We describe an experimental approach to study the relation between emojis and online language use based on tweets from a public social media platform. The novel contribution of our work derives from the application of an entropy measure to compare lexical diversity (vocabulary richness) in messages with and without emojis. We find that lexical diversity is reliably attenuated when emojis are added to text. The pattern holds both when influence of tweet length or number of emojis is removed. That emojis serve to alter the lexical diversity of text in an online public context could demonstrate an interdependency between emojis and accompanying text in the service of communication. Our findings are consistent with the claim that the drop in lexical diversity may reflect a compensatory relation between emojis and words in communication. This outcome sets boundary conditions on the extent to which emoji patterning in isolation from text can qualify as a language.