Research activity in 1999
During 1999, the research on Natural Language Processing has concerned
the following topics:
- Formal syntax and Cognitive Architectures
- Cognitive modeling of Pragmatic Phenomena
- Agents and dialog
Formal Syntax and Cognitive Architectures
This research has been carried on in strict cooperation with Vincenzo Lombardo, of the University of Piemonte Orientale ‘Amedeo Avogadro’.
During 1999, we have continued the investigation on the definition of a syntactic theory based on the dependency paradigm, and we have applied the theoretical constructs to the development of linguistic resources for Italian. In particular, we have implemented two tools that help in the linguistic analysis, and interact with the human annotators in the construction of the resources. In the definition of the syntactic theory, we have investigated the role of non-lexical units in the dependency tree. Non-lexical units permit a clear representation of the covert dependencies in the sentence. The insertion of such units, though uncommon in the dependency-based paradigm, has revealed to be particularly useful in practical applications of language processing.
The first software tool, devoted to treebank annotation, is based on psycholinguistic modeling. Annotators work with an interactive parser, which pursues a strongly incremental strategy by maintaining a totally connected parse at all times. The annotation format is in the dependency paradigm, and thus oriented to the representation of the predicate-argument structure. The syntactic base is a grammar of individual dependency relations between categories; more complex structural chunks are aggregated as encountered during the syntactic annotation. Attachment ambiguities are solved by the annotators, but the tool presentation is enhanced through the use of psycholinguistic preferences. The graphic interface of the tool, for the display of partial trees to the annotators, is realized by the program DaVinci.
The second software tool implements the automatic assignment of dependency labels (grammatical functions for verbs). It is based a set of generic assignment rules, on a set of 'subcategorization frames' of Italian verbs, and on a set of 'transformations' which produce the possible surface realizations of the verbal case frames. Currently, this tool has been tested on about 20.000 words, with a percentage of errors of about 16% on the test set (18.000 words used as corpus for the - manual - learning of the various rules, and 2.000 words in the test set). The set of rules is currently under extension.
Cognitive modeling of Pragmatic Phenomena
The use of plan-based representations in the study of semantic phenomena leads to an intersection of semantics and pragmatics, where the contributions of the representation and reasoning framework constituted by mental models (Johnson-Laird 1983) has proved fruitful. In particular, we have examined the following
phenomena:
Tense and aspect
By using a plan-based representation of action verbs to model the aspectual and temporal composition of a sentence, it is possible to account for a number of elements that contribute to determining its resulting aspect.This can be done without introducing an explicit ontology of aspectual classes [Boella and Damiano 1999].
Action verbs are interpreted as instances of action schemata; these instances can be underspecified, depending on the description that a certain sentence gives of an event.
The interpretation process builds the action schema instance incrementally as the elements of the description are encountered (verb complements, temporal adverbial phrases, tense), and adds some constraints to it, depending on the presence and on the value of these elements. Pragmatic factors (like reference to the described agent's intentions and telicity as a conversational implicature) are also taken into account in the model.
Mental models and presuppositions
Our proposal exploits mental models and conversational implicatures, together with a plan-based representation of action verbs, to explain how lexical presuppositions arise in the presence of action verbs [Boella, Damiano and Lesmo 1999]. Plans offer a natural and rich representation of the semantics of action verbs, while mental models provide the reasoning framework.
The interpretation process proceeds in an incremental way: first, a set of mental models is built; it embodies the characteristics of the described event, expressed in the sentence via aspectual and temporal features or via other linguistic features, like negation; so, what is shared by all models constitutes what the sentence entails. On this representation, conversational implicatures are computed, and their effect is to discharge the mental models that could be described more appropriately by alternative linguistic expressions: what is shared by the remaining models is what the sentence presupposes.
Mental models and anaphora
Since mental models are able to account both for the representation of a linguisitic expression and for the reasoning process carried out on this representation, they provide the adequate framework to explain complex anaphoric phenomena, as the ones exemplified by donkey sentences [Boella, Damiano and Lesmo 1999b]. We designed a procedure to construct mental models of donkey sentences that allows to account for dependency of reference in and across sentences in an explicit way, by means of unification and fleshing out of tokens and relations. Moreover, several related phenomena are covered by this model, including the interpretation of generalized quantifiers, universal/existential readings, and the connection between the interpretation of anaphora and presuppositions.
