Susan Windisch Brown

VerbNet class assignment as a WSD task

The VerbNet lexical resource classifies English verbs based on semantic and syntactic regularities and has been used for numerous NLP tasks, most notably, semantic role labeling. Since, in addition to thematic roles, it also provides semantic predicates, it can serve as a foundation for further inferencing. Many verbs belong to multiple VerbNet classes, with each class membership corresponding roughly to a different sense of the verb. A VerbNet token classifier is essential for current applications using the resource and could provide the basis for a deep semantic parsing system, one that made full use of VerbNets extensive syntactic and semantic information. We describe our VerbNet classifier, which uses rich syntactic and semantic features to label verb instances with their appropriate VerbNet class. It achieves an accuracy of 88.67% with multiclass verbs, which is a 49% error reduction over the most frequent class baseline.

Criteria for the Manual Grouping of Verb Senses

In this paper, we argue that clustering WordNet senses into more coarse-grained groupings results in higher inter-annotator agreement and increased system performance. Clustering of verb senses involves examining syntactic and semantic features of verbs and arguments on a case-by-case basis rather than applying a strict methodology. Determining appropriate criteria for clustering is based primarily on the needs of annotators.

Choosing Sense Distinctions for WSD: Psycholinguistic Evidence

Supervised word sense disambiguation requires training corpora that have been tagged with word senses, which begs the question of which word senses to tag with. The default choice has been WordNet, with its broad coverage and easy accessibility. However, concerns have been raised about the appropriateness of its fine-grained word senses for WSD. WSD systems have been far more successful in distinguishing coarsegrained senses than fine-grained ones (Navigli, 2006), but does that approach neglect necessary meaning differences? Recent psycholinguistic evidence seems to indicate that closely related word senses may be represented in the mental lexicon much like a single sense, whereas distantly related senses may be represented more like discrete entities. These results suggest that, for the purposes of WSD, closely related word senses can be clustered together into a more general sense with little meaning loss. The current paper will describe this psycholinguistic research and its implications for automatic word sense disambiguation.

VerbNet overview, extensions, mappings and applications

The goal of this tutorial is to introduce and discuss VerbNet, a broad coverage verb lexicon freely available on-line. VerbNet contains explicit syntactic and semantic information for classes of verbs and has mappings to several other widely-used lexical resources, including WordNet, PropBank, and FrameNet. Since its first release in 2005 VerbNet is being used by a large number of researchers as a means of characterizing verbs and verb classes. The first part of the tutorial will include an overview of the original Levin verb classification; introduce the main VerbNet components, such as thematic roles and syntactic and semantic representations, and present a comparison with other available lexical resources. During the second part of the tutorial, we will explore VerbNet extensions (how new classes were derived and created through manual and semi-automatic processes), and we will present on-going work on automatic acquisition of Levin-style classes in corpora. The latter is useful for domain-adaptation and tuning of VerbNet for real-world applications which require this. The last part of the tutorial will be devoted to discussing the current status of VerbNet; including recent work mapping to other lexical resources, such as PropBank, FrameNet, WordNet, OntoNotes sense groupings, and the Omega ontology. We will also present changes designed to regularize the syntactic frames and to make the naming conventions more transparent and user friendly. Finally, we will describe some applications in which VerbNet has been used.

The Rich Event Ontology.

In this paper we describe a new lexical semantic resource, The Rich Event On-tology, which provides an independent conceptual backbone to unify existing semantic role labeling (SRL) schemas and augment them with event-to-event causal and temporal relations. By unifying the FrameNet, VerbNet, Automatic Content Extraction, and Rich Entities, Relations and Events resources, the ontology serves as a shared hub for the disparate annotation schemas and therefore enables the combination of SRL training data into a larger, more diverse corpus. By adding temporal and causal relational information not found in any of the independent resources, the ontology facilitates reasoning on and across documents, revealing relationships between events that come together in temporal and causal chains to build more complex scenarios. We envision the open resource serving as a valuable tool for both moving from the ontology to text to query for event types and scenarios of interest, and for moving from text to the ontology to access interpretations of events using the combined semantic information housed there.

Multimodal Use of an Upper-Level Event Ontology

We describe the ongoing development of a lexically-informed, upper-level event ontology and explore use cases of the ontology. This ontology draws its lexical sense distinctions from VerbNet, FrameNet and the Rich Entities, Relations and Events Project. As a result, the ontology facilitates interoperability and the combination of annotations done for each independent resource. While this ontology is intended to be practical for a variety of applications, here we take the initial steps in determining whether or not the event ontology could be utilized in multimodal applications, specifically to recognize and reason about events in both text and video. We find that the ontology facilitates the generalization of potentially noisy or sparse individual realizations of events into larger categories of events and enables reasoning about event relations and participants, both of which are useful in event recognition and interpretation regardless of modality.

The relevance of a cognitive model of the mental lexicon to automatic word sense disambiguation: invited talk

Supervised word sense disambiguation requires training corpora that have been tagged with word senses, and these word senses typically come from a pre-existing sense inventory. Space limitations imposed by dictionary publishers have biased the field towards lists of discrete senses for an individual lexeme. Although some dictionaries use hierarchical entries to emphasize relations between senses, many do not. WordNet, which has been the default choice of NLP researchers for sense tagging because of its broad coverage and easy accibility, does not have hierarchical entries. Could the relations between senses that are captured by a hierarchy be useful to NLP systems? Concerns have also been raised about whether or not WordNets word senses are unnecessarily fine-grained. WSD systems are obviously more successful in distinguishing coarse-grained senses than fine-grained ones (Navigli, 2006), but important information could be lost if fine-grained distinctions are ignored. Recent psycholinguistic evidence seems to indicate that closely related word senses may be represented in the mental lexicon much like a single sense, whereas distantly related senses may be represented more like discrete entities (Brown, 2008). These results suggest that, for the purposes of WSD, closely related word senses can be clustered together into a more general sense with little meaning loss. This talk will describe this psycholinguistic research and its current implications for automatic word sense disambiguation, as well as plans for future research and its possible impact.