{"id":{"repo_id":"cambridge","oai_identifier":"oai:www.repository.cam.ac.uk:1810/380613"},"canonical_url":"https://search.dev.ndltd.org/etd/cambridge/oai:www.repository.cam.ac.uk:1810/380613","repository":{"repo_id":"cambridge","name":"Cambridge University","base_url":"https://api.repository.cam.ac.uk/server/oai/request"},"display":{"title":"Modelling Metaphorical Extension","abstract":"Humans have an uncanny ability to push the semantics of words beyond their boundaries. This ability manifests itself clearly when people extend words metaphorically. Computational models of metaphorical extension are rare, in large part due to limitations with the methodology that is currently used to treat metaphor. Existing computational work treats metaphor as a property belonging to tokens in a sentence. However, knowing whether or not a word is metaphorical tells you nothing about what the metaphor means, and this information is critical if we are to build models of metaphorical extension. To address this, I put forward a new formulation of metaphor, in which metaphoricity is treated as a property of word senses rather than tokens. First, I explore the task of identifying metaphorical senses in the lexicon. I develop a model for this task, which classifies senses in the English WordNet as metaphorical or literal. To train this model, I design a procedure that combines token-based metaphor identification with word sense disambiguation. I collect evaluation data for this task and find that my model significantly outperforms an existing high-performing token-based metaphor identification model. Labelling senses as metaphorical or literal does not fully describe the ways in which a word's senses can be related. In particular, successive sense extensions form complex chain-like structures. To capture these structures, I develop a more sophisticated formalism of sense metaphoricity, in which metaphor is not a property of a sense but instead a relation between a pair of senses. Using this formalism, I collect a new annotation layer for WordNet which I call ChainNet. ChainNet reveals how the senses of a word are derived from one another: every nominal sense of a word is either connected to another sense by metaphor or metonymy, or is disconnected in the case of homonymy. WordNet senses have preexisting links to grounded information that captures their meaning, including images and textual examples. ChainNet identifies which of these senses are metaphorical, and therefore makes it possible to use this grounded information to study metaphorical extension. To this end, I define a new task, novel sense generation. In this task, a model is exposed to a multimodal representation of an unseen concept, and must extend a word to describe it. I create evaluation datasets not only for metaphor, but also for six other types of sense extension: metonymy, semantic drift, slang, jargon, zero derivation, and overextension. I demonstrate the feasibility of novel sense generation by evaluating a simple cognitive model and two deep neural architectures: a perceptron and a transformer. The perceptron outperformed both of the other models across all seven phenomena. This thesis establishes the methodological foundations that are needed to computationally study metaphorical extension. In the future, this methodology could be extended to study other lexical phenomena, or could be applied to other languages in order to study cognitive variation.","abstract_html":"Humans have an uncanny ability to push the semantics of words beyond their boundaries. This ability manifests itself clearly when people extend words metaphorically. Computational models of metaphorical extension are rare, in large part due to limitations with the methodology that is currently used to treat metaphor. Existing computational work treats metaphor as a property belonging to tokens in a sentence. However, knowing whether or not a word is metaphorical tells you nothing about what the metaphor means, and this information is critical if we are to build models of metaphorical extension. To address this, I put forward a new formulation of metaphor, in which metaphoricity is treated as a property of word senses rather than tokens. First, I explore the task of identifying metaphorical senses in the lexicon. I develop a model for this task, which classifies senses in the English WordNet as metaphorical or literal. To train this model, I design a procedure that combines token-based metaphor identification with word sense disambiguation. I collect evaluation data for this task and find that my model significantly outperforms an existing high-performing token-based metaphor identification model. Labelling senses as metaphorical or literal does not fully describe the ways in which a word's senses can be related. In particular, successive sense extensions form complex chain-like structures. To capture these structures, I develop a more sophisticated formalism of sense metaphoricity, in which metaphor is not a property of a sense but instead a relation between a pair of senses. Using this formalism, I collect a new annotation layer for WordNet which I call ChainNet. ChainNet reveals how the senses of a word are derived from one another: every nominal sense of a word is either connected to another sense by metaphor or metonymy, or is disconnected in the case of homonymy. WordNet senses have preexisting links to grounded information that captures their meaning, including images and textual examples. ChainNet identifies which of these senses are metaphorical, and therefore makes it possible to use this grounded information to