... state-of-the-art abbreviationrecognizers as baselines: Schwartz and Hearst’smethod (SH) (2003), SaRAD (Adar, 2004), AL-ICE (Ao and Takagi, 2005), Chang and Sch¨utze’smethod (CS) (Chang and ... var-ious methods by which to extract abbreviationdefinitions that appear in actual texts (Taghva and Gilbreth, 1999; Park and Byrd, 2001; Wren and Garner, 2002; Schwartz and Hearst, 2003;Adar, ... forms asthe training/evaluation data.The evaluation metrics used in the abbreviationgeneration are exact-match accuracy (hereinafteraccuracy), including top-1 accuracy, top-2 accu-racy, and...