Inducing Example-based Semantic Frames from a Massive Amount of Verb Uses

Abstract

We present an unsupervised method for inducing semantic frames from verb uses in giga-word corpora. Our semantic frames are verb-specific example-based frames that are distinguished according to their senses. We use the Chinese Restaurant Process to automatically induce these frames from a massive amount of verb instances. In our experiments, we acquire broad-coverage semantic frames from two giga-word corpora, the larger comprising 20 billion words. Our experimental results indicate the effectiveness of our approach.

A Step-wise Usage-based Method for Inducing Polysemy-aware Verb Classes

“…The original method in Kawahara et al (2014) defined w as pairs of slots and words, e.g., "nsubj:child" and "dobj:bird," but does not consider slot-only features, e.g., "nsubj" and "dobj," which ignore lexical information. Here we experiment with both representations and compare the results.…”

“…We induce verb-specific semantic frames from verb uses based on the method of Kawahara et al (2014). Our semantic frames consist of case slots, each of which consists of word instances that can be filled.…”

CzEngClass – Towards a Lexicon of Verb Synonyms with Valency Linked to Semantic Roles

“…We will create a lexicon of such synonymous verb pairs around representatively selected "seed" verbs; such similarity will be tested primarily against the translational equivalents in context, as found in the parallel corpora. We will compare the results with the approach of [18] as embodied, e.g., in the VerbNet [24], [14], as a representative of classes of semantically and syntactically similar verbs based on monolingual resources and research on one language (English). Last, but certainly not least, we will compare the resulting classes and their properties to the VALLEx lexicon [20], [19] and VerbaLex [13] on the Czech side.…”

“…mff.cuni.cz/pdt2.0) and PCEDT [11], and the valency lexicons linked to these treebanks: PDT-Vallex [32], EngVallex [5], and a parallel valency lexicon CzEngVallex [33], [34]. We also take advantage of another fGD-based lexicon VALLEx [20], [19], [15] and other available resources, such as VerbNet [24], [14], frameNet [3], English [6] and Czech WordNet [22], [23].…”

“…Some researchers even go so far as to declare that they "do not believe in word senses" (legendary quote by the lexicographer Sue Atkins [2], explained by an article with the same title by Kilgarriff [16] that it should be interpreted as not believing in predetermined, fixed set of word senses). Others try to find a sweet spot between a hardto-agree-on, fine-grained set, represented e.g., by WordNet [6], and a coarse(r)-grained set, which does not provide enough detail-such as VerbNet lexicon [24], [14]. frameNet [3], [8], [9] an English lexical resource which adds roles and uses semantic frames to group verbs and provide examples of use (based on attested corpus examples) is another well-known resource.…”