Many social medias provide machine translation to help the users understand any caption made in another languages. But it is something gives inaccurate translation which makes the meaning is not delivered properly. The study aims to identify types of semantic errors in lexis made by Instagram Machine Translation found in the National Geographic Indonesia’s Instagram captions. Observation method and note taking technique is used in to collect the data from National Geographic Indonesia Instagram Account from May until October. This study used a descriptive qualitative method in analyzing types of semantic errors in lexis and applying the theory types of semantic errors in lexis by James (1998). The result shows that the confusion of sense relations in choosing a wrong synonym is mostly found in data source. This phenomenon occurred because Instagram Machine Translation literally translated the source language to the target language which have several synonyms and different meaning in each use without considering the overall context of the caption.

Keywords: semantic errors in lexis, Instagram, machine translation