Contradictory, My Dear Watson is a Getting Started competition focused on Natural Language Inferencing (NLI) using multilingual text. The challenge asks participants to determine the relationship between pairs of sentences: whether one entails the other, contradicts it, or if they are unrelated.
The task involves creating an NLI model that assigns labels to pairs of premises and hypotheses: - 0: Entailment - 1: Neutral - 2: Contradiction
A unique aspect of this competition is that the training and test sets include text in fifteen different languages. Participants are encouraged to use modern approaches such as embeddings and transformers like BERT. To facilitate this, Kaggle provides a starter notebook designed to run on Tensor Processing Units (TPUs), which are hardware accelerators specialized for deep learning tasks. All users have access to TPU Quota at no cost.
This competition is designed for individuals with little to no machine learning background, serving as a non-competitive way to get familiar with Kaggle's platform, learn basic concepts, and engage with the community. There are no cash prizes, medals, or points associated with leaderboard rankings; the focus is purely on learning and experimentation. The dataset contains text that may be considered profane, vulgar, or offensive.
Submissions must be generated as an output from a Kaggle Notebook. Participants should submit a CSV file containing exactly 5195 entries plus a header row, with two columns: id and prediction. The score is calculated based on accuracy—the percentage of relationships correctly predicted.