The potential of large language models (LLMs) as decision support tools is increasingly being explored in fields such as business, engineering, and medicine, which often face challenging tasks of *decision-making under uncertainty*. In this paper, we show that directly prompting LLMs on these types of decision-making problems can yield poor results, especially as the problem complexity increases. To aid in these tasks, we propose DeLLMa (Decision-making Large Language Model assistant), a framework designed to enhance decision-making accuracy in uncertain environments. DeLLMa involves a multi-step reasoning procedure that integrates recent best practices in scaling *inference-time reasoning*, drawing upon principles from decision theory and utility theory, to provide an accurate and human-auditable decision-making process. We validate our procedure on multiple realistic decision-making environments, demonstrating that DeLLMa can consistently enhance the decision-making performance of leading language models, and achieve up to a 40% increase in accuracy over competing methods. Additionally, we show how performance improves when scaling compute at test time, and carry out human evaluations to benchmark components of DeLLMa.

Modern language models can process inputs across diverse languages and modalities. We hypothesize that models acquire this capability through learning a _shared representation space_ across heterogeneous data types (e.g., different languages and modalities), which places semantically similar inputs near one another, even if they are from different modalities/languages. We term this the _semantic hub hypothesis_, following the hub-and-spoke model from neuroscience (Patterson et al., 2007) which posits that semantic knowledge in the human brain is organized through a transmodal semantic "hub" which integrates information from various modality-specific ``spokes'' regions. We first show that model representations for semantically equivalent inputs in different languages are similar in the intermediate layers, and that this space can be interpreted using the model's dominant pretraining language via the logit lens. This tendency extends to other data types, including arithmetic expressions, code, and visual/audio inputs. Interventions in the shared representation space in one data type also predictably affect model outputs in other data types, suggesting that this shared representations space is not simply a vestigial byproduct of large-scale training on broad data, but something that is actively utilized by the model during input processing.

Denoising diffusion models enable conditional generation and density modeling of complex relationships like images and text. However, the nature of the learned relationships is opaque making it difficult to understand precisely what relationships between words and parts of an image are captured, or to predict the effect of an intervention. We illuminate the fine-grained relationships learned by diffusion models by noticing a precise relationship between diffusion and information decomposition. Exact expressions for mutual information and conditional mutual information can be written in terms of the denoising model. Furthermore, ${pointwise}$ estimates can be easily estimated as well, allowing us to ask questions about the relationships between specific images and captions. Decomposing information even further to understand which variables in a high-dimensional space carry information is a long-standing problem. For diffusion models, we show that a natural non-negative decomposition of mutual information emerges, allowing us to quantify informative relationships between words and pixels in an image. We exploit these new relations to measure the compositional understanding of diffusion models, to do unsupervised localization of objects in images, and to measure effects when selectively editing images through prompt interventions.

We analyze the masked language modeling pretraining objective function from the perspective of the distributional hypothesis.We investigate whether better sample efficiency and the better generalization capability of models pretrained with masked language modeling can be attributed to the semantic similarity encoded in the pretraining data's distributional property.Via a synthetic dataset, our analysis suggests that distributional property indeed leads to the better sample efficiency of pretrained masked language models, but does not fully explain the generalization capability.We also conduct analyses over two real-world datasets and demonstrate that the distributional property does not explain the generalization ability of pretrained natural language models either.Our results illustrate our limited understanding of model pretraining and provide future research directions. 1

Abstract We introduce ART, a new corpus-level autoencoding approach for training dense retrieval models that does not require any labeled training data. Dense retrieval is a central challenge for open-domain tasks, such as Open QA, where state-of-the-art methods typically require large supervised datasets with custom hard-negative mining and denoising of positive examples. ART, in contrast, only requires access to unpaired inputs and outputs (e.g., questions and potential answer passages). It uses a new passage-retrieval autoencoding scheme, where (1) an input question is used to retrieve a set of evidence passages, and (2) the passages are then used to compute the probability of reconstructing the original question. Training for retrieval based on question reconstruction enables effective unsupervised learning of both passage and question encoders, which can be later incorporated into complete Open QA systems without any further finetuning. Extensive experiments demonstrate that ART obtains state-of-the-art results on multiple QA retrieval benchmarks with only generic initialization from a pre-trained language model, removing the need for labeled data and task-specific losses.1 Our code and model checkpoints are available at: https://github.com/DevSinghSachan/art.

Hao Peng, Jungo Kasai, Nikolaos Pappas, Dani Yogatama, Zhaofeng Wu, Lingpeng Kong, Roy Schwartz, Noah Smith. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Text generation is of great importance to many natural language processing applications. However, maximization-based decoding methods (e.g., beam search) of neural language models often lead to degenerate solutions---the generated text is unnatural and contains undesirable repetitions. Existing approaches introduce stochasticity via sampling or modify training objectives to decrease the probabilities of certain tokens (e.g., unlikelihood training). However, they often lead to solutions that lack coherence. In this work, we show that an underlying reason for model degeneration is the anisotropic distribution of token representations. We present a contrastive solution: (i) SimCTG, a contrastive training objective to calibrate the model's representation space, and (ii) a decoding method---contrastive search---to encourage diversity while maintaining coherence in the generated text. Extensive experiments and analyses on three benchmarks from two languages demonstrate that our proposed approach outperforms state-of-the-art text generation methods as evaluated by both human and automatic metrics.

Abstract We present a memory-augmented approach to condition an autoregressive language model on a knowledge graph. We represent the graph as a collection of relation triples and retrieve relevant relations for a given context to improve text generation. Experiments on WikiText-103, WMT19, and enwik8 English datasets demonstrate that our approach produces a better language model in terms of perplexity and bits per character. We also show that relational memory improves coherence, is complementary to token-based memory, and enables causal interventions. Our model provides a simple yet effective way to combine an autoregressive language model and a knowledge graph for more coherent and logical generation.

