Kelantan and Sarawak Malay Dialects: Parallel Dialect Text Collection and Alignment Using Hybrid Distance-Statistical-Based Phrase Alignment Algorithm

Parallel texts corpora are essential resources especially in translation and multilingual information retrieval. However, the publicly available parallel text corpora are limited to certain types and domains. Besides, Malay dialects are not standardized in term of writing. The existing alignment algorithms that is used to analayze the writing will require a large training data to obtain a good result. The paper describes our methodology in acquiring a parallel text corpus of Standard Malay and Malay dialects, particularly Kelantan Malay and Sarawak Malay. Second, we propose a hybrid of distance-based and statistical-based alignment algorithm to align words and phrases of the parallel text. The proposed approach has a better precision and recall than the state-of-the-art GIZA++. In the paper, the alignment obtained were also compared to find out the lexical similarities and differences between SM and the two dialects.

Supervector Extraction for Encoding Speaker and Phrase Information with Neural Networks for Text-Dependent Speaker Verification

In this paper, we propose a new differentiable neural network with an alignment mechanism for text-dependent speaker verification. Unlike previous works, we do not extract the embedding of an utterance from the global average pooling of the temporal dimension. Our system replaces this reduction mechanism by a phonetic phrase alignment model to keep the temporal structure of each phrase since the phonetic information is relevant in the verification task. Moreover, we can apply a convolutional neural network as front-end, and, thanks to the alignment process being differentiable, we can train the network to produce a supervector for each utterance that will be discriminative to the speaker and the phrase simultaneously. This choice has the advantage that the supervector encodes the phrase and speaker information providing good performance in text-dependent speaker verification tasks. The verification process is performed using a basic similarity metric. The new model using alignment to produce supervectors was evaluated on the RSR2015-Part I database, providing competitive results compared to similar size networks that make use of the global average pooling to extract embeddings. Furthermore, we also evaluated this proposal on the RSR2015-Part II. To our knowledge, this system achieves the best published results obtained on this second part.

A Civil Code Article Information Retrieval System based on Phrase Alignment with Article Structure Analysis and Ensemble Approach

In this paper, we introduce a system for COLIEE task phase 1 that retrieves relevant civil code article(s) for making correct entailment to the questions of Japanese Bar Exam. This system is an extended version of our previous system that based on legal terminology and civil code article structure. However, the performance of the previous system is not as good as best performance system of the task. In this paper, we introduce concept of phrase alignment that takes into account the civil code article structure. In addition, due to the variations of the question types, the settings that are good for particular type of questions may not be good for other types of questions. Therefore, we propose to use systems with different settings and generate final answer by aggregating the output of different systems based on ensemble approach. Finally, we also discuss the difference between English task and Japanese task based on the retrieval results of Indri, one of the state-of-the-art information retrieval system.