Figure - PMC

Skip to main content

An official website of the United States government

Here's how you know

Here's how you know

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS
A lock ( ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

View full-text article in PMC

. 2023 Oct 30;26(12):2213–2225. doi: 10.1038/s41593-023-01468-4

Search in PMC
Search in PubMed
View in NLM Catalog
Add to search

© The Author(s). 2023

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

PMC Copyright notice

Fig. 4 — a, Sample speech sentence text, waveform and phonemic annotations. The segmentations of phonemic and syllabic contexts to the current timeframe (black arrow) are marked in different colors: phoneme(0), current phoneme (gray); phoneme(−1), previous phoneme (purple); phoneme(−2), second to the previous phoneme (blue); syllable(0), current syllable (excluding the current phoneme; green); syllable(−1), previous syllable (orange); syllable(−2), second to the previous syllable (red). b, Template attention-weight matrices for different contextual structures as shown in a. ‘Query’ indicates the target sequence. ‘Key’ indicates the source sequence. Colored blocks correspond to different contexts. c, Averaged AS (Pearson’s correlation coefficient between attention weights and templates) across all English sentences for each transformer layer corresponding to each type of attention template. r values in the top left of each panel indicate the correlation between the AS and the layer index (n = 12 different layers, permutation test). Black line indicates the averaged AS from the same DNN architecture with randomized weights (mean ± s.e.m., n = 499 independent sentences). d–g, Scatter plots of AS versus BPS across layers for the AN (d), IC (e), HG (f) and STG (g) areas. Each dot indicates a transformer layer, and each panel corresponds to one type of attention pattern. The r and P values correspond to the AS–BPS correlation across layers (Pearson’s correlation, permutation test, one-sided. Red fonts indicate significant positive correlations).

Source data