Towards Making a Dependency Parser See
Towards Making a Dependency Parser See
复制标题
DOI：
10.18653/v1/d19-1160
复制
发表时间：
2019-09
期刊：
影响因子：
--
通讯作者：
Michalina Strzyz;David Vilares;Carlos Gómez-Rodríguez
中科院分区：
文献类型：
--
作者：
Michalina Strzyz;David Vilares;Carlos Gómez-Rodríguez
我们探索是否有可能在RNN依赖解析器（英语）中利用眼动跟踪数据，当这些信息仅在训练期间可用时-即在推理时不使用聚合或标记级凝视特征。为此，我们训练了一个多任务学习模型，该模型将句子解析为序列标签，并利用凝视特征作为辅助任务。我们的方法还学习从不相交的数据集进行训练，即它可以用于测试已经收集的凝视特征是否有助于提高新的非凝视注释树库的性能。准确度的提高是适度的，但积极的，表明该方法的可行性。它可以作为架构的第一步，可以更好地利用眼动跟踪数据或其他仅用于训练句子的补充信息，可能会导致句法分析的改进。
We explore whether it is possible to leverage eye-tracking data in an RNN dependency parser (for English) when such information is only available during training - i.e. no aggregated or token-level gaze features are used at inference time. To do so, we train a multitask learning model that parses sentences as sequence labeling and leverages gaze features as auxiliary tasks. Our method also learns to train from disjoint datasets, i.e. it can be used to test whether already collected gaze features are useful to improve the performance on new non-gazed annotated treebanks. Accuracy gains are modest but positive, showing the feasibility of the approach. It can serve as a first step towards architectures that can better leverage eye-tracking data or other complementary information available only for training sentences, possibly leading to improvements in syntactic parsing.