Speech-to-Text Translation (S2TT) has typically been addressed with cascade systems, where speech recognition systems generate a transcription that is subsequently passed to a translation model. While there has been a growing interest in developing direct speech translation systems to avoid propagating errors and losing non-verbal content, prior work in direct S2TT has struggled to conclusively establish the advantages of integrating the acoustic signal directly into the translation process. This work proposes using contrastive evaluation to quantitatively measure the ability of direct S2TT systems to disambiguate utterances where prosody plays a crucial role. Specifically, we evaluated Korean-English translation systems on a test set containing wh-phrases, for which prosodic features are necessary to produce translations with the correct intent, whether it’s a statement, a yes/no question, a wh-question, and more. Our results clearly demonstrate the value of direct translation systems over cascade translation models, with a notable 12.9% improvement in overall accuracy in ambiguous cases, along with up to a 15.6% increase in F1 scores for one of the major intent categories. To the best of our knowledge, this work stands as the first to provide quantitative evidence that direct S2TT models can effectively leverage prosody. The code for our evaluation is openly accessible and freely available for review and utilisation.

Large pretrained models enable transfer learning to low-resource domains for language generation tasks. However, previous end-to-end approaches do not account for the fact that some generation sub-tasks, specifically aggregation and lexicalisation, can benefit from transfer learning in different extents. To exploit these varying potentials for transfer learning, we propose a new hierarchical approach for few-shot and zero-shot generation. Our approach consists of a three-moduled jointly trained architecture: the first module independently lexicalises the distinct units of information in the input as sentence sub-units (e.g. phrases), the second module recurrently aggregates these sub-units to generate a unified intermediate output, while the third module subsequently post-edits it to generate a coherent and fluent final text. We perform extensive empirical analysis and ablation studies on few-shot and zero-shot settings across 4 datasets. Automatic and human evaluation shows that the proposed hierarchical approach is consistently capable of achieving state-of-the-art results when compared to previous work.

Large-scale document retrieval systems often utilize two styles of neural network models which live at two different ends of the joint computation vs. accuracy spectrum. The first style is dual encoder (or two-tower) models, where the query and document representations are computed completely independently and combined with a simple dot product operation. The second style is cross-attention models, where the query and document features are concatenated in the input layer and all computation is based on the joint query-document representation. Dual encoder models are typically used for retrieval and deep re-ranking, while cross-attention models are typically used for shallow re-ranking. In this paper, we present a lightweight architecture that explores this joint cost vs. accuracy trade-off based on multi-vector attention (MVA). We thoroughly evaluate our method on the MS-MARCO passage retrieval dataset and show how to efficiently trade off retrieval accuracy with joint computation and offline document storage cost. We show that a highly compressed document representation and inexpensive joint computation can be achieved through a combination of learned pooling tokens and aggressive downprojection. Our code and model checkpoints are open-source and available on GitHub.

Deep-learning models for language generation tasks tend to produce repetitive output. Various methods have been proposed to encourage lexical diversity during decoding, but this often comes at a cost to the perceived fluency and adequacy of the output. In this work, we propose to ameliorate this cost by using an Imitation Learning approach to explore the level of diversity that a language generation model can reliably produce. Specifically, we augment the decoding process with a meta-classifier trained to distinguish which words at any given timestep will lead to high-quality output. We focus our experiments on concept-to-text generation where models are sensitive to the inclusion of irrelevant words due to the strict relation between input and output. Our analysis shows that previous methods for diversity underperform in this setting, while human evaluation suggests that our proposed method achieves a high level of diversity with minimal effect on the output’s fluency and adequacy.

Concept-to-text Natural Language Generation is the task of expressing an input meaning representation in natural language. Previous approaches in this task have been able to generalise to rare or unseen instances by relying on a delexicalisation of the input. However, this often requires that the input appears verbatim in the output text. This poses challenges in multilingual settings, where the task expands to generate the output text in multiple languages given the same input. In this paper, we explore the application of multilingual models in concept-to-text and propose Language Agnostic Delexicalisation, a novel delexicalisation method that uses multilingual pretrained embeddings, and employs a character-level post-editing model to inflect words in their correct form during relexicalisation. Our experiments across five datasets and five languages show that multilingual models outperform monolingual models in concept-to-text and that our framework outperforms previous approaches, especially in low resource conditions.

This paper presents our submission to the WebNLG Challenge 2020 for the English and Russian RDF-to-text generation tasks. Our first of three submissions is based on Language Agnostic Delexicalisation, a novel delexicalisation method that match values in the input to their occurrences in the corresponding text through comparison of pretrained multilingual embeddings, and employs a character-level post-editing model to inflect words in their correct form during relexicalisation. Our second submission forfeits delexicalisation and uses SentencePiece subwords as basic units. Our third submission combines the previous two by alternating between the output of the delexicalisation-based system when the input contains unseen entities and/or properties and the output of the SentencePiece-based system when the input is seen during training.