present a case study in which a new scenario supporting negotiation between two human role players and two virtual humans was developed over a period of 12 weeks. We discuss the methodology and development process that were employed, from storyline design through role play and iterative development of the virtual humans’ semantic and task representations and natural language processing capabilities. We analyze the effort, expertise, and time required for each development step, and discuss opportunities to further streamline the development process.

@InProceedings{pluss:SemDial-2011:a,
  author= "Brian Pl{\"u}ss and David DeVault and David Traum",
  title = "Toward Rapid Development of Multi-Party Virtual Human Negotiation Scenarios",
  booktitle = "Proceedings of SemDial 2011, the 15th Workshop on the Semantics and
      Pragmatics of Dialogue",
  location = "Los Angeles, California",
  numpages = "10",
  year = "2011",
  month = sep
}

We have applied Fastest, an implementation of the Test Template Framework, to five real case studies of aerospace software. This involved the formalization in the Z notation of nontrivial parts of each system. One of these models, for instance, formalizes a significant portion of the ECSS-E-70-41A aerospace standard. The models were then fed into Fastest, which automatically generated detailed functional abstract test cases. Since these test cases are independent of any implementation, they can be used to test any of them. Furthermore, we were able to semi-automatically translate them into English so they can be used by domain experts performing independent validation and verification activities.

@InProceedings{cristia:SEW34-2010,
  author= "Maximiliano Cristi{\'a} and Pablo Albertengo and Claudia Frydman and
           Brian Pl{\"u}ss and Pablo Rodr{\'i}guez Monetti",
  title = "Applying the Test Template Framework to Aerospace Software",
  booktitle = "Proceedings of the 34th IEEE Annual Software Engineering Workshop",
  series = "SEW 2011",
  location = "Limerick, Ireland",
  numpages = "10",
  year = "2011",
  month = jun,
  organization = "IEEE"
}

[ pdf | slides | poster ]

This paper presents ongoing research on computational models for non-cooperative dialogue. We start by analysing different levels of cooperation in conversation. Then, inspired by findings from an empirical study, we propose a technique for measuring non-cooperation in political interviews. Finally, we describe a research programme towards obtaining a suitable model and discuss previous accounts for conflictive dialogue, identifying the differences with our work.

@InProceedings{pluss:ACLSRW-2010,
  author= "Brian Pl{\"u}ss",
  title = "Non-cooperation in dialogue",
  booktitle = "Proceedings of the ACL 2010 Student Research Workshop",
  series = "ACL-SRW 2010",
  location = "Uppsala, Sweden",
  pages = "1--6",
  numpages = "6",
  year = "2010",
  month = jul,
  organization = "Association for Computational Linguistics",
  address = "Stroudsburg, PA, USA"
}

[ pdf | teaser | poster ]

Critical software most often requires an independent validation and verification (IVV). IVV is usually performed by domain experts, who are not familiar with specific, many times formal, development technologies. In addition, model-based testing (MBT) is a promising testing technique for the verification of critical software. Test cases generated by MBT tools are logical descriptions. The problem is, then, to provide natural language (NL) descriptions of these test cases, making them accessible to domain experts. In this paper, we present ongoing research aimed at finding a suitable method for generating NL descriptions from test cases in a formal specification language. A first prototype has been developed and applied to a real-world project in the aerospace sector.

@InProceedings{cristia:pluss:INLG-2010,
  author= "Maximiliano Cristi{\'a} and Brian Pl{\"u}ss",
  title = "Generating natural language descriptions of Z test cases",
  booktitle = "Proceedings of the 6th International Natural Language Generation Conference",
  series = "INLG 2010",
  location = "Dublin, Ireland",
  pages = "173--177",
  numpages = "5",
  year = "2010",
  month = jul,
  organization = "Association for Computational Linguistics",
  address = "Stroudsburg, PA, USA"
}

[ pdf ]

