This text proposes a method for automatic analysis of predicates for discovery (PD) in Spanish. A PD is a predicative unit that projects an argument structure (AS) whose meaning alludes to ‘something that is found by someone -or something- somewhere’ (e.g., ‘encontrar’, ‘hallar’). This type of task is useful in fields such as medicine, since it offers the possibility of automatically identifying findings of interest (diseases, test results, etc.) in large text corpora. The present work is based on Lexicon Grammar (LG), which proposes a formalization from the nature of arguments (object classes) and transformational possibilities. The methodology is carried out as follows: (i) manual identification of PDs from a corpus of gynecology and obstetrics; (ii) elaboration of LG tables for each PD, where object classes are categorized and possible transformations are listed; and (iii) computational modeling. For the last stage, electronic dictionaries and computer-generated grammars were built in NooJ. The algorithm with automatically detected and generated ASs from PDs (325 grammatical sentences) was evaluated against an annotated corpus (1000 manually-annotated sentences, randomly extracted from a corpus of 5 million words). Results gave 98% accuracy, 88% coverage, and 92% F-measure.