Uncovering Noun-Noun Compound Relations by Gamification

Johan Bos, Malvina Nissim
Can relations described by English noun-noun compounds be adequately captured by prepositions? We attempt to answer this question in a data-driven way, using gamification to annotate a set of about a thousand noun-noun compound examples. Annotators could make a choice out of five prepositions generated with the help of paraphrases found in the Google n-gram corpus. We show that there is substantial agreement among the players of our linguistic annotation game, and that their answers differ in
more » ... out 50% of raw frequency counts of the Google n-gram corpus. Prepositions can be used to describe the majority of the implicit relations present in noun-noun compounds, but not all relations are captured by natural prepositions and some compounds are not easy to paraphrase with the use of a preposition.