Figure 1:
![An example of the masked LM task, where [MASK] = chase, and window is an attractor (red). We suppress phrase-structure annotations and corruptions on the context tokens for clarity.](https://mitp.silverchair-cdn.com/mitp/content_public/journal/tacl/8/10.1162_tacl_a_00345/5/00345f01.png?Expires=1716925094&Signature=FpLF7u7BcW4e7aD~0GC0kTYy87FpevSRzBKj2F6~H2HgkouKF5Ca7FMP~NZOdNyvqB46xphMGCa5ALiwORMUm5hgrmH4KNjyxLAieIV76crXhm8vQSHfkurxvs5uK5BeQtU6SBHdxlAQ-h6eArZD-ZgkuYY8GZ6dBt5-tBfYIL-Z7MS-oAqZ12jPPQVEyUERZB46b1fij1XRvfJN~jCzZTaeSoXym9P8ohWnxLrqLTkbluQ~KhCTJx2HkCkDtCORCfqUvbBa5~2U7M12077QNu00nUI0cHCDO6yIj3dhpGzMDWJPUM-wTOgpxON3Vkvx8KjxpZgIjANuPDVVKZXncQ__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
An example of the masked LM task, where [MASK] = chase, and window is an attractor (red). We suppress phrase-structure annotations and corruptions on the context tokens for clarity.
An example of the masked LM task, where [MASK] = chase, and window is an attractor (red). We suppress phrase-structure annotations and corruptions on the context tokens for clarity.
