We present easy-to-use retrieval focused multilingual sentence embedding models, made available on TensorFlow Hub. The models embed text from 16 languages into a shared semantic space using a multi-task trained dualencoder that learns tied cross-lingual representations via translation bridge tasks (Chidambaram et al., 2018). The models achieve a new state-of-the-art in performance on monolingual and cross-lingual semantic retrieval (SR). Competitive performance is obtained on the related tasks of translation pair bitext retrieval (BR) and retrieval question answering (ReQA). On transfer learning tasks, our multilingual embeddings approach, and in some cases exceed, the performance of English only sentence embeddings.

No abstract

Popular QA benchmarks like SQuAD have driven progress on the task of identifying answer spans within a specific passage, with models now surpassing human performance. However, retrieving relevant answers from a huge corpus of documents is still a challenging problem, and places different requirements on the model architecture. There is growing interest in developing scalable answer retrieval models trained end-to-end, bypassing the typical document retrieval step. In this paper, we introduce Retrieval Question-Answering (ReQA), a benchmark for evaluating large-scale sentence-level answer retrieval models. We establish baselines using both neural encoding models as well as classical information retrieval techniques. We release our evaluation code to encourage further work on this challenging task.

In Natural Language Processing (NLP), topic modeling is the technique to extract abstract information from documents with huge amount of text. This abstract information leads towards the identification of the topics in the document. One way to retrieve topics from documents is keyphrase extraction. Keyphrases are a set of terms which represent high level description of a document. Different techniques of keyphrase extraction for topic prediction have been proposed for multiple languages i.e. English, Arabic, etc. However, this area needs to be explored for other languages e.g. Urdu. Therefore, in this paper, a novel unsupervised approach for topic prediction for Urdu language has been introduced which is able to extract more significant information from the documents. For this purpose, the proposed TOP-Rank system extracts keywords from the document and ranks them according to their position in a sentence. These keywords along with their ranking scores are utilized to generate keyphrases by applying syntactic rules to extracts more meaningful topics. These keyphrases are ranked according to the keywords scores and re-ranked with respect to their positions in the document. Finally, our proposed model identifies topranked keyphrases as topical significance and keyphrase with the highest score is selected as the topic of the document. Experiments are performed on two different datasets and performance of the proposed system is compared with existing state-of-the-art techniques. Results have shown that our proposed system outperforms existing techniques and holds the ability to produce more meaningful topics.

RemoAct is a wearable depth sensing and projection system that enables interaction on many surfaces. It makes interaction with the environment more intuitive through sharing and sending data with surrounding devices by applying certain gestures. This system offers a mobile and intuitive solution for interacting using a projected surface on habitual flat surfaces. Every user has their public and private areas, where the user can create tiles on the fly and share it with others and these public tiles are shown to other users through augmented reality. Interaction is made through hand gestures, finger tracking and hand tracking. This gives the user more freedom in movement. Different experiments were conducted to calculate the accuracy and RemoAct ran against different conventional methods to compare its accuracy, time and user experience. RemoAct takes less time for two users to draw one chart. As the system enables the users to work simultaneously, it reduces the needed time, short compared to successive drawing. For gesture recognition, accuracy reached 90-95\%. Object recognition and face identification accuracy varied with the variation of light.