Improving Neural Machine Translation with AMR Semantic Graphs

<div>Experimental results on the effect of heads <svg height="8.68572pt" id="M155" style="vertical-align:-0.0498209pt" version="1.1" viewbox="-0.0498162 -8.6359 11.0606 8.68572" width="11.0606pt" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink"><g transform="matrix(.013,0,0,-0.013,0,0)"><path d="M828 650H570L564 622C646 614 650 609 634 524L604 366H253L281 524C296 608 308 615 382 622L388 650H132L126 622C211 615 214 609 198 524L121 122C106 42 102 35 23 28L17 0H276L282 28C196 35 191 40 206 122L243 324H597L559 123C544 42 537 35 452 28L446 0H710L716 28C632 35 628 43 643 122L719 524C735 610 741 615 822 622L828 650Z"></path></g></svg> of the graph attention and the depth of the LightConvSeq2Seq-AMR model. (a) LightConvSeq2Seq-AMR with two graph coding layers. (b) The model with three layers.</div>

Mathematical Problems in Engineering

Improving Neural Machine Translation with AMR Semantic Graphs

Figure 6