OBJECTIVES: Literature analysis could benefit from machine-learning (ML) methods that parse medical text to extract reported results. Therefore, we benchmark one ML approach against an incoming medical school student in the same task, as a comparative study. METHODS: We ran an ML algorithm, using Named Entity Recognizers, against a corpus of PubMed abstracts from Randomized Controlled Trials. We then sampled 102 resulting sentences, limiting to those with at least two interventions compared to one another. The ML parsed the disease, end-points, interventions studied, and the reported numerator and denominator (e.g., the sentence, “The total response rate was 75.0% (6/8) in non-transplantation group and 37.5% (3/8) in transplantation group, respectively.” yields two results. The first has 6 as a numerator, 8 as a denominator, “total response rate” as the end-point, and “non-transplantation” as the intervention. The second result has 6, 8, “total response rate”, and “transplantation” for its attributes.) We then trained the student on the same task and gave him the 102 sentences to process. RESULTS: The student and the ML algorithm yielded the same end-points for 97/102 sentences (95%). We then compared triples of (numerator, denominator, intervention) across the sentences. The ML and student match on 87/102 (85%) of sentences (e.g., both pairs matched for each sentence 85% of the time between the ML and student). Accounting for errors associated with only the numerator and denominator (e.g., the interventions match), this number improves to 94/102 sentences (92%). These errors are mostly associated with incorrectly assigning one intervention’s quantities to the other. CONCLUSIONS: Although preliminary, we demonstrate that ML can be almost as effective as an incoming medical student in the task of turning written text into structured results.