The quality of parallel corpora used to train a Neural Machine Translation (NMT) model can critically influence the model's performance. Various approaches for refining parallel corpora have been introduced, but there is still much room for improvements, such as enhancing the efficiency and the quality of refinement. We introduce VANT, a novel visual analytics system for refining parallel corpora used in training an NMT model. Our system helps users to readily detect and filter noisy parallel corpora by (1) aiding the quality estimation of individual sentence pairs within the corpora by providing diverse quality metrics (e.g., cosine similarity, BLEU, length ratio) and (2) allowing users to visually examine and manage the corpora based on the pre-computed metrics scores. Our system's effectiveness and usefulness are demonstrated through a qualitative user study with eight participants, including four domain experts with real-world datasets.

We introduce Parallel Histogram Plot (PHP), a technique that overcomes the innate limitations of parallel coordinates plot (PCP) by attaching stacked-bar histograms with discrete color schemes to PCP. The color-coded histograms enable users to see an overview of the whole data without cluttering or scalability issues. Each rectangle in the PHP histograms is color coded according to the data ranking by a selected attribute. This color-coding scheme allows users to visually examine relationships between attributes, even between those that are displayed far apart, without repositioning or reordering axes. We adopt the Visual Information Seeking Mantra so that the polylines of the original PCP can be used to show details of a small number of selected items when the cluttering problem subsides. We also design interactions, such as a focus+context technique, to help users investigate small regions of interest in a space-efficient manner. We provide a real-world example in which PHP is effectively utilized compared with other visualizations, and we perform a controlled user study to evaluate the performance of PHP in helping users estimate the correlation between attributes. The results demonstrate that the performance of PHP was consistent in the estimation of correlations between two attributes regardless of the distance between them.

Most smartphones support multi-tasking with several means to switch between apps (e.g., a "recent apps" button or a "back" button). However, switching between apps is cumbersome when one has to do it frequently for example, when notifications keep interrupting one's current task. We introduce Peek-a-View, a fully transparent flipping screen cover that can reduce task switching overhead by providing an additional virtual screen space for subtasks. We assessed its feasibility in handling notifications. Upon receiving a notification, users can peek into the content of the notification without actually switching apps by slightly lifting the cover. If necessary, users can completely flip the cover to switch to the app that fired the notification. Two user studies showed that flipping and peeking interaction provided improved performance and proved to be useful for tasks that involve subtasks.