Agents and dialog
This area covers the construction of agent based systems which are applied for modeling dialogic phenomena and, more recently, modeling discourse and narrative.
Dialog coherence
The reasearch on dialogic phenomena started in the previous years has continued and the results have been published in a forthcoming article, Plan based agent architecture for interpreting natural language dialogue, to appear in Int.l Journal of Human-Computer Studies, Academic Press, by L. Ardissono, G. Boella, and L. Lesmo.
The proposal is based on the idea that when the user utters a new turn, the system looks for a coherence link with the previous dialog context. Coherence links are based on the participants' underlying intentions, both at linguistic and domain level. The goals underlying a certain turn could be related to the interlocutor's pending goals (goal adherence and goal adoption), or could be related to a plan that the speaker is carrying on.
Agents and cooperation
We are currently developing a utility-based approach to cooperation among agents [Boella, Damiano and Lesmo 1999, Boella, Damiano and Lesmo 1999b, Boella 1999]. Agents - we assume - are rational, and they choose among alternative courses of action by exploiting a multi-attribute utility function that expresses the extent to which a certain course of action contributes to the achievement of the agent's goals. Most interactions among agents, however, involve some form of cooperation: when a group of agents is cooperating to a shared plan, a combination of individual utility and
group utility should be used. Each socially responsible agent, when evaluating alternative courses of action, considers also the consequences on the partners’ choices, in the light of this hybrid measure of group and individual utility. The effort to maximize utility leads to some interesting behaviors, including goal adoption, helpfulness, appropriate generation of communicative acts, and so on: all these phenomena contibute to improve coordination and to reduce conflicts among the actions of the invlved agents.
Narrative and plans
The dialog agent for interpreting Natural Language, whose knowledge is encoded in plan libraries, can be extended to create a system that understands narratives that contain descriptions of domain and linguistic actions [Boella, Damiano and Lesmo 1999c, Boella, Damiano and Lesmo 1999d]. The system incrementally reconstructs coherence links among described actions, by attributing intentions to the described agents. Moreover, the intention attribution activity can be extended to the narrator's linguistic planning, by identifying communicative intentions underlying the structure of the narrative. At the same time, the dialog agent architecture lends itself to building a system that produces a description his own activity.
1999 Publications
International Conferences and Workshops
L. Ardissono, G. Boella, and L. Lesmo. The role of social goals in planning polite speech acts. In Workshop on Attitude, Personality and Emotions in User-Adapted Interaction at UM'99 Conference, Banff, 1999.
G. Boella and R. Damiano. Plan-based event representations for the analysis of tense and aspect. In Proceedings of 6th Congress of AI*IA, Bologna, 1999.
G. Boella, R. Damiano, and L. Lesmo. Beating a donkey: a mental model approach to complex anaphorical phenomena. In Proceedings of European Congress of Cognitive Science of ECCS'99, Pontignano, 1999.
G. Boella, R. Damiano, and L. Lesmo. Cooperating to the group's utility. In N.R. Jennings and Y. Lespérance, editors, Intelligent Agents VI - Proceedings of the Sixth International Workshop on Agent Theories, Architectures, and Languages ATAL-99, Orlando (FL), 1999.
G. Boella, R. Damiano, and L. Lesmo. Dialog modeling and reported speech in narrative. In Proceedings of AAAI Workshop on Mixed Initiative Intelligence at AAAI'99 Conference, Orlando (FL), 1999.
G. Boella, R. Damiano, and L. Lesmo. Mental models and pragmatics: the case of presuppositions. In Proceedings of the 21st Annual Meeting of the Cognitive Science Society, Vancouver (BC), 1999.
G. Boella, R. Damiano, and L. Lesmo. Understanding narrative is like observing agents. In Proceedings of AAAI Fall Symposium on Narrative Intelligence, Cape Cod, 1999.
G. Boella, R. Damiano, and L. Lesmo. A utility based approach to cooperation among agents. In Proceedings of Workshop on Collective Agent Based Systems at ESSLLI'99, Utrecht, 1999.
G. Boella, R. Damiano, L. Lesmo, and L. Ardissono. Conversational cooperation: the leading role of intentions. In Amstelogue'99 Workshop on Dialogue, Amsterdam, 1999.
Other publications
G. Boella. Cooperation among economically rational agents. PhD thesis, Università di Torino, Italy, 1999.
