1. Structuring the English lexicon: The Berkeley FrameNet project in its 20th year
This talk explores the workflow, methodology, tools, and results of the Berkeley FrameNet project (
http://framenet.icsi.berkeley.edu). Founded by Charles Fillmore, the founder of Frame Semantics, at the International Computer Science Institute in Berkeley, California, in 1997, FrameNet is building a lexical database of English verbs, nouns, adjectives, adverbs, and prepositions using the principles of Frame Semantics. This talk first provides some theoretical background regarding Frame Semantics as developed by Charles Fillmore during the 1980s and 1990s. It then discusses how frame-semantic principles influenced the design and workflow of FrameNet (concept of frames, definition of frame elements, drawing borders between frames, etc.). Next, this talk shows what types of information are contained in lexical entries in FrameNet (specifically: Frame Element Configurations, Null Instantiations, and Valency Reports), and it discusses several important structural properties of FrameNet such as frame-to-frame relations, the FrameGrapher, and the expansion towards covering grammatical constructions. The talk then discusses domain-specific FrameNets for legal and medical language and multilingual FrameNets. Finally, we address some challenges and opportunities regarding the ongoing activities of the Berkeley FrameNet project and we discuss an ongoing international effort towards a multilingual FrameNet for more than ten languages.
2. Construction Grammar and Frame Semantics
This talk discusses the relationship between Construction Grammar and Frame Semantics, which both have their roots in the early work of Fillmore’s (1968) seminal paper “The Case for Case.” While Construction Grammar emerged in the 1980s as a viable alternative to mainstream generative transformational grammar approaches and continued its success from the 1990s until today, Frame Semantics has not received that much attention. This talk explores the reasons for this state of affairs. Starting with a discussion of how Frame Semantics developed during the 1980s out of Fillmore’s earlier research of the 1960s and 1970s, we argue that the concept of “semantic frame” was initially quite difficult to define. We then review how Construction Grammar and Frame Semantics emerged together as a viable cognitive alternative to formal syntactico-centric approaches during the 1980s. Next, we show how Frame Semantics developed in the early 1990s, eventually laying the ground for the Berkeley FrameNet project, a corpus-based online lexical database for English. Finally, we show how the lexical annotation and documentation process of FrameNet has been extended to also cover the annotation and documentation of grammatical constructions, in the so-called constructicon.
3. The structure of Multilingual FrameNets
4. Verb classification using constructions and semantic frames
This talk discusses various methodologies for classifying verbs. The first part discusses a number of related approaches to verb classification that all claim a close connection between form and meaning. On this view, verbs that are closely related in meaning should all exhibit similar types of syntactic patterns, including alternations. I show how the approaches of Pinker (1989), Levin (1993), Pustejovsky (1995), and Rappaport Hovav and Levin (2005) deal with syntactic alternations and discuss the role of verb classification in each of these approaches. The second part reviews some data that are problematic for these approaches, because not all verbs that are supposed to belong to the same class do in fact behave similarly at the syntactic level. These observations form the basis for the third part of my talk in which I first present an alternative way of classifying verbs using frame-semantic criteria (Boas 2006). On this view, semantic factors are primary when it comes to classifying verbs (and not syntactic criteria), and syntactic alternations are “only” an epiphenomenon caused by different verbs evoking different types of semantic frames. The last part of my talk addresses the question of how frame-semantic classifications can be employed for the description and categorization of grammatical constructions.
5. Extending the lexicon: The constructicon between valency and argument structures
Most theories of grammar make a systematic distinction between the lexicon and syntax. This talk shows that this idea is difficult to maintain once one is forced to provide an analysis of complete texts. The first part of the talk introduces the FrameNet project at the International Computer Science Institute in Berkeley, California, and shows how for the past 20 years it has been compiling lexical entries of English words to be stored in an online database structured with semantic frames (Fillmore 1982). The second part of the talk discusses a number of phenomena that are both lexical and non-lexical in nature. In the third part of the talk, I show how such data have been employed to argue for an expansion of the lexical coverage of FrameNet to also include grammatical constructions of different types as known through research in Construction Grammar. The fourth part of the talk discusses the architecture of the so-called constructicon, which uses the same data and annotation format to annotate and document grammatical constructions of different types and of different degrees of abstraction (Fillmore et al. 2012, Boas 2014). More specifically, I show how the interaction of constructional and lexical knowledge is crucial for understanding and analyzing full texts (Ziem et al. 2014).
