Machine translation (MT) is one of the main tasks in natural language processing whose objective is to translate texts automatically from one natural language to another. Nowadays, using deep neural networks for MT tasks has received great attention. These networks require lots of data to learn abstract representations of the input and store it in continuous vectors. This paper presents the first relatively large-scale Amharic-English parallel sentence dataset. Using these compiled data, we build bi-directional Amharic-English translation models by fine-tuning the existing Facebook M2M100 pre-trained model achieving a BLEU score of 37.79 in Amharic-English 32.74 in English-Amharic translation. Additionally, we explore the effects of Amharic homophone normalization on the machine translation task. The results show that the normalization of Amharic homophone characters increases the performance of Amharic-English machine translation in both directions.

Although a machine translation model trained with a large in-domain parallel corpus achieves remarkable results, it still works poorly when no in-domain data are available. This situation restricts the applicability of machine translation when the target domain's data are limited. However, there is great demand for high-quality domain-specific machine translation models for many domains. We propose a framework that efficiently and effectively collects parallel sentences in a target domain from the web with the help of crowdworkers. With the collected parallel data, we can quickly adapt a machine translation model to the target domain. Our experiments show that the proposed method can collect target-domain parallel data over a few days at a reasonable cost. We tested it with five domains, and the domain-adapted model improved the BLEU scores to +19.7 by an average of +7.8 points compared to a general-purpose translation model.

Active learning aims to deliver maximum benefit when resources are scarce. We use COMET-QE, a reference-free evaluation metric, to select sentences for low-resource neural machine translation. Using Swahili, Kinyarwanda and Spanish for our experiments, we show that COMET-QE significantly outperforms two variants of Round Trip Translation Likelihood (RTTL) and random sentence selection by up to 5 BLEU points for 20k sentences selected by Active Learning on a 30k baseline. This suggests that COMET-QE is a powerful tool for sentence selection in the very low-resource limit.

Adapting a neural text-to-speech (TTS) model to a target speaker typically involves fine-tuning most if not all of the parameters of a pretrained multi-speaker backbone model. However, serving hundreds of fine-tuned neural TTS models is expensive as each of them requires significant footprint and separate computational resources (e.g., accelerators, memory). To scale speaker adapted neural TTS voices to hundreds of speakers while preserving the naturalness and speaker similarity, this paper proposes a parameter-efficient few-shot speaker adaptation, where the backbone model is augmented with trainable lightweight modules called residual adapters. This architecture allows the backbone model to be shared across different target speakers. Experimental results show that the proposed approach can achieve competitive naturalness and speaker similarity compared to the full fine-tuning approaches, while requiring only $\sim$0.1% of the backbone model parameters for each speaker.

Segmentation for continuous Automatic Speech Recognition (ASR) has traditionally used silence timeouts or voice activity detectors (VADs), which are both limited to acoustic features. This segmentation is often overly aggressive, given that people naturally pause to think as they speak. Consequently, segmentation happens mid-sentence, hindering both punctuation and downstream tasks like machine translation for which high-quality segmentation is critical. Model-based segmentation methods that leverage acoustic features are powerful, but without an understanding of the language itself, these approaches are limited. We present a hybrid approach that leverages both acoustic and language information to improve segmentation. Furthermore, we show that including one word as a look-ahead boosts segmentation quality. On average, our models improve segmentation-F0.5 score by 9.8% over baseline. We show that this approach works for multiple languages. For the downstream task of machine translation, it improves the translation BLEU score by an average of 1.05 points.

The many-to-many multilingual neural machine translation can translate between language pairs unseen during training, i.e., zero-shot translation. Improving zero-shot translation requires the model to learn universal representations and cross-mapping relationships to transfer the knowledge learned on the supervised directions to the zero-shot directions. In this work, we propose the state mover's distance based on the optimal theory to model the difference of the representations output by the encoder. Then, we bridge the gap between the semantic-equivalent representations of different languages at the token level by minimizing the proposed distance to learn universal representations. Besides, we propose an agreement-based training scheme, which can help the model make consistent predictions based on the semantic-equivalent sentences to learn universal cross-mapping relationships for all translation directions. The experimental results on diverse multilingual datasets show that our method can improve consistently compared with the baseline system and other contrast methods. The analysis proves that our method can better align the semantic space and improve the prediction consistency.

Neurons can code for multiple variables simultaneously and neuroscientists are often interested in classifying neurons based on their receptive field properties. Statistical models provide powerful tools for determining the factors influencing neural spiking activity and classifying individual neurons. However, as neural recording technologies have advanced to produce simultaneous spiking data from massive populations, classical statistical methods often lack the computational efficiency required to handle such data. Machine learning (ML) approaches are known for enabling efficient large scale data analyses; however, they typically require massive training sets with balanced data, along with accurate labels to fit well. Additionally, model assessment and interpretation are often more challenging for ML than for classical statistical methods. To address these challenges, we develop an integrated framework, combining statistical modeling and machine learning approaches to identify the coding properties of neurons from large populations. In order to demonstrate this framework, we apply these methods to data from a population of neurons recorded from rat hippocampus to characterize the distribution of spatial receptive fields in this region.

In this work, we define a new style transfer task: perspective shift, which reframes a dialogue from informal first person to a formal third person rephrasing of the text. This task requires challenging coreference resolution, emotion attribution, and interpretation of informal text. We explore several baseline approaches and discuss further directions on this task when applied to short dialogues. As a sample application, we demonstrate that applying perspective shifting to a dialogue summarization dataset (SAMSum) substantially improves the zero-shot performance of extractive news summarization models on this data. Additionally, supervised extractive models perform better when trained on perspective shifted data than on the original dialogues. We release our code publicly.

At least 21 European languages are in danger of digital extinction due to a severe lack of technology support, concluded the META-NET 2012 reports prepared by a group of more than 230 experts from all over Europe. For the past decade, the introduction of neural technologies in automatic translation has precipitated a revolution in digital language services, allowing for ever faster and more accurate automatic speech recognition (ASR) and machine translation (MT) results. Yet, a stark imbalance persists in technology support between the five most spoken EU languages (English, French, German, Spanish and Italian), and the remaining 19 official ones. This digital inequality further increases when regional and minority languages are considered, leading to a dearth of online technological support, both in spoken (audio, video) and written (text) form. As digital services become an ever so integral part of our lives, such digital language inequalities could eventually threaten the digital survival of EU languages.

Developing Piping and Instrumentation Diagrams (P&IDs) is a crucial step during the development of chemical processes. Currently, this is a tedious, manual, and time-consuming task. We propose a novel, completely data-driven method for the prediction of control structures. Our methodology is inspired by end-to-end transformer-based human language translation models. We cast the control structure prediction as a translation task where Process Flow Diagrams (PFDs) are translated to P&IDs. To use established transformer-based language translation models, we represent the P&IDs and PFDs as strings using our recently proposed SFILES 2.0 notation. Model training is performed in a transfer learning approach. Firstly, we pre-train our model using generated P&IDs to learn the grammatical structure of the process diagrams. Thereafter, the model is fine-tuned leveraging transfer learning on real P&IDs. The model achieved a top-5 accuracy of 74.8% on 10,000 generated P&IDs and 89.2% on 100,000 generated P&IDs. These promising results show great potential for AI-assisted process engineering. The tests on a dataset of 312 real P&IDs indicate the need of a larger P&IDs dataset for industry applications.