Successful and effective communication between humans and AI relies on a shared experience of the world. By training solely on written text, current language models (LMs) miss the grounded experience of humans in the real-world -- their failure to relate language to the physical world causes knowledge to be misrepresented and obvious mistakes in their reasoning. We present Mind's Eye, a paradigm to ground language model reasoning in the physical world. Given a physical reasoning question, we use a computational physics engine (DeepMind's MuJoCo) to simulate the possible outcomes, and then use the simulation results as part of the input, which enables language models to perform reasoning. Experiments on 39 tasks in a physics alignment benchmark demonstrate that Mind's Eye can improve reasoning ability by a large margin (27.9% zero-shot, and 46.0% few-shot absolute accuracy improvement on average). Smaller language models armed with Mind's Eye can obtain similar performance to models that are 100x larger. Finally, we confirm the robustness of Mind's Eye through ablation studies.
We evaluate the reasoning abilities of large language models in multilingual settings. We introduce the Multilingual Grade School Math (MGSM) benchmark, by manually translating 250 grade-school math problems from the GSM8K dataset (Cobbe et al., 2021) into ten typologically diverse languages. We find that the ability to solve MGSM problems via chain-of-thought prompting emerges with increasing model scale, and that models have strikingly strong multilingual reasoning abilities, even in underrepresented languages such as Bengali and Swahili. Finally, we show that the multilingual reasoning abilities of language models extend to other tasks such as commonsense reasoning and word-in-context semantic judgment. The MGSM benchmark is publicly available at https://github.com/google-research/url-nlp.
Semantic representation learning for sentences is an important and well-studied problem in NLP. The current trend for this task involves training a Transformer-based sentence encoder through a contrastive objective with text, i.e., clustering sentences with semantically similar meanings and scattering others. In this work, we find the performance of Transformer models as sentence encoders can be improved by training with multi-modal multi-task losses, using unpaired examples from another modality (e.g., sentences and unrelated image/audio data). In particular, besides learning by the contrastive loss on text, our model clusters examples from a non-linguistic domain (e.g., visual/audio) with a similar contrastive loss at the same time. The reliance of our framework on unpaired non-linguistic data makes it language-agnostic, enabling it to be widely applicable beyond English NLP. Experiments on 7 semantic textual similarity benchmarks reveal that models trained with the additional non-linguistic (images/audio) contrastive objective lead to higher quality sentence embeddings. This indicates that Transformer models are able to generalize better by doing a similar task (i.e., clustering) with unpaired examples from different modalities in a multi-task fashion.
Suicide is a major public health crisis. With more than 20,000,000 suicide attempts each year, the early detection of suicidal intent has the potential to save hundreds of thousands of lives. Traditional mental health screening methods are time-consuming, costly, and often inaccessible to disadvantaged populations; online detection of suicidal intent using machine learning offers a viable alternative. Here we present Robin, the largest non-keyword generated suicidal corpus to date, consisting of over 1.1 million online forum postings. In addition to its unprecedented size, Robin is specially constructed to include various categories of suicidal text, such as suicide bereavement and flippant references, better enabling models trained on Robin to learn the subtle nuances of text expressing suicidal ideation. Experimental results achieve state-of-the-art performance for the classification of suicidal text, both with traditional methods like logistic regression (F1=0.85), as well as with large-scale pre-trained language models like BERT (F1=0.92). Finally, we release the Robin dataset publicly as a machine learning resource with the potential to drive the next generation of suicidal sentiment research.
Machine learning (ML) models have been applied to a wide range of natural language processing (NLP) tasks in recent years. In addition to making accurate decisions, the necessity of understanding how models make their decisions has become apparent in many applications. To that end, many interpretability methods that help explain the decision processes of ML models have been developed. Yet, there currently exists no widely-accepted metric to evaluate the quality of explanations generated by these methods. As a result, there currently is no standard way of measuring to what degree an interpretability method achieves an intended objective. Moreover, there is no accepted standard of performance by which we can compare and rank the current existing interpretability methods. In this paper, we propose a novel metric for quantifying the quality of explanations generated by interpretability methods. We compute the metric on three NLP tasks using six interpretability methods and present our results.