study metaphorical extension. To this end, I define a new task, novel sense generation. In this task, a model is exposed to a multimodal representation of an unseen concept, and must extend a word to describe it. I create evaluation datasets not only for metaphor, but also for six other types of sense extension: metonymy, semantic drift, slang, jargon, zero derivation, and overextension. I demonstrate the feasibility of novel sense generation by evaluating a simple cognitive model and two deep neural architectures: a perceptron and a transformer. The perceptron outperformed both of the other models across all seven phenomena. This thesis establishes the methodological foundations that are needed to computationally study metaphorical extension. In the future, this methodology could be extended to study other lexical phenomena, or could be applied to other languages in order to study cognitive variation.","abstract_has_math":false,"creators":["Hall Maudslay, Rowan"],"institution":"University of Cambridge","degree_name":"Doctor of Philosophy (PhD)","degree_level":"Doctoral","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Teufel, Simone"],"committee_chairs":[],"committee_members":[],"year":2024,"date_issued":"2024-11-29","date_published":"2024-11-29","updated_at":"2026-07-22T22:24:01Z","subjects":["computational linguistics","lexicography","metaphor","metonymy","natural language processing","WordNet"],"languages":["eng"],"rights":[],"rights_urls":["https://www.repository.cam.ac.uk/bitstreams/99670643-8944-40a2-9247-1d1a35064563/download","http://purl.org/NET/rdflicense/allrightsreserved"],"identifier_entries":[]},"links":{"outbound_url":"https://doi.org/10.17863/CAM.116178","outbound_label":"DOI","outbound_source":"dc:identifier.doi"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Teufel, Simone"]},{"key":"dc:contributor.sponsor","label":"Sponsor","values":["Leslie Wilson Vice Chancellor's Scholarship Alan Turing Institute Enrichment Award Isaac Newton Trust Language Sciences Incubator Fund Magdalene College Bye-Fellowship"]},{"key":"dc:creator","label":"Author","values":["Hall Maudslay, Rowan"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2024-11-29"]},{"key":"dc:publisher.institution","label":"Dc Publisher Institution","values":["University of Cambridge"]},{"key":"dc:relation.isreferencedby.uri","label":"Dc Relation Isreferencedby URI","values":["https://www.repository.cam.ac.uk/handle/1810/380613"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"dc:type.qualificationlevel","label":"Dc Type Qualificationlevel","values":["Doctoral"]},{"key":"dc:type.qualificationname","label":"Dc Type Qualificationname","values":["Doctor of Philosophy (PhD)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["computational linguistics","lexicography","metaphor","metonymy","natural language processing","WordNet"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["https://www.repository.cam.ac.uk/bitstreams/99670643-8944-40a2-9247-1d1a35064563/download","http://purl.org/NET/rdflicense/allrightsreserved"]},{"key":"dc:rights.embargodate","label":"Dc Rights Embargodate","values":["2028-02-25"]},{"key":"dc:rights.embargotype","label":"Dc Rights Embargotype","values":["embargo"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.doi","label":"DOI","values":["https://doi.org/10.17863/CAM.116178"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://www.repository.cam.ac.uk/bitstreams/12abeb9f-3d5d-47f1-bc0c-a6beaa3e9118/download"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Humans have an uncanny ability to push the semantics of words beyond their boundaries. This ability manifests itself clearly when people extend words metaphorically. Computational models of metaphorical extension are rare, in large part due to limitations with the methodology that is currently used to treat metaphor. Existing computational work treats metaphor as a property belonging to tokens in a sentence. However, knowing whether or not a word is metaphorical tells you nothing about what the metaphor means, and this information is critical if we are to build models of metaphorical extension. To address this, I put forward a new formulation of metaphor, in which metaphoricity is treated as a property of word senses rather than tokens. First, I explore the task of identifying metaphorical senses in the lexicon. I develop a model for this task, which classifies senses in the English WordNet as metaphorical or literal. To train this model, I design a procedure that combines token-based metaphor identification with word sense disambiguation. I collect evaluation data for this task and find that my model significantly outperforms an existing high-performing token-based metaphor identification model. Labelling senses as metaphorical or literal does not fully describe the ways in which a word's senses can be related. In particular, successive sense extensions form complex chain-like structures. To capture these structures, I develop a more sophisticated formalism of sense metaphoricity, in which metaphor is not a property of a sense but instead a relation between a pair of senses. Using this formalism, I collect a new annotation layer for WordNet which I call ChainNet. ChainNet reveals how the senses of a word are derived from one another: every nominal sense of a word is either connected to another sense by metaphor or metonymy, or is disconnected in the case of homonymy. WordNet senses have preexisting links to grounded information that captures their meaning, including images and textual examples. ChainNet identifies which of these senses are metaphorical, and