Multi-layer multi-head self-attention mechanism is widely applied in modern neural language models. Attention redundancy has been observed among attention heads but has not been deeply studied in the literature. Using BERT-base model as an example, this paper provides a comprehensive study on attention redundancy which is helpful for model interpretation and model compression. We analyze the attention redundancy with Five-Ws and How. (What) We define and focus the study on redundancy matrices generated from pre-trained and fine-tuned BERT-base model for GLUE datasets. (How) We use both token-based and sentence-based distance functions to measure the redundancy. (Where) Clear and similar redundancy patterns (cluster structure) are observed among attention heads. (When) Redundancy patterns are similar in both pre-training and fine-tuning phases. (Who) We discover that redundancy patterns are task-agnostic. Similar redundancy patterns even exist for randomly generated token sequences. (“Why”) We also evaluate influences of the pre-training dropout ratios on attention redundancy. Based on the phase-independent and task-agnostic attention redundancy patterns, we propose a simple zero-shot pruning method as a case study. Experiments on fine-tuning GLUE tasks verify its effectiveness. The comprehensive analyses on attention redundancy make model understanding and zero-shot model pruning promising.

Simultaneous speech-to-speech translation is an extremely challenging but widely useful scenario that aims to generate target-language speech only a few seconds behind the source-language speech. In addition, we have to continuously translate a speech of multiple sentences, but all recent solutions merely focus on the single-sentence scenario. As a result, current approaches will accumulate more and more latencies in later sentences when the speaker talks faster and introduce unnatural pauses into translated speech when the speaker talks slower. To overcome these issues, we propose Self-Adaptive Translation which flexibly adjusts the length of translations to accommodate different source speech rates. At similar levels of translation quality (as measured by BLEU), our method generates more fluent target speech latency than the baseline, in both Zh<->En directions.