13197 research outputs found
Sort by
Disentangling Pretrained Representation to Leverage Low-Resource Languages in Multilingual Machine Translation
No full textMultilingual neural machine translation aims to encapsulate multiple languages into a single model. However, it requires an enormous dataset, leaving the low-resource language (LRL) underdeveloped. As LRLs may benefit from shared knowledge of multilingual representation, we aspire to find effective ways to integrate unseen languages in a pre-trained model. Nevertheless, the intricacy of shared representation among languages hinders its full utilisation. To resolve this problem, we employed target language prediction and a central language-aware layer to improve representation in integrating LRLs. Focusing on improving LRLs in the linguistically diverse country of Indonesia, we evaluated five languages using a parallel corpus of 1,000 instances each, with experimental results measured by BLEU showing zero-shot improvement of 7.4 from the baseline score of 7.1 to a score of 15.5 at best. Further analysis showed that the gains in performance are attributed more to the disentanglement of multilingual representation in the encoder with the shift of the target language-specific representation in the decoder.conference pape
Comparing Likert Scale and Pairwise Comparison for Human Evaluation in Rapport-Building Dialogue Systems
Get PDFHuman evaluation plays a critical role in dialogue systems research, especially in non-task-oriented systems such as rapport-building dialogue systems. Current evaluations often rely on Likert scales to assess user experience, but this method introduces challenges such as inconsistent scale perception, inefficiency, and central tendency bias. Moreover, it is difficult to compare the agent's performance across multiple criteria due to the problem of uneven scoring interpretations by participants on the Likert scale. On the other hand, pairwise comparison emphasizes direct item-to-item evaluation based on defined criteria, producing scores that more closely align with participants' preferences and minimizing biases. This paper compares an evaluation framework for rapport-building dialogue systems using pairwise comparison with a conventional Likert scale system. These approaches are tested through dialogue experiments involving six participants and four dialogue systems embedded in a conversational robot: CommA, CommI, CommO, and CommE, to measure human-agent rapport. Our experimental results indicated that the pairwise comparison method better represented systems' overall performance compared to the Likert scale. It also demonstrated lower variability, higher reliability, and a shorter completion time.journal articl
Advancing Primate Behavior Analysis: Developing a Unified Dataset and Multi-Animal Tracking System for Japanese Macaques and Broader Primate Species
Get PDF奈良先端科学技術大学院大学博士(工学)doctoral thesi
Adaptive Cognitive Behavior Therapy With a Virtual Agent Considering User’s Psychological Distress
Get PDF奈良先端科学技術大学院大学博士(工学)doctoral thesi
A Study on Adaptive and Robust Privacy-Enhancing Technologies for Spatio-Temporal Data Aggregation
Get PDF奈良先端科学技術大学院大学博士(工学)doctoral thesi
Transcriptome-wide identification of substrates of the Drosophila RNA export factor Exportin5
Get PDF奈良先端科学技術大学院大学博士(バイオサイエンス)doctoral thesi
Estimating Gene Regulatory Network in Vertebrate Development Based on LASSO Regression
Get PDF奈良先端科学技術大学院大学博士(工学)doctoral thesi
Cellular uptake of spontaneously formed lipid nanodiscs toward their molecular delivery application
Get PDF奈良先端科学技術大学院大学博士(工学)doctoral thesi
バーチャル ロボット トノ ニチジョウテキ テキスト タイワ ノ LLM オ モチイタ コウリツカ ト シャカイ ジッケン オ トオシタ タイワ ケイゾク ヒョウカ
Get PDF奈良先端科学技術大学院大学修士(工学)master thesi