Language tagging for NLP

The challenge

Current language services go beyond just translation. In Qabiria we are used to tackling projects that straddle different disciplines.

In 2020, we were contacted by a leading company in natural language processing (NLP) that needed to label a corpus of texts, taken from conversations on a social network, that had various languages mixed together. Labeling, or tagging, involved identifying the languages, parts of speech, and functions performed by each element.

The project was particularly complex for several reasons: first, the client’s very tight timelines. Under pressure from the final client, a large international group, the project of about 500,000 words had to be completed in just 3 weeks.

Secondly, we were required to use brat rapid annotation tool, a specific structured annotation tool, which we had never used until then. In this case, it was sufficient to connect via a browser to the brat instance on the client’s servers to access the assigned tasks and perform them in collaborative mode.

Finally, the presence of other languages, in addition to Italian, entailed an additional complication, as one had to distinguish the use of simple Anglicisms, which are very frequent in Italian, from the parts properly in other languages.

The solution

To cope with the client’s urgency, we prepared in a few days a team of 8 linguists who worked in a coordinated way. After training them on the activity to be done and the system to be used, we assigned each person his or her part and set up a system to keep track of the hours spent.

Qabiria has also assigned a dedicated project manager to this position. This person maintained contact with the client, monitored adherence to timelines, responded to linguists' concerns, and oversaw the entire project.

The result

After completing a first part of the project on time and in the required manner, the client expressed full satisfaction and sent other similar tagging work in the following months.

To tackle these new jobs, we have further optimized processes and significantly reduced delivery times, while still maintaining the same level of quality.