# Multilingual Models for Compositional Distributed Semantics

Source: [Multilingual Models for Compositional Distributed Semantics](<https://www.sota2.com/research/paper/68fa61501e209131ce9f3c05>)

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

Data updated: 2026-01-01T12:01:16.028Z

Authors: Karl Moritz Hermann, Phil Blunsom

Published: 2014-04-17T20:18:03.000Z

## Abstract

We present a novel technique for learning semantic representations, which extends the distributional hypothesis to multilingual data and joint-space embeddings. Our models leverage parallel data and learn to strongly align the embeddings of semantically equivalent sentences, while maintaining sufficient distance between those of dissimilar sentences. The models do not rely on word alignments or any syntactic information and are successfully applied to a number of diverse languages. We extend our approach to learn semantic representations at the document level, too. We evaluate these models on two cross-lingual document classification tasks, outperforming the prior state of the art. Through qualitative analysis and the study of pivoting effects we demonstrate that our representations are semantically plausible and can capture semantic relationships across languages without parallel data.

## 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.

- [Document Classification on TED corpus L2 -\> en (test)](<https://www.sota2.com/research/sota/document-classification-on-ted-corpus-l2-en-test.md>): Document Classification; TED corpus L2 -\> en (test); Arabic: 45.2.

- [Document Classification on TED corpus en -\> L2 (test)](<https://www.sota2.com/research/sota/document-classification-on-ted-corpus-en-l2-test.md>): Document Classification; TED corpus en -\> L2 (test); Arabic: 41.

- [Document Classification on TED corpus](<https://www.sota2.com/research/sota/document-classification-on-ted-corpus.md>): Document Classification; TED corpus; English: 47.5.

- [Crosslingual Document Classification on RCV1 RCV2 DE -\> EN (1,000 documents subset, test)](<https://www.sota2.com/research/sota/crosslingual-document-classification-on-rcv1-rcv2-de-en-1-000-documents-subset-test.md>): Crosslingual Document Classification; RCV1 RCV2 DE -\> EN 1,000 documents per language (test); Accuracy: 79.2.

- [Crosslingual Document Classification on RCV1 RCV2 EN -\> DE (test)](<https://www.sota2.com/research/sota/crosslingual-document-classification-on-rcv1-rcv2-en-de-test.md>): Crosslingual Document Classification; RCV1 RCV2 EN -\> DE 1,000 documents per language (test); Accuracy: 88.1.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/1404.4641>)

[PDF](<https://arxiv.org/pdf/1404.4641.pdf>)

[Implementation](<http://www.karlmoritz.com/>)