Interoperability = f(community, division of labour)

This paper aims to motivate the hypothesis that practical interoperability can be seen as a function of whether and how stakeholder communities duplicate or divide work in a given area or market. We focus on the area of language processing which traditionally produces many diverse tools that are not immediately interoperable. However, there is also a strong desire to combine these tools into processing pipelines and to apply these to a wide range of different corpora. The space opened between generic, inherently “empty” interoperability frameworks that offer no NLP capabilities themselves and dedicated NLP tools gave rise to a new class of NLP-related projects that focus specifically on interoperability: component collections. This new class of projects drives interoperability in a very pragmatic way that could well be more successful than, e.g., past efforts towards standardised formats which ultimately saw little adoption or support by software tools.