therefore makes it possible to use this grounded information to study metaphorical extension. To this end, I define a new task, novel sense generation. In this task, a model is exposed to a multimodal representation of an unseen concept, and must extend a word to describe it. I create evaluation datasets not only for metaphor, but also for six other types of sense extension: metonymy, semantic drift, slang, jargon, zero derivation, and overextension. I demonstrate the feasibility of novel sense generation by evaluating a simple cognitive model and two deep neural architectures: a perceptron and a transformer. The perceptron outperformed both of the other models across all seven phenomena. This thesis establishes the methodological foundations that are needed to computationally study metaphorical extension. In the future, this methodology could be extended to study other lexical phenomena, or could be applied to other languages in order to study cognitive variation."]},{"key":"dc:format.checksum.md5","label":"Dc Format Checksum Md5","values":["a4438bc86f7b3e8a35e00763c06a0b9c","87eda9de84448d1f82354d60eee3eb5f"]},{"key":"dc:title","label":"Title","values":["Modelling Metaphorical Extension"]}]}],"canonical_facts":{"dc:contributor.advisor":["Teufel, Simone"],"dc:contributor.sponsor":["Leslie Wilson Vice Chancellor's Scholarship Alan Turing Institute Enrichment Award Isaac Newton Trust Language Sciences Incubator Fund Magdalene College Bye-Fellowship"],"dc:creator":["Hall Maudslay, Rowan"],"dc:date.issued":["2024-11-29"],"dc:description.abstract":["Humans have an uncanny ability to push the semantics of words beyond their boundaries. This ability manifests itself clearly when people extend words metaphorically. Computational models of metaphorical extension are rare, in large part due to limitations with the methodology that is currently used to treat metaphor. Existing computational work treats metaphor as a property belonging to tokens in a sentence. However, knowing whether or not a word is metaphorical tells you nothing about what the metaphor means, and this information is critical if we are to build models of metaphorical extension. To address this, I put forward a new formulation of metaphor, in which metaphoricity is treated as a property of word senses rather than tokens. First, I explore the task of identifying metaphorical senses in the lexicon. I develop a model for this task, which classifies senses in the English WordNet as metaphorical or literal. To train this model, I design a procedure that combines token-based metaphor identification with word sense disambiguation. I collect evaluation data for this task and find that my model significantly outperforms an existing high-performing token-based metaphor identification model. Labelling senses as metaphorical or literal does not fully describe the ways in which a word's senses can be related. In particular, successive sense extensions form complex chain-like structures. To capture these structures, I develop a more sophisticated formalism of sense metaphoricity, in which metaphor is not a property of a sense but instead a relation between a pair of senses. Using this formalism, I collect a new annotation layer for WordNet which I call ChainNet. ChainNet reveals how the senses of a word are derived from one another: every nominal sense of a word is either connected to another sense by metaphor or metonymy, or is disconnected in the case of homonymy. WordNet senses have preexisting links to grounded information that captures their meaning, including images and textual examples. ChainNet identifies which of these senses are metaphorical, and therefore makes it possible to use this grounded information to study metaphorical extension. To this end, I define a new task, novel sense generation. In this task, a model is exposed to a multimodal representation of an unseen concept, and must extend a word to describe it. I create evaluation datasets not only for metaphor, but also for six other types of sense extension: metonymy, semantic drift, slang, jargon, zero derivation, and overextension. I demonstrate the feasibility of novel sense generation by evaluating a simple cognitive model and two deep neural architectures: a perceptron and a transformer. The perceptron outperformed both of the other models across all seven phenomena. This thesis establishes the methodological foundations that are needed to computationally study metaphorical extension. In the future, this methodology could be extended to study other lexical phenomena, or could be applied to other languages in order to study cognitive variation."],"dc:format.checksum.md5":["a4438bc86f7b3e8a35e00763c06a0b9c","87eda9de84448d1f82354d60eee3eb5f"],"dc:identifier.doi":["https://doi.org/10.17863/CAM.116178"],"dc:identifier.uri":["https://www.repository.cam.ac.uk/bitstreams/12abeb9f-3d5d-47f1-bc0c-a6beaa3e9118/download"],"dc:language":["eng"],"dc:publisher.institution":["University of Cambridge"],"dc:relation.isreferencedby.uri":["https://www.repository.cam.ac.uk/handle/1810/380613"],"dc:rights":["https://www.repository.cam.ac.uk/bitstreams/99670643-8944-40a2-9247-1d1a35064563/download","http://purl.org/NET/rdflicense/allrightsreserved"],"dc:rights.embargodate":["2028-02-25"],"dc:rights.embargotype":["embargo"],"dc:subject":["computational linguistics","lexicography","metaphor","metonymy","natural language processing","WordNet"],"dc:title":["Modelling Metaphorical Extension"],"dc:type":["Thesis"],"dc:type.qualificationlevel":["Doctoral"],"dc:type.qualificationname":["Doctor of Philosophy (PhD)"]},"updated_at":"2026-07-22T22:24:01Z"}