Attribution graph comparison

clinicaloff-targetstructuralnode size = attribution masscurve width/opacity = |attribution weight|red curve = negativehover or tap any node for details
Baseline (standard phrasing)“After that greasy fry-up my dyspepsia flared up, so at the counter I grabbed an”prob(“apple”) = 0.17embL0L5L10L15L20L25logit⋮“apple” · prob 0.17“orange” · prob 0.13“egg” · prob 0.09“ice” · prob 0.06“iced” · prob 0.04ClinicalOff-targetStructural<bos>After that greasy fry-up my dyspepsia flared up, so at the counter I grabbed an454773 weakest edges omittedRung 1 - fully formal clinical register“Following ingestion of that greasy fried meal, my dyspepsia became symptomatic, necessitating that I procure at the pharmacy counter an”embL0L5L10L15L20L25logit⋮“ant” · prob 0.32“over” · prob 0.23“anti” · prob 0.10“OTC” · prob 0.03“Ant” · prob 0.02ClinicalOff-targetStructural<bos>Following ingestion of that greasy fried meal, my dyspepsia became symptomatic, necessitating that I procure at the pharmacy counter an417854 weakest edges omittedRung 2 - mostly clinical, one everyday word“After that greasy fry-up, my dyspepsia became symptomatic, so at the counter I purchased an”embL0L5L10L15L20L25logit⋮“ant” · prob 0.11“anti” · prob 0.10“over” · prob 0.05“aspirin” · prob 0.04“orange” · prob 0.04ClinicalOff-targetStructural<bos>After that greasy fry-up, my dyspepsia became symptomatic, so at the counter I purchased an426108 weakest edges omittedDialect Δ vs. baseline (Rung 3 - even mix of clinical and everyday phrasing): -3% probability (0.17 → 0.14)Rung 3 - even mix of clinical and everyday phrasing“After that greasy fry-up my dyspepsia flared up, so at the counter I picked up an”prob(“apple”) = 0.14embL0L5L10L15L20L25logit⋮“apple” · prob 0.14“orange” · prob 0.10“egg” · prob 0.08“ice” · prob 0.05“iced” · prob 0.04ClinicalOff-targetStructural<bos>After that greasy fry-up my dyspepsia flared up, so at the counter I picked up an439930 weakest edges omittedDialect Δ vs. baseline (Rung 4 - mostly casual everyday phrasing): -3% probability (0.17 → 0.14)Rung 4 - mostly casual everyday phrasing“After that greasy fry-up my dyspepsia flared up again, so I grabbed an”prob(“apple”) = 0.14embL0L5L10L15L20L25logit⋮“apple” · prob 0.14“early” · prob 0.08“ice” · prob 0.06“aspirin” · prob 0.06“orange” · prob 0.05ClinicalOff-targetStructural<bos>After that greasy fry-up my dyspepsia flared up again, so I grabbed an419102 weakest edges omittedDialect Δ vs. baseline (Rung 5 - maximally casual spoken phrasing with slang): +3% probability (0.17 → 0.19)Rung 5 - maximally casual spoken phrasing with slang“After that greasy fry-up my dyspepsia was acting up somethin' fierce, so I just grabbed an”prob(“apple”) = 0.19embL0L5L10L15L20L25logit⋮“apple” · prob 0.19“old” · prob 0.11“empty” · prob 0.07“ice” · prob 0.07“orange” · prob 0.05ClinicalOff-targetStructural<bos>After that greasy fry-up my dyspepsia was acting up somethin' fierce, so I just grabbed an397520 weakest edges omitted