Thesis
Cultural machine translation using MRF Gibbs model & Bayesian learning
Master of Science (M.S.), Drexel University
Jun 2017
DOI:
https://doi.org/10.17918/etd-7434
Abstract
In this thesis, we introduce a novel Gibbs language model constructed on a multi-layered dependency semantic graph to lexically disambiguate word, phrases, and sentences that lend themselves to different possible meaning and interpretations for use in machine translation (MT). The model looks at semantic cliques of words (key words) and assigns Gibbs potentials and conditional probabilities in proportion to the importance and degree of interactions between a given word and its neighbors within the semantic cliques. Efficient estimates (maximum likelihood estimators (MLE)) for the Gibbs clique parameters are obtained using bilingual parallel corpora. Our method also naturally factors in the beliefs of expert translators, maps them into expert Gibbs parameters, and updates the MLE to maximum a posterior probability (MAP) estimates. Experimental results using our model and method are reported on testbeds in the medical and literary fiction domains and our results fare more than favorably when compared to the state-of-the-art long short-term memory (LSTM) Neural Network (NN) approach.
Metrics
43 File views/ downloads
24 Record Views
Details
- Title
- Cultural machine translation using MRF Gibbs model & Bayesian learning
- Creators
- Zheng Zhong - DU
- Contributors
- Fernand S. Cohen (Advisor) - Drexel University (1970-)
- Awarding Institution
- Drexel University
- Degree Awarded
- Master of Science (M.S.)
- Publisher
- Drexel University; Philadelphia, Pennsylvania
- Number of pages
- vii, 42 pages
- Resource Type
- Thesis
- Language
- English
- Academic Unit
- College of Engineering (1970-2026); Electrical (and Computer) Engineering (1970-2026); Drexel University
- Other Identifier
- 7434; 991014632226704721