I'm trying to build a cognitive model of how people learn a event representation from a sentence describing the event, based on St. John & McClelland's Sentence Gestalt Model.

However, while that model generated its own sentences based on the event to be described, I'm hoping to find a corpus that supplies me with the descriptive sentences required.

Since the event description I end up trying to extract from these sentences is likely to be rather ad hoc, I'm not simply looking for a corpus tagged with semantic information like FrameNet or PropBank (although any tagging is always a plus).

Rather, I'm hoping to get a corpus with a very limited scope of topic (e.g. traffic/weather reports, sport/game commentary, a catalog of product descriptions) and then use a combination of heuristics and Amazon's Mechanical Turk to get whatever semantic information I end up needing to train my model.

2 Answers 2


Have you looked at this paper? I quote the abstract:

we describe the Baseball Announcers’ Language Linked with General Annotation of Meaningful Events (BALLGAME) project – a text corpus for research in computional semantics. We collected pitch-by-pitch event data for a sample of baseball games and used this data to build an annotated corpus composed of transcripts of radio broadcasts of these games. Our annotation links text from the broadcast to events in a formal representation of the semantics of the baseball game. We describe our corpus model, the annotation tool used to create the corpus, and conclude by discussing applications of this corpus in semantics research and natural language processing.

Besides this, many people have worked on creating and/or analysing RoboCup commentary data. I haven't looked for links for that. But I think they ought to be easy enough to find. Try this website where they have the corpus as an XML file. The text within <nl lang="en"> is what you are looking for, I guess.

  • The BALLGAME project's corpus looks like the exact thing I need! Thanks a bundle for the paper, but would you happen to know where I might get my hands on the actual corpus? The paper strangely fails to mention this...
    – zergylord
    Commented Sep 16, 2011 at 18:36
  • Ezra Keshret's homepage (ezrakeshet.com) seems to be malfunctioning... you could probably contact him to find out.
    – prash
    Commented Sep 16, 2011 at 19:03

If you want limited and descriptive then look fora narrative corpus. Your best bet would be to look for something like "Frog Story" - which is the wordless picture book "Frog Where Are You" by Mercer Mayer a fairly standard text in linguistic elicitation. This way you also have controlled elicitation because you know exactly the scope of the visual information available for each event description.

I'm assuming you want English, but this task has been done in many many languages over the years.

Here's a text file of a child and adult corpus from Tania Renner and Virginia Marchman at U Cal: http://www.docstoc.com/docs/53044454/ENGLISH-FROG-STORY-DATA

OLAC shows one from Lowry Hemphill: http://www.language-archives.org/item/oai:childes.psy.cmu.edu:Frogs-English-WolfHemp

The CHILDES project - a major child language corpus - also has a collection of frog stories (this opens as a PDF) http://childes.psy.cmu.edu/manuals/frog.pdf

Yuji Kawaguchi et al.'s "Corpus-based perspectives in linguistics" has an article and corpus on Frog Stories.

Of course, a bit more googling and investigating might turn up more - you'll find plenty of researchers have used Frog Story and even if their corpus isn't publicly available you could contact them to see if they'll share it!

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.