Datasets are foundational to many breakthroughs in modern artificial intelligence. Many recent achievements in the space of natural language processing (NLP) can be attributed to the fine-tuning of pre-trained models on a diverse set of tasks that enables a large language model (LLM) to respond to instructions. Instruction fine-tuning (IFT) requires specifically constructed and annotated datasets. However, existing datasets are almost all in the English language. In this work, our primary goal is to bridge the language gap by building a human-curated instruction-following dataset spanning 65 languages. We worked with fluent speakers of languages from around the world to collect natural instances of instructions and completions. Furthermore, we create the most extensive multilingual collection to date, comprising 513 million instances through templating and augmenting existing datasets across 114 languages. In total, we contribute three key resources: we develop and open-source the Aya Dataset, the Aya Collection, and the Aya Evaluation Suite. The Aya initiative also serves as a valuable case study in participatory research, involving collaborators from 119 countries. We see this as an important framework for future research collaborations that aim to bridge gaps in resources.

AI alignment in the shape of Reinforcement Learning from Human Feedback (RLHF) is increasingly treated as a crucial ingredient for high performance large language models. Proximal Policy Optimization (PPO) has been installed by the seminal literature as the standard method for the RL part of RLHF. However, it involves both high computational cost and sensitive hyperparameter tuning. We posit that most of the motivational principles that led to the development of PPO are less of a practical concern in RLHF and advocate for a less computationally expensive method that preserves and even increases performance. We revisit how alignment from human preferences is formulated in the context of RL. Keeping simplicity as a guiding principle, we show that many components of PPO are unnecessary in an RLHF context and that far simpler REINFORCE-style optimization variants outperform both PPO and newly proposed “RL-free” methods such as DPO and RAFT. Our work suggests that careful adaptation to LLMs alignment characteristics allows benefiting from online RL optimization at low cost.

Recent breakthroughs in large language models (LLMs) have centered around a handful of data-rich languages. What does it take to broaden access to breakthroughs beyond first-class citizen languages? Our work introduces Aya, a massively multilingual generative language model that follows instructions in 101 languages of which over 50% are considered as lower-resourced. Aya outperforms mT0 and BLOOMZ on the majority of tasks while covering double the number of languages. We introduce extensive new evaluation suites that broaden the state-of-art for multilingual eval across 99 languages —— including discriminative and generative tasks, human evaluation, and simulated win rates that cover both held-out tasks and in-distribution performance. Furthermore, we conduct detailed investigations on the optimal finetuning mixture composition, data pruning, as well as the toxicity, bias, and safety of our models.

In Natural Language Processing (NLP), the Elo rating system, well-established for ranking dynamic competitors in games like chess, has seen increasing adoption for evaluating Large Language Models (LLMs) through “A vs B” paired comparisons. However, while popular, the system’s suitability for assessing entities with constant skill levels, such as LLMs, remains relatively unexplored. Our study investigates the sensitivity and reproducibility of Elo scores for LLMs, integrating both synthetic and human feedback. We show that Elo ratings for LLMs stabilize with 100 or more comparison permutations. A lower K-factor is preferable for closely matched models, whereas a higher K-factor better distinguishes models with clear performance differences. We also report that transitivity (A B and B C implies A C) does not consistently hold, particularly when models demonstrate similar performance. Our empirical findings provide guidelines for more reliable LLM evaluation.

To provide AI researchers with modern tools for dealing with the explosive growth of the research literature in their field, we introduce a new platform, AI Research Navigator, that combines classical keyword search with neural retrieval to discover and organize relevant literature. The system provides search at multiple levels of textual granularity, from sentences to aggregations across documents, both in natural language and through navigation in a domain specific Knowledge Graph. We give an overview of the overall architecture of the system and of the components for document analysis, question answering, search, analytics, expert search, and recommendations.

Recent works have shown that Neural Machine Translation (NMT) models achieve impressive performance, however, questions about understanding the behavior of these models remain unanswered. We investigate the unexpected volatility of NMT models where the input is semantically and syntactically correct. We discover that with trivial modifications of source sentences, we can identify cases where unexpected changes happen in the translation and in the worst case lead to mistranslations. This volatile behavior of translating extremely similar sentences in surprisingly different ways highlights the underlying generalization problem of current NMT models. We find that both RNN and Transformer models display volatile behavior in 26% and 19% of sentence variations, respectively.

Neural Machine Translation has achieved state-of-the-art performance for several language pairs using a combination of parallel and synthetic data. Synthetic data is often generated by back-translating sentences randomly sampled from monolingual data using a reverse translation model. While back-translation has been shown to be very effective in many cases, it is not entirely clear why. In this work, we explore different aspects of back-translation, and show that words with high prediction loss during training benefit most from the addition of synthetic data. We introduce several variations of sampling strategies targeting difficult-to-predict words using prediction losses and frequencies of words. In addition, we also target the contexts of difficult words and sample sentences that are similar in context. Experimental results for the WMT news translation task show that our method improves translation quality by up to 1.7 and 1.2 Bleu points over back-translation using random sampling for German-English and English-German, respectively.

Distributed word representations are widely used for modeling words in NLP tasks. Most of the existing models generate one representation per word and do not consider different meanings of a word. We present two approaches to learn multiple topic-sensitive representations per word by using Hierarchical Dirichlet Process. We observe that by modeling topics and integrating topic distributions for each document we obtain representations that are able to distinguish between different meanings of a given word. Our models yield statistically significant improvements for the lexical substitution task indicating that commonly used single word representations, even when combined with contextual information, are insufficient for this task.

The quality of a Neural Machine Translation system depends substantially on the availability of sizable parallel corpora. For low-resource language pairs this is not the case, resulting in poor translation quality. Inspired by work in computer vision, we propose a novel data augmentation approach that targets low-frequency words by generating new sentence pairs containing rare words in new, synthetically created contexts. Experimental results on simulated low-resource settings show that our method improves translation quality by up to 2.9 BLEU points over the baseline and up to 3.2 BLEU over back-translation.