Jungo Kasai, Hao Peng, Yizhe Zhang, Dani Yogatama, Gabriel Ilharco, Nikolaos Pappas, Yi Mao, Weizhu Chen, Noah A. Smith. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. 2021.

Our world is open-ended, non-stationary, and constantly evolving; thus what we talk about and how we talk about it change over time. This inherent dynamic nature of language contrasts with the current static language modelling paradigm, which trains and evaluates models on utterances from overlapping time periods. Despite impressive recent progress, we demonstrate that Transformer-XL language models perform worse in the realistic setup of predicting future utterances from beyond their training period, and that model performance becomes increasingly worse with time. We find that, while increasing model size alone—a key driver behind recent progress—does not solve this problem, having models that continually update their knowledge with new information can indeed mitigate this performance degradation over time. Hence, given the compilation of ever-larger language modelling datasets, combined with the growing list of language-model-based NLP applications that require up-to-date factual knowledge about the world, we argue that now is the right time to rethink the static way in which we currently train and evaluate our language models, and develop adaptive language models that can remain up-to-date with respect to our ever-changing and non-stationary world. We publicly release our dynamic, streaming language modelling benchmarks for WMT and arXiv to facilitate language model evaluation that takes temporal dynamics into account.

We present an end-to-end differentiable training method for retrieval-augmented open-domain question answering systems that combine information from multiple retrieved documents when generating answers. We model retrieval decisions as latent variables over sets of relevant documents. Since marginalizing over sets of retrieved documents is computationally hard, we approximate this using an expectation-maximization algorithm. We iteratively estimate the value of our latent variable (the set of relevant documents for a given question) and then use this estimate to update the retriever and reader parameters. We hypothesize that such end-to-end training allows training signals to flow to the reader and then to the retriever better than staged-wise training. This results in a retriever that is able to select more relevant documents for a question and a reader that is trained on more accurate documents to generate an answer. Experiments on three benchmark datasets demonstrate that our proposed method outperforms all existing approaches of comparable size by 2-3% absolute exact match points, achieving new state-of-the-art results. Our results also demonstrate the feasibility of learning to retrieve to improve answer generation without explicit supervision of retrieval decisions.

Abstract We present a language model that combines a large parametric neural network (i.e., a transformer) with a non-parametric episodic memory component in an integrated architecture. Our model uses extended short-term context by caching local hidden states—similar to transformer-XL—and global long-term memory by retrieving a set of nearest neighbor tokens at each timestep. We design a gating function to adaptively combine multiple information sources to make a prediction. This mechanism allows the model to use either local context, short-term memory, or long-term memory (or any combination of them) on an ad hoc basis depending on the context. Experiments on word-based and character-based language modeling datasets demonstrate the efficacy of our proposed method compared to strong baselines.

State-of-the-art unsupervised multilingual models (e.g., multilingual BERT) have been shown to generalize in a zero-shot cross-lingual setting. This generalization ability has been attributed to the use of a shared subword vocabulary and joint training across multiple languages giving rise to deep multilingual abstractions. We evaluate this hypothesis by designing an alternative approach that transfers a monolingual model to new languages at the lexical level. More concretely, we first train a transformer-based masked language model on one language, and transfer it to a new language by learning a new embedding matrix with the same masked language modeling objective, freezing parameters of all other layers. This approach does not rely on a shared vocabulary or joint training. However, we show that it is competitive with multilingual BERT on standard cross-lingual classification benchmarks and on a new Cross-lingual Question Answering Dataset (XQuAD). Our results contradict common beliefs of the basis of the generalization ability of multilingual models and suggest that deep monolingual models learn some abstractions that generalize across languages. We also release XQuAD as a more comprehensive cross-lingual benchmark, which comprises 240 paragraphs and 1190 question-answer pairs from SQuAD v1.1 translated into ten languages by professional translators.

We review motivations, definition, approaches, and methodology for unsupervised crosslingual learning and call for a more rigorous position in each of them.An existing rationale for such research is based on the lack of parallel data for many of the world's languages.However, we argue that a scenario without any parallel data and abundant monolingual data is unrealistic in practice.We also discuss different training signals that have been used in previous work, which depart from the pure unsupervised setting.We then describe common methodological issues in tuning and evaluation of unsupervised cross-lingual models and present best practices.Finally, we provide a unified outlook for different types of research in this area (i.e., cross-lingual word embeddings, deep multilingual pretraining, and unsupervised machine translation) and argue for comparable evaluation of these models.

Textual representation learners trained on large amounts of data have achieved notable success on downstream tasks; intriguingly, they have also performed well on challenging tests of syntactic competence. Hence, it remains an open question whether scalable learners like BERT can become fully proficient in the syntax of natural language by virtue of data scale alone, or whether they still benefit from more explicit syntactic biases. To answer this question, we introduce a knowledge distillation strategy for injecting syntactic biases into BERT pretraining, by distilling the syntactically informative predictions of a hierarchical—albeit harder to scale—syntactic language model. Since BERT models masked words in bidirectional context, we propose to distill the approximate marginal distribution over words in context from the syntactic LM. Our approach reduces relative error by 2–21% on a diverse set of structured prediction tasks, although we obtain mixed results on the GLUE benchmark. Our findings demonstrate the benefits of syntactic biases, even for representation learners that exploit large amounts of data, and contribute to a better understanding of where syntactic biases are helpful in benchmarks of natural language understanding.

Po-Sen Huang, Robert Stanforth, Johannes Welbl, Chris