Large vocabulary domain-agnostic Automatic Speech Recognition (ASR) systems often mistranscribe domain-specific words and phrases. Since these generic ASR systems are the first component of most voice assistants in production, building Natural Language Understanding (NLU) systems that are robust to these errors can be a challenging task. In this paper, we focus on handling ASR errors in named entities, specifically person names, for a voicebased collaboration assistant. We demonstrate an effective method for resolving person names that are mistranscribed by black-box ASR systems, using character and phonemebased information retrieval techniques and contextual information, which improves accuracy by 40.8% on our production system. We provide a live interactive demo to further illustrate the nuances of this problem and the effectiveness of our solution. 1

In this paper, we investigate the tendency of end-to-end neural Machine Reading Comprehension (MRC) models to match shallow patterns rather than perform inference-oriented reasoning on RC benchmarks. We aim to test the ability of these systems to answer questions which focus on referential inference. We propose ParallelQA, a strategy to formulate such questions using parallel passages. We also demonstrate that existing neural models fail to generalize well to this setting. Johannes Welbl, Pontus Stenetorp, and Sebastian Riedel. 2017. Constructing datasets for multi-hop reading comprehension across documents. arXiv preprint arXiv:1710.06481 . . 2015. Towards ai-complete question answering: A set of prerequisite toy tasks. arXiv preprint arXiv:1502.05698 . Adina Williams, Nikita Nangia, and Samuel R Bowman. 2017. A broad-coverage challenge corpus for sentence understanding through inference. arXiv preprint arXiv:1704.05426 . Zhilin Yang, Junjie Hu, Ruslan Salakhutdinov, and William W Cohen. 2017. Semi-supervised qa with generative domain-adaptive nets. arXiv preprint arXiv:1702.02206 .

When a set of genes are identified to be related to a disease, say through gene expression analysis, it is common to examine the average distance among their protein products in the human interactome as a measure of biological relatedness of these genes. The reasoning for this is that, genes associated with a disease would tend to be functionally related, and that functionally related genes would be closely connected to each other in the interactome. Typically, average shortest path length (ASPL) of disease genes* is compared to ASPL of randomly selected genes or to ASPL in a randomly permuted network. We examined whether the ASPL of a set of genes is indeed a good measure of biological relatedness or whether it is simply a characteristic of the degree distribution of those genes. We examined the ASPL of genes sets of some disease and pathway associations and compared them to ASPL of three types of randomly selected control sets: uniform selection, from entire proteome, degree-matched selection and random permutation of the network. We found that disease associated genes and their degree-matched random genes have comparable ASPL. In other words, ASPL is a characteristic of the degree of the genes and the network topology, and not that of functional coherence.

No abstract

Abstractive summarization has been explored only to some extent in recent years in English, Japanese and other foreign languages. This paper shows that abstraction can be accomplished for Indian Languages, specifically Kannada, using guided summarization approach. The sArAmsha system involves analyzing the given Kannada document and performing parts of speech tagging and stemming operations, identification of named entities such as person, location and date, usage of abstraction schemes and Information Extraction (IE) rules created by hand to extract information and finally, creation of a summary by forming sentences based on domain templates. The method can be expanded to a large variety of categories and complexities to deal with more aspects in the future.