M Tech Dissertations

Permanent URI for this collectionhttp://ir.daiict.ac.in/handle/123456789/3

Browse

Filters

2020 - 20212

Settings

Subject: search.filters.subject.Transfer learning

Search Results

Now showing 1 - 2 of 2
  • Thumbnail Image
    ItemOpen Access
    Development of Neural Machine Translation Systems for Indian Languages
    (2021) Prajapati, Raj B.; Majumder, Prasenjit
    Machine Translation has became a very promising field with the recent advent of Neural Machine Translation (NMT) systems.A lot of work has been done in a small span of time, starting from bilingual models to complex multilingual models which can incorporate many languages at once. Building NMT systems for Indian languages is not an easy task and we often have to embed linguistic information to make it more robust and effective.The literature survey enlists different neural machine translation approaches as well as some research done on direct speech to speech translation. However, in the recent years the standard practices in NMT which are supervised in nature have not been upto mark so semi-supervised and unsupervised efforts have also been in existence. In this report we have compiled all the efforts and experiments done so far to develop neural machine translation systems for Indian languages and related components , starting from baseline modelling to use of feature injections , applying transfer learning, use of language models
  • Thumbnail Image
    ItemOpen Access
    Augmenting dialogue generation using dialogue act embeddings: a transfer learning approach
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2020) Bisht, Abhimanyu Singh; Majumder, Prasenjit
    The following work looks at contemporary end-to-end dialogue systems with the aim of improving dialogue generation in an open-domain setting. It provides an overview of popular literature in the domain of dialogue generation, followed by a brief look at how human dialogue is understood from the perspective of Linguistics and Cognitive Science. We try to extract useful ideas from these domains of research and implement them in a transfer learning approach where a pretrained language model is supplemented with dialogue act information using special embeddings. The hypothesis behind the proposed approach is that the dialogue act information will aid the generation process. The proposed approach is then compared with a baseline approach on their performance on the DailyDialog[12] dataset using perplexity as the evaluation metric. Though the proposed approach is a significant improvement over the baseline, the contribution of the Dialogue Act Embeddings in the development is shown to be marginal via ablation analysis.