The impressive performance of GPT-3 using natural language prompts and in-context learning has inspired work on better fine-tuning of moderately-sized models under this paradigm. Following this line of work, we present a contrastive learning framework that clusters inputs from the same class for better generality of models trained with only limited examples. Specifically, we propose a supervised contrastive framework that clusters inputs from the same class under different augmented "views" and repel the ones from different classes. We create different "views" of an example by appending it with different language prompts and contextual demonstrations. Combining a contrastive loss with the standard masked language modeling (MLM) loss in prompt-based few-shot learners, the experimental results show that our method can improve over the state-of-the-art methods in a diverse set of 15 language tasks. Our framework makes minimal assumptions on the task or the base model, and can be applied to many recent methods with little modification. The code will be made available at: https://github.com/yiren-jian/LM-SupCon.
Few-shot language learners adapt knowledge from a pre-trained model to recognize novel classes from a few-labeled sentences. In such settings, fine-tuning a pre-trained language model can cause severe over-fitting. In this paper, we propose an Embedding Hallucination (EmbedHalluc) method, which generates auxiliary embedding-label pairs to expand the fine-tuning dataset. The hallucinator is trained by playing an adversarial game with the discriminator, such that the hallucinated embedding is indiscriminative to the real ones in the fine-tuning dataset. By training with the extended dataset, the language learner effectively learns from the diverse hallucinated embeddings to overcome the over-fitting issue. Experiments demonstrate that our proposed method is effective in a wide range of language tasks, outperforming current fine-tuning methods. Further, we show that EmbedHalluc outperforms other methods that address this over-fitting problem, such as common data augmentation, semi-supervised pseudo-labeling, and regularization. The code will be made available at: https://github.com/yiren-jian/EmbedHalluc.
The performance of existing text style transfer models is severely limited by the non-parallel datasets on which the models are trained. In non-parallel datasets, no direct mapping exists between sentences of the source and target style; the style transfer models thus only receive weak supervision of the target sentences during training, which often leads the model to discard too much style-independent information, or utterly fail to transfer the style. In this work, we propose LaMer, a novel text style transfer framework based on large-scale language models. LaMer first mines the roughly parallel expressions in the non-parallel datasets with scene graphs, and then employs MLE training, followed by imitation learning refinement, to leverage the intrinsic parallelism within the data. On two benchmark tasks (sentiment & formality transfer) and a newly proposed challenging task (political stance transfer), our model achieves qualitative advances in transfer accuracy, content preservation, and fluency. Further empirical and human evaluations demonstrate that our model not only makes training more efficient, but also generates more readable and diverse expressions than previous models.
Pre-trained language models (LMs) have been shown to memorize a substantial amount of knowledge from the pre-training corpora; however, they are still limited in recalling factually correct knowledge given a certain context. Hence, they tend to suffer from counterfactual or hallucinatory generation when used in knowledge-intensive natural language generation (NLG) tasks. Recent remedies to this problem focus on modifying either the pre-training or task fine-tuning objectives to incorporate knowledge, which normally require additional costly training or architecture modification of LMs for practical applications. We present Knowledge Infused Decoding (KID) -- a novel decoding algorithm for generative LMs, which dynamically infuses external knowledge into each step of the LM decoding. Specifically, we maintain a local knowledge memory based on the current context, interacting with a dynamically created external knowledge trie, and continuously update the local memory as a knowledge-aware constraint to guide decoding via reinforcement learning. On six diverse knowledge-intensive NLG tasks, task-agnostic LMs (e.g., GPT-2 and BART) armed with KID outperform many task-optimized state-of-the-art models, and show particularly strong performance in few-shot scenarios over seven related knowledge-infusion techniques. Human evaluation confirms KID's ability to generate more relevant and factual language for the input context when compared with multiple baselines. Finally, KID also alleviates exposure bias and provides stable generation quality when generating longer sequences. Code for KID is available at https://github.com/microsoft/KID.
Artificial intelligence, particularly through recent advancements in deep learning, has achieved exceptional performances in many tasks in fields such as natural language processing and computer vision. In addition to desirable evaluation metrics, a high level of interpretability is often required for these models to be reliably utilized. Therefore, explanations that offer insight into the process by which a model maps its inputs onto its outputs are much sought-after. Unfortunately, current black box nature of machine learning models is still an unresolved issue and this very nature prevents researchers from learning and providing explicative descriptions for a model's behavior and final predictions. In this work, we propose a novel framework utilizing Adversarial Inverse Reinforcement Learning that can provide global explanations for decisions made by a Reinforcement Learning model and capture intuitive tendencies that the model follows by summarizing the model's decision-making process.