Browsing by Subject "lexical semantics"

Sort by: Order: Results:

Now showing items 1-3 of 3
  • Narkevich, Dmitry (Helsingin yliopisto, 2021)
    Hypernymy is a relationship between two words, where the hyponym carries a more specific meaning, and entails a hypernym that carries a more general meaning. A particular kind of verbal hypernymy is troponymy, where troponyms are verbs that encode a particular manner or way of doing something, such as “whisper” meaning “to speak in a quiet manner”. Recently, contextualized word vectors have emerged as a powerful tool for representing the semantics of words in a given context, in contrast to earlier static embeddings where every word is represented by a single vector regardless of sense. BERT, a pre-trained language model that uses contextualized word representations, achieved state of the art performance on various downstream NLP tasks such as question answering. Previous research identified knowledge of scalar adjective intensity in BERT, but not systematic knowledge of nominal hypernymy. In this thesis, we investigate systematic knowledge of troponymy and verbal hypernymy in the base English version of BERT. We compare the similarity of vector representations for manner verbs and adverbs of interest, to see if troponymy is represented in the vector space. Then, we evaluate BERT’s predictions for cloze tasks involving troponymy and verbal hypernymy. We also attempt to train supervised models to probe vector representations for this knowledge. Lastly, we perform clustering analyses on vector representations of words in hypernymy pairs. Data on troponymy and hypernymy relationships is extracted from WordNet and HyperLex, and sentences containing instances of the relevant words are obtained from the ukWaC corpus. We were unable to identify any systematic knowledge about troponymy and verb hypernymy in BERT. It was reasonably successful at predicting hypernyms in the masking experiments, but a general inability to go in the other direction suggests that this knowledge is not systematic. Our probing models were unsuccessful at recovering information related to hypernymy and troponymy from the representations. In contrast with previous work that finds type-level semantic information to be located in the lower layers of BERT, our cluster-based analyses suggest that the upper layers contain stronger or more accessible representations of hypernymy.
  • Koskinen, Anne (Helsingin yliopisto, 2021)
    In the linguistic research of the last decades one can observe a development towards image-centricity. However, there is no clarity about what is meant by the word image. The varying lexicalization of the complex phenomenon in different languages (Bild in German, image/picture in English) additionally contributes to the confusion. Building on the idea of linguistics as cultural studies, in this study the concept of the image is approached as a cultural phenomenon constituted on a linguistic basis. Following the objective of contrastive lexical semantics, the goal of the work is to investigate language-bound differences in image concepts and word fields, and to identify lexical/conceptual gaps in German- and English-language image theories. Due to the traditionally central role of the German language in philosophy and the long tradition of German-language image research (Bildwissenschaft), the German concept of Bild (image/picture) has a special relevance, which also stimulates international discussions about the German word Bild. As can be seen from the research literature, words are not only used as a vehicle of theorization, but the image sciences (Bildwissenschaften) deal intensively with various aspects of the word Bild, its etymology, word formation, semantics, and word history as well as its equivalents in English and other languages. In order to trace the lexical influences and semantic problems, the considerations of the image theorists are supplemented and commented on with linguistic sources. As a general summary of Chapter 2, it could be said that etymologies and word histories offer a (with time growing) selection of word (sub) meanings for various theoretical needs, which can be used as required to emphasize suitable readings and thus etymologically substantiate one's own argumentation, so to speak. Subsequently, the question of an intuitively used word field concept arises, which serves as an invitation to further discussion of the lexeme Bild from the point of view of lexicology and word field theory using on corpus-based resources of contemporary German. Chapter 3 introduces the central concepts of lexical semantics and presents the background as well as fundamental methodological considerations and newer, especially ideologically critical approaches to word field research. Building on these thoughts, in Chapter 4 a philosophically oriented (i.e. not a conventional) word field analysis is carried out based on the procedure of word field modelling proposed by Lutzeier (2007) and Staffeldt (2017). In other words: the word field analysis is adapted for the linguistic foundation of image sciences. The complexity of the German concept of the image is reflected in the analysis of the selected sub-areas of the word field around Bild in the reading 'artistic representation' ('künstlerische Darstellung'). The role of encyclopedic knowledge in word field modelling is discussed using as an example the new digital art form of NFTs (Non-Fungible Tokens), which challenge not only the conventional understanding of the image/picture (Bild), but also legislation and the art market. This raises questions of image migration as well as the agency and subject status of the image, which require further consideration. An important challenge for research is to look not only at the different modes and media, but also at what surrounds the media: the system in which the media are embedded. In contrast to the considerations of the image theorists that sculptures can also be referred to with the word Bild in German, in the analyzed work "Blinder Gallerist" (König 2019), the word Bild is largely used 1) to refer to two-dimensional artworks, and 2) to be synonymic with the word Gemälde (painting). Thus, the use is in line with the discussed translations of the research texts as well as the results of the word field analysis. In the contemporary German language the word Bild in the reading 'artistic representation' is mainly used in the meaning 'painting' and Skulptur (sculpture) is to be understood as a cohyponym to Bild. How to deal with expressions for which there are no equivalents in other languages is a central challenge for a global image science. As demonstrated here, a comparison of language-related research questions may prove useful, since concepts that are diffuse in one language can be made clearer using another language. In addition to the Western concepts, the image science (Bildwissenschaft) discourse not only increasingly discusses selected aspects of image concepts outside the Indo-European language area, but also strives for a dialogue and polylogue, which counters the challenges of an inter-/transcultural and dia-/polylogical image science. The research must include not only translinguistic aspects, but also the multi- or transmodality of communication instead of building on a hegemony of one mode, medium, language, or scientific discipline. Thus, the desiderata that can be formulated as a result of this study can be described as translinguistic multimodality research.
  • Crane, Thera Marie; Fleisch, Axel Gustav (Finnish Oriental Society, 2019)
    Studia Orientalia
    Verbal lexical aspectual structure is a domain in which infinite meaning possibilities meet a closed set of grammatical categories. It is therefore a fruitful area for investigations of subtle cross-linguistic semantic differences, as well as of contact-induced semantic change. Bantu lexical aspectual systems typically include "complex" lexicalizations denoting both a coming-to-be phase and a resultant state (e.g. the same verb in different frame can encode both 'get angry' and 'be angry'). At least some Bantu lexical aspectual types are, therefore, difficult to account for using traditional classifications (e.g. Vendler 1957), and the tests frequently used to arrive at classifications may also not be applicable. In this paper, we describe our pilot study of lexical aspect in isiNdebele, a Bantu language of South Africa, and our comparative with a related South African language, Sindebele. We describe some of the tests we used, and suggest general guidelines for developing and applying tests of lexical aspect within and across languages. We also describe and illustrate the semi-structured interview process we used, showing that hybrid elicitation/ethnographic discussions are helpful in developing and appropriately applying tests for lexical aspect.