Actors Challenge is a new game on Lingo Boingo. Players record themselves "auditioning" for a part or "casting" others. The object of the game is to convey different meanings with intonation alone. To play, click here.
Below is a Q&A interview with Roberto Zamparelli, one of the developers of Actors Challenge.
Q: What gave you the idea for Actors Challenge?
A: At the time I was working on a project, Vinko, which used the web to collect oral data from regional dialects. I had recently reread Roman Jakobson's essay "Linguistics and Poetics", where he tells how actors trained in the Stanislavsky acting method would go on stage after giving the audience a list of 40 different situations. They then proceeded to utter the words "Sevodnija vecerom", in 40 different ways, and the audience was supposedly able to match each utterance to the corresponding situation in the list. It occurred to me that Vinko could be the blueprint for a system where participants would alternate between the roles of actor and audience, that audience recognition was a natural way to quantify acting success, and that this setup could be used to collect utterance-meaning pairs.
Q: What was the game development and research process like?
A: The game started as an internship project done by a master's student, which we continued and collaborated on after she graduated. We then hired a person to create the graphic interface. After that, writing the data was the fun part; the Italian and English contents were developed in parallel, and the translations were done by students passionate about the project.
Q: What kinds of research do you hope this game will contribute to?
A: Anything where prosody meets semantics, in the broadest sense. My PhD student is using it to study the acoustic correlates of emotions in speech, but it can be used to study the prosodic means to disambiguate syntactic or lexico-syntactic ambiguities, or topic/focus structures, which are quite complex in Italian. One advantage of the system is that it produces minimal pairs (same text, different prosody), which are nearly impossible with naturally collected data. Another is that you have the recognition data: even before opening the sound files you can do studies on which semantic aspects are best recognized by which type of user, with variables like language, region of provenance, age gap and gender. The cons to this method of data collection are lower sound fidelity and the fact that any scientific question has to be cast in the same mold: 1 target, multiple context.
Q: When enough people participate in this game, what do you hope to do with the linguistic data?
A: We will release them as anonymized open data on the platform Zenodo. We make participants aware at sign-up time that their voices might be recognized. Fortunately, tracing a voice to its owner is much harder than tracing a photo. If enough data is produced it could be used to improve computer speech recognition and production. Text-to-speech has become quite good lately, but not necessarily at the type of complex emotional scenarios that Actors Challenge can easily produce.