The 20th Spanish Congress on Fuzzy Technologies and Logic awards a CiTIUS paper
The jury acknowledges the contribution of the study presented by the centre’s predoctoral researcher Mariña Canabal, an empirical analysis focused on human perception of the inconsistencies of conversational agents in their dialogue with people.
A CiTIUS paper has just been awarded at the twentieth edition of the XX Spanish Congress on Fuzzy Technologies and Logic (ESTYLF 20/21). The award recognises the contribution with the third prize for the best paper in the Student Category, an analysis entitled Empirical evaluation of fuzzy quantification models applied to a conversational agent. The prize-giving ceremony took place today in Malaga during the closing session of the congress, a discussion forum where results, ideas and projects related to the field were presented.
Advantages of an ‘imprecise dialogue’
In the award-winning work, the centre’s scientific team has carried out an empirical study that assesses whether people who engage in a dialogue with a conversational agent do or do not perceive inconsistencies in the conversation, provided that these are due to the use of certain models of imprecise quantifiers.
Imprecise quantifiers are concepts used in a vague and generic way in conversation; expressions such as “all”, “few”, “some”, or “most” are extraordinarily frequent in human language. So much so that it would, in fact, be unthinkable to hold a normal conversation without using this kind of term at some point.
In the field of fuzzy logic, the properties of these quantifier models have so far been studied and defined; thus, for the scientific community in the area it was well known that some of the most common quantifiers failed to meet certain properties that are intuitively expected. The main contribution of the CiTIUS work that has just been awarded is that, for the first time, the problem is approached from a pragmatic point of view; that is, seeking to measure the effect that the violation of these properties has on speakers.
To this end, the researchers have built the conversational agent Quanversa, which is capable of holding conversations about the weather in which it uses quantifiers modelled with the Zadeh model, one of the most widely used paradigms. Using the conversations generated by Quanversa, a study was carried out with 132 users, showing that these people perceive significant differences in terms of the consistency and usefulness of the conversations held in two scenarios: on the one hand, when the conversational agent uses Zadeh’s quantifier model; and on the other, when Quanversa uses a corrected model. This provides empirical evidence that the design of a conversational agent that uses quantified expressions must take into account the fulfilment of certain properties, in order to avoid generating conversations that users may perceive as incoherent.
A CiTIUS paper has just been awarded at the twentieth edition of the XX Spanish Congress on Fuzzy Technologies and Logic (ESTYLF 20/21). The award recognises the contribution with the third prize for the best paper in the Student Category, an analysis entitled Empirical evaluation of fuzzy quantification models applied to a conversational agent. The prize-giving ceremony took place today in Malaga during the closing session of the congress, a discussion forum where results, ideas and projects related to the field were presented.
Advantages of an ‘imprecise dialogue’
In the award-winning work, the centre’s scientific team has carried out an empirical study that assesses whether people who engage in a dialogue with a conversational agent do or do not perceive inconsistencies in the conversation, provided that these are due to the use of certain models of imprecise quantifiers.
Imprecise quantifiers are concepts used in a vague and generic way in conversation; expressions such as “all”, “few”, “some”, or “most” are extraordinarily frequent in human language. So much so that it would, in fact, be unthinkable to hold a normal conversation without using this kind of term at some point.
In the field of fuzzy logic, the properties of these quantifier models have so far been studied and defined; thus, for the scientific community in the area it was well known that some of the most common quantifiers failed to meet certain properties that are intuitively expected. The main contribution of the CiTIUS work that has just been awarded is that, for the first time, the problem is approached from a pragmatic point of view; that is, seeking to measure the effect that the violation of these properties has on speakers.
To this end, the researchers have built the conversational agent Quanversa, which is capable of holding conversations about the weather in which it uses quantifiers modelled with the Zadeh model, one of the most widely used paradigms. Using the conversations generated by Quanversa, a study was carried out with 132 users, showing that these people perceive significant differences in terms of the consistency and usefulness of the conversations held in two scenarios: on the one hand, when the conversational agent uses Zadeh’s quantifier model; and on the other, when Quanversa uses a corrected model. This provides empirical evidence that the design of a conversational agent that uses quantified expressions must take into account the fulfilment of certain properties, in order to avoid generating conversations that users may perceive as incoherent.