Kuiper, Jesper (2024) Using Conceptors to Extract Abstraction Hierarchies from Corpora of Natural Text: Combatting Word Polysemy Using Word Sense Disambiguation Techniques. Master's Thesis / Essay, Artificial Intelligence.
|
Text
mAI_2024_KuiperJS.pdf Download (1MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (125kB) |
Abstract
This thesis documents a project wherein a method was developed for extracting hierarchies of abstraction between concepts from large unlabeled corpora of text. To this end conceptors were employed, which includes a measure for estimating concept-to-concept abstraction. Previous research has focused on using conceptors of word embeddings to estimate word abstraction, but this fails to consider word polysemy. We therefore move from word-to-word abstraction to sense-to-sense abstraction by incorporating findings from the linguistic problem of word sense disambiguation. Secondly, a graded extension of the available binary conceptor abstraction measure was developed, to enable a finer-grained insight in abstraction values. This was integrated into a single abstraction destilation pipeline, on which initial experiments were conducted. Results have shown that conceptors are suitable for estimating abstraction relations between word senses, but for the present experiment need a considerable amount of exemplar occurrences before the abstraction estimation becomes accurate. Additionally, the word sense classification step seems to work, but this cannot be quantified in the current experiments due to a lack of quantitative accuracy measures on unlabeled data. Points for improvements revolve around detecting compounds, metaphors and open named entities. Finally, a discussion is held on how formal abstraction definitions might not align with human abstraction interpretation in discourse.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Jaeger, H. and Rij-Tange, J.C. van |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 23 Jan 2024 14:46 |
| Last Modified: | 23 Jan 2024 14:46 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/31844 |
Actions (login required)
 |
View Item |