Most work in linguistics has approached dialogue on the assumption that participants share a common goal and cooperate to achieve it by means of conversation. In computational linguistics this assumption is even stronger. For instance, most dialogue systems rely on the interlocutor’s full coopera- tion to model interaction. The research described here is aimed at the other cases, at those escaping the norms. Failure to cooperate can happen for many reasons. A non-native speaker trying to engage in a complex discussion might provide contribu- tions which are not as clear and precise as would be expected. A student not quite sure about the topic he is supposed to elaborate on in an oral exami- nation might provide information which is not entirely truthful or relevant. Someone suffering from dementia might produce utterances which are irrelevant or uninformative for the current exchange. These examples have to do with incompetence, ignorance and irrationality, all of which lie outside the scope of our study. We will focus on situations in which non-cooperative conversational behaviour is rational, competent and well-informed. This report is part of the first-year probation assessment for a full-time Ph.D. programme. It provides details about the proposed research question, a review of the relevant literature, the proposed research methodology and a work plan.

@TechReport{pluss:TR-2009,
  author= "Brian Pl{\"u}ss",
  title = "Towards a Computational Pragmatics for Non-Cooperative Dialogue",
  number = {2009/13},
  year = "2009",
  month = jul,
  day = "21"
  url = {http://computing-reports.open.ac.uk/2009/TR2009-13.pdf},
  type = {PhD Probation Report}
}

[ pdf | slides (in Spanish) ]

Program verification refers to proving formally that a program is correct with respect to its specification. It is assumed almost universally that the work required for these proofs is so complex, time consuming and expensive that it can not be done but for the most critical parts of a software system.
Invariant Based Programming (IBP) comes to fill the gap between programming and verification. While other approaches to verification are based on a usually large number of complex rules that make them unnatural and unappealing resulting in low popularity, IBP proposes a natural way of obtaining software which is correct by construction.
Procedures constitute a natural extension once a formalism has been defined for the most basic programming constructs. They provide access to abstraction, giving any programming method increased power to deal with more complex programs and algorithms. Procedures also serve as the starting layer, upon which higher constructs can be defined (e.g. classes, data modules, and so forth) and allow (mutual) recursion and exception handling.
However, most proof rules for procedure calls in the literature contain considerable complexity in order to handle parameter binding, different parameter passing conventions and recursion. This makes verification in the presence of procedures difficult in practice.
In this work, we present a practical method for reasoning about mutually recursive procedures with multiple exits in a visual formalism. Based on the refinement calculus, we model procedure calls as multi-exit statements and parameter binding as assignments to fresh variables in an extended state space. We apply the method to a case study; a set of procedures for string manipulation, and suggest how it could be implemented in the SOCOS research tool.

@mastersthesis{pluss:masters-2008,
  author= "Brian Pl{\"u}ss",
  title = "A Practical Method for Reasoning About Procedures in Invariant Based Programming",
  school = "Universidad Nacional de Rosario",
  number = {2009/13},
  year = "2008",
  month = sep,
  address = "Argentina"
}

[ pdf | short | slides | poster ] (all in Spanish)

Este trabajo presenta un propuesta novedosa para la detección y corrección de errores ortográficos basada en la automatización de las reglas que rigen la ortografía de la lengua castellana. Para ello, se analiza un subconjunto de las reglas tal como aparecen en la literatura
y se introducen los conceptos lingüísticos a los cuales las reglas hacen referencia. Se describen brevemente las técnicas computacionales existentes para el procesamiento de lenguajes naturales, explicando cómo éstas pueden ser utilizadas en la tarea de determinar si una palabra viola alguna de las reglas establecidas y corregir el error. La factibilidad de la propuesta es apoyada con la implementación de un prototipo escrito en Prolog y su utilidad expuesta con algunos ejemplos ilustrativos.

@InProceedings{pluss:pomponio:EST-2008,
  author= "Brian Pl{\"u}ss and Laura Pomponio",
  title = "Automating Spelling Rules for Detection and Correction of Errors in Spanish",
  booktitle = "EST 2008 at the 37th Argentine Conference on Informatics and Operations Research",
  year = "2008",
  month = sep,
  address = "Santa Fe, Argentina"
}