# Differentiable Reasoning on Large Knowledge Bases and Natural Language

Source: [Differentiable Reasoning on Large Knowledge Bases and Natural Language](<https://www.sota2.com/research/paper/68fa66d01e209131cea1285a>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-04T03:12:29.150Z

Authors: Pasquale Minervini, Matko Bo\\v{s}njak, Tim Rockt\\"aschel, Sebastian Riedel, Edward Grefenstette

Published: 2019-12-17T23:01:54.000Z

## Abstract

Reasoning with knowledge expressed in natural language and Knowledge Bases (KBs) is a major challenge for Artificial Intelligence, with applications in machine reading, dialogue, and question answering. General neural architectures that jointly learn representations and transformations of text are very data-inefficient, and it is hard to analyse their reasoning process. These issues are addressed by end-to-end differentiable reasoning systems such as Neural Theorem Provers (NTPs), although they can only be used with small-scale symbolic KBs. In this paper we first propose Greedy NTPs (GNTPs), an extension to NTPs addressing their complexity and scalability limitations, thus making them applicable to real-world datasets. This result is achieved by dynamically constructing the computation graph of NTPs and including only the most promising proof paths during inference, thus obtaining orders of magnitude more efficient models. Then, we propose a novel approach for jointly reasoning over KBs and textual mentions, by embedding logic facts and natural language sentences in a shared embedding space. We show that GNTPs perform on par with NTPs at a fraction of their cost while achieving competitive link prediction results on large datasets, providing explanations for predictions, and inducing interpretable models. Source code, datasets, and supplementary material are available online at https://github.com/uclnlp/gntp.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Link Prediction on Nations](<https://www.sota2.com/research/sota/link-prediction-on-nations.md>): Link Prediction; Nations; MRR: 0.658.

- [Link Prediction on FB122 (test-II)](<https://www.sota2.com/research/sota/link-prediction-on-fb122-test-ii.md>): Link Prediction; FB122 (test-II); Hits@3: 98.2.

- [Link Prediction on FB122 (test)](<https://www.sota2.com/research/sota/link-prediction-on-fb122-test.md>): Link Prediction; FB122 (test); Hits@3: 69.2.

- [Link Prediction on Countries S1](<https://www.sota2.com/research/sota/link-prediction-on-countries-s1.md>): Link Prediction; Countries S1; AUC-PR: 1.

- [Link Prediction on Countries S2](<https://www.sota2.com/research/sota/link-prediction-on-countries-s2.md>): Link Prediction; Countries S2; AUC-PR: 93.48.

- [Link Prediction on Countries S3](<https://www.sota2.com/research/sota/link-prediction-on-countries-s3.md>): Link Prediction; Countries S3; AUC-PR: 0.9127.

- [Link Prediction on Kinship](<https://www.sota2.com/research/sota/link-prediction-on-kinship.md>): Link Prediction; Kinship; MRR: 0.759.

- [Link Prediction on FB122 (test-I)](<https://www.sota2.com/research/sota/link-prediction-on-fb122-test-i.md>): Link Prediction; FB122 (test-I); Hits@3: 33.7.

- [Link Prediction on UMLS](<https://www.sota2.com/research/sota/link-prediction-on-umls.md>): Link Prediction; UMLS; Hits@10: 98.6.

- [Link Prediction on WN18RR (test)](<https://www.sota2.com/research/sota/link-prediction-on-wn18rr-test.md>): Link Prediction; WN18RR (test); Hits@10: 48.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/1912.10824>)

[PDF](<https://arxiv.org/pdf/1912.10824.pdf>)

[Implementation](<https://github.com/uclnlp/gntp>)