Submission

Participants can submit up to ten different models:

For each model, each team can submit a maximum of three different versions (runs).

Format: each run will be an output file, and the format for each output file will have to be as follows:

• <id> tab <prediction>

<id> is the instance id as provided in the test files

<prediction> is one of "F" or "M"

For example, for the following instances in a Twitter test file:

you will have to produce lines as such, assuming your system has predicted male, female, male, respectively:

Optimally, you will preserve the order as provided in the test files.

File names: each team has to make sure the submitted files obey the following standardised format:

• <team-name>_<setting>_<target-genre>_<run-number>

<team-name> is your team name, for instance "UniABC"

<setting> is one of either "IN" (for in-genre) or "OUT" (for cross-genre)

<target-genre> is one of "TW" (Twitter), "YT" (YouTube), "DI" (Diary), "JO" (Journalism), and "CH" (Children)

<run-number> is one of 1,2,3

For example, team "UniABC" can submit three runs for in-genre gender detection in Diaries and name the files:

• UniABC_IN_DI_1
• UniABC_IN_DI_2
• UniABC_IN_DI_3

For cross-genre, teams should specify the genre they are testing on, thus:

• UniABC_OUT_TW_1 is a run where genre detection is done on the Twitter test data with models trained on anything but Twitter
• UniABC_OUT_YT_3 is a run where genre detection is done on the YouTube test data with models trained on anything but YouTube