← Back to the problem

Kazakhstan FAIO 2025 · Statement

Problem 4: Who Speaks What?

Formatted by IOAI Training from the original (BeksultanSagyndyk/faio-tasks, licence Apache-2.0). The original is authoritative.

Story

Two interns, Adilkhan and Darkhan, have just started working as data scientists in the NLP team at a large telecom company. Their team lead assigned their very first task: improve the company’s customer service chatbot.

Since the telecom has millions of customers speaking Kazakh, Russian, and English, a crucial requirement is to detect the customer’s language correctly from the very first query. Without accurate language identification, the chatbot might respond in the wrong language, frustrating the user.

To solve this, Adilkhan and Darkhan compiled a dataset of real conversational messages collected from the company’s support system. They prepared two files: train.csv and test.csv, each containing short conversational texts labeled with one of three languages: "ru", "kaz", or "eng".

Now it is your mission to build a model that can accurately predict the language of any incoming text.
Your help is vital—if Adilkhan and Darkhan succeed, they will pass their internship and become full-time NLP engineers!

Formal Problem Statement

You are given two CSV files:

  • train.csv: Contains training data in the form of text samples with their corresponding labels (ru, kaz, eng).
  • test.csv: Contains only text samples, without labels. Your model must predict the correct label for each test example.

Dataset:

Input

  • train.csv with columns: text, label.
  • test.csv with column: text.
  • Labels are one of: ru, kaz, eng.

Output

Participants must submit only a CSV file named solution.csv, with the following format:

id,label
0,eng
1,kaz
2,ru
...