BERT-based rankers have been shown very effective as rerankers in information retrieval tasks. In order to extend these models to full-ranking scenarios, the ColBERT model has been recently proposed, which adopts a late interaction mechanism. This mechanism allows for the representation of documents to be precomputed in advance. However, the late-interaction mechanism leads to large index size, as one needs to save a representation for each token of every document. In this work, we focus on token pruning techniques in order to mitigate this problem. We test four methods, ranging from simpler ones to the use of a single layer of attention mechanism to select the tokens to keep at indexing time. Our experiments show that for the MS MARCO-passages collection, indexes can be pruned up to 70% of their original size, without a significant drop in performance. We also evaluate on the MS MARCO-documents collection and the BEIR benchmark, which reveals some challenges for the proposed mechanism.

The semantic matching capabilities of neural information retrieval can ameliorate synonymy and polysemy problems of symbolic approaches.However, neural models' dense representations are more suitable for re-ranking, due to their inefficiency.Sparse representations, either in symbolic or latent form, are more efficient with an inverted index.Taking the merits of the sparse and dense representations, we propose an ultra-high dimensional (UHD) representation scheme equipped with directly controllable sparsity.UHD's large capacity and minimal noise and interference among the dimensions allow for binarized representations, which are highly efficient for storage and search.Also proposed is a bucketing method, where the embeddings from multiple layers of BERT are selected/merged to represent diverse linguistic aspects.We test our models with MS MARCO and TREC CAR, showing that our models outperforms other sparse models.

We present Bendi, a shape-changing device for a tactile-visual phone conversation. Bendi enables users to deliver shape-changing movements (e.g., upward or downward bending, left or right tilting, and shrinking) from the user's joystick input to the other party's device in real time during phone conversations. We conducted a user study to observe how seven couples used it over three days in a coffeehouse. Our field trial of Bendi in a coffeehouse showed the private and natural uses, and integrated uses of tactile and visual expressions along with the uses of the vocabularies developed through Bendi. In addition, there were active uses even in negative and serious conversational context with its pleasant tactile feelings and movement representations. Lastly, we discuss issues for the future designs and real-world deployment of shape-changing mobile devices for daily use.

We introduce Wrigglo, a shape-changing smart phone peripheral that allows pairs of users to share wriggling movements with one another. Attached to a smart phone, Wrigglo captures the sender's motions and activates the receiver's Wrigglo which repeats the motion simultaneously. The result of our in-lab use observation with twelve couples showed that Wrigglo supported emotional and functional roles of body gestures and postures, creating vocabularies related to the motion of specific body parts and, to some extent, reflected the connected user's presence through the device's movement. Through its peripheral anthropomorphization, Wrigglo can deliver new forms of telepresence by embodied posturing and gesturing in mobile communication.