6. Capturing the meaning of absent words with semantic frames
This talk investigates how the meaning of absent words can be captured systematically using the principles of Frame Semantics (Fillmore 1982). The first part of my talk discusses a range of examples in which words are missing from sentences and asks how these missing words can be interpreted systematically. Part two of my talk argues that each word, sentence, and text builds on pre-existing knowledge of different kinds. Using the principles of Frame Semantics, I show that the types of information contained in FrameNet can be used to systematically account for the types of absent information (Fillmore 1986). A key component of capturing the meaning of absent words includes reference to cultural concepts that are often intimately tied to a word’s meaning as well as to the meanings of idiomatic phrases and other types of figurative language (Boas 2013). Part three of my talk shows how different types of frame-semantic information can be employed to account for both idiosyncratic and more general types of information that is missing from texts. Finally, I address the question of how frame-semantically structured information about words can be used in the interpretation of texts.
7. Polysemy is the enemy: A frame-semantic approach to word meanings and syntactic alternations
Most theories of language classify word meanings based on syntactic criteria. For English, Levin (1993) provides one of the most popular accounts of verb classification based on so-called syntactic alternations. On the other hand, semantic-based approaches such as Frame Semantics (Fillmore 1982) claim that semantic criteria are primary. This talk compares the two approaches and argues that neither approach is currently fine-grained enough to capture both generalizations and medium and low level exceptions to those generalizations. Based on data from Dux (2016), I argue that verb classes and alternations are both epiphenomenon that arise because of specific interactions of type and token frequencies in language. On this view, a usage-based bottom-up approach is a viable alternative to existing accounts that focus either on syntactic or semantic criteria. Building on research in Lexicon Grammar (Salkoff 1983, Gross 1990), I argue for a frame-constructional approach towards verb classifications that captures both relevant verb classes and syntactic alternations at the same time.
8. Construction families and networks: Capturing generalizations and idiosyncratic patterns
This talk compares how constructional research deals with regular and irregular structures that are typically located in the so-called “lexicon” and the so-called “grammar.” Part one reviews some of the basic assumptions of generative transformational grammar, specifically the notions of core, periphery, principles, parameters, and the idealized speaker (Chomsky 1965, 1981). Part two compares how Construction Grammar differs from generative transformational grammar in a number of important ways: strict separation between lexicon and syntax, derivations, levels of representation, the linguistic sign, and modularity (Goldberg 2006). Based on this comparison, part three presents a number of case studies illustrating the methodologies and procedures employed in constructional research, including the use of corpora, the identification of constructions, and the role of constructional networks in determining the range of a verb’s argument realizations. The final part of the talk addresses different strands of Construction Grammar, including Berkeley Construction Grammar, Sign-based Construction Grammar, Cognitive Construction Grammar, and Radical Construction Grammar (see contributions in Hoffmann and Trousdale 2013).
9. Using semantic frames for building online learners' dictionaries of German and English
How can Frame Semantics be applied to the learning of foreign languages? This talk discusses the German Frame-based Online Lexicon (G-FOL)(
http://coerll.utexas.edu/frames)“Deutsch imBlick” as its basis, G-FOL is in the process of creating dictionary entries for about 2,000 lexical units. Each of the dictionary entries is centered around the notion of a semantic frame, together with sentence templates, grammar notes, cultural notes, and contrastive examples that help the students learning German acquire their vocabulary more easily inside and outside of the classroom. We also discuss a number of online exercises and a crowd-sourcing strategy aimed at speeding up our workflow. Finally, we discuss how our approach can be employed for the teaching of other languages, and we show how the lexicon approach can be extended to also cover the teaching of grammatical constructions in the foreign language classroom (Boas et al. 2016).), which is currently under development at the University of Texas at Austin. The first part of the talk provides some necessary background information about the theory of Frame Semantics and the Berkeley FrameNet project, which seeks to build a lexical database for English (Fillmore and Baker 2010). The second part of the talk discusses how semantic frames derived on the basis of English can be applied for the description and analysis of other languages. Part three presents the structure and goals of G-FOL, which is currently in the process of building a learner’s dictionary for the first year of German as a second language students at the University of Texas. Using the vocabulary of the first year online German textbook.
10. The English and German resultative construction networks
This talk discusses how one grammatical phenomenon, the resultative construction, has both similar yet at the same time different properties in two related languages, namely English and German. Part one reviews the pertinent literature on resultatives, including event-structure and constructional approaches, showing that existing approaches rely on abstract mechanisms that have problems restricting the distribution of resultatives. Part two proposes a frame-based approach to resultatives that highlights the importance of so-called mini-constructions as the basis of constructional networks. On this alternative view, mini-constructions come with syntactic, lexical, semantic, and pragmatic specifications, including collocational restrictions that account for the distribution of so-called conventionalized resultatives. Non-conventionalized resultatives are licensed by analogy, based on existing conventionalized resultatives. A case study involving air emission verbs shows how members of specific verb classes exhibit different restrictions in the resultative. Finally, we discuss the differences between constructional resultative networks in English and German to provide further arguments in favor of the proposal that most resultatives are conventionalized and must therefore be captured at the level of mini-constructions.