Attribution graph comparison clinical off-target structural node size = attribution mass curve width/opacity = |attribution weight|red curve = negative hover or tap any node for details
Clinical wording “<bos>Grandma's been constipated for a week, so before dinner she took a” emb L0 L5 L10 L15 L20 L25 logit ⋮ “lax” · prob 0.27 “dose” · prob 0.07 “couple” · prob 0.07 “stool” · prob 0.04 “big” · prob 0.04 Clinical Off-target Structural <bos> Grandma's been constipated for a week, so before dinner she took a 427397 weakest edges omitted Patient wording “<bos>Grandma's been all bunged up for a week, so before dinner she took a” emb L0 L5 L10 L15 L20 L25 logit ⋮ “nap” · prob 0.28 “shower” · prob 0.09 “bath” · prob 0.06 “little” · prob 0.03 “walk” · prob 0.03 Clinical Off-target Structural <bos> Grandma's been all bunged up for a week, so before dinner she took a 415535 weakest edges omitted Translated wording “<bos>Grandmother has been constipated for a week, so before dinner she took a” emb L0 L5 L10 L15 L20 L25 logit ⋮ “lax” · prob 0.47 “bowl” · prob 0.06 “stool” · prob 0.04 “spoonful” · prob 0.03 “few” · prob 0.03 Clinical Off-target Structural <bos> Grandmother has been constipated for a week, so before dinner she took a 477936 weakest edges omitted