Table of Contents

2.2 Layers of information

WhatsApp messages are built up in a hierarchy: a chat contains messages that contain tokens that contain characters. A corpus of WhatsApp chats should allow for all these layers to be queried. Additionally, there is meta-data about the chats (e.g. number of messages) and about the messages (e.g. the timestamp when they were written) and about the informant (e.g. his/her age) and about the tokens (e.g. part of speech). This makes our corpus a rather challenging and complex resource.

These layers can nicely be seen when browsing results from a query: Figure 1: Representation of layers when browsing results

Chats

In this example, you find the chat back as an ID (chat138) at the top in pink in figure 1. If you want to see the whole chat, you see two options at the very bottom: chat (in context) (faster) or chat (complete) (can be slow). When you click on the little i in the top bar, you can also see meta data about the chat, such as the number of speakers, languages, total messages etc.

Messages

In the chat in figure 1, you see three selected messages in blue:

As you can see, these messages have meta data assigned to them as well, e.g. the message ID and the speaker (these pieces of information are always available) as well as information provided by the informant such as age, mothertongue etc.

Tokens

The individual tokens are marked in green in figure 1 and they are aligned to the message to which they belong.

Tokens, too, (can) have annotations that are assigned to them. In figure 1 you have the following meta data:

The red token di, by the way, is the one that we queried for to create this screen shot.

Labels

On all three layers, i.e. for chats, messages and tokens, as well as for all the meta data, you see the labels, e.g. msg_id, gloss, home_postcode etc. These labels are used for queries.

Examples:

To see the query-labels for the chat as well as all the labels available in a specific sub-corpus, check the information for the sub-corpus.