Argumentative “microtexts” are short texts produced by students in response
to a trigger question that invites taking a stance and providing arguments in
support of that stance. Our original corpus of 115 texts of this kind, collected
in a classroom setting, was published in 2015, together with annotations of
the underlying argumentation structure. In the talk, I first describe our
experiments on automatically classifying those structures, which make use of
the minimum-spanning tree algorithm, and of Integer Linear Programming.
Then I turn to various extensions made to the corpus just recently: New
annotation layers that allow for computing correlations with properties of
argumentation structure, as well as an extension of the overall data set by 200
new texts that were obtained via crowdsourcing. I will summarize our
experiences with this approach to text production and explain the extra steps
needed to make the crowdsourced data compatible to the original corpus.