Predicting Typological Features in WALS using Language Embeddings and Conditional Probabilities: \'UFAL Submission to the SIGTYP 2020 Shared Task

TL;DR

This paper introduces a system for predicting linguistic typological features using a combination of correlation-based methods and neural language embeddings, achieving top accuracy in the SIGTYP 2020 shared task.

Contribution

It presents a novel hybrid approach combining correlation analysis and neural embeddings for typological feature prediction, ranking first in the shared task.

Findings

Achieved 70.7% accuracy on test data.

Outperformed other systems in the shared task.

Demonstrated effectiveness of combining correlation and neural methods.

Abstract

We present our submission to the SIGTYP 2020 Shared Task on the prediction of typological features. We submit a constrained system, predicting typological features only based on the WALS database. We investigate two approaches. The simpler of the two is a system based on estimating correlation of feature values within languages by computing conditional probabilities and mutual information. The second approach is to train a neural predictor operating on precomputed language embeddings based on WALS features. Our submitted system combines the two approaches based on their self-estimated confidence scores. We reach the accuracy of 70.7% on the test data and rank first in the shared task.