We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: Deep Neural Networks Evolve Human-like Attention Distribution during Reading Comprehension

Abstract: Attention is a key mechanism for information selection in both biological brains and many state-of-the-art deep neural networks (DNNs). Here, we investigate whether humans and DNNs allocate attention in comparable ways when reading a text passage to subsequently answer a specific question. We analyze 3 transformer-based DNNs that reach human-level performance when trained to perform the reading comprehension task. We find that the DNN attention distribution quantitatively resembles human attention distribution measured by fixation times. Human readers fixate longer on words that are more relevant to the question-answering task, demonstrating that attention is modulated by top-down reading goals, on top of lower-level visual and text features of the stimulus. Further analyses reveal that the attention weights in DNNs are also influenced by both top-down reading goals and lower-level stimulus features, with the shallow layers more strongly influenced by lower-level text features and the deep layers attending more to task-relevant words. Additionally, deep layers' attention to task-relevant words gradually emerges when pre-trained DNN models are fine-tuned to perform the reading comprehension task, which coincides with the improvement in task performance. These results demonstrate that DNNs can evolve human-like attention distribution through task optimization, which suggests that human attention during goal-directed reading comprehension is a consequence of task optimization.
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Cite as: arXiv:2107.05799 [cs.CL]
  (or arXiv:2107.05799v1 [cs.CL] for this version)

Submission history

From: Jiajie Zou [view email]
[v1] Tue, 13 Jul 2021 01:07:22 GMT (2748kb)

Link back to: arXiv, form interface, contact.