It is an indication of their primary language but it might be misleading. For example, a user might select French as their mother tongue but compose their Tweets in English. DataSift offers an augmentation called Language which determines the language based on a statistical analysis of groups of characters.
The drop-down list shows the full name of the language (English, French, German, and so on) but the twitter.user.lang target contains the code for the language.
Take a look at the list of languages that Twitter supports.
Filter for Tweets from authors who specify French as their preferred language and mention Paris:
interaction.author.language == "fr" and interaction.content contains "Paris"
Filter for Tweets from authors who specify French as their preferred language but Tweet in English:
interaction.author.language == "fr" and language.tag == "en"
Target service: The Common Target: Interaction
Target object: Interaction: Author
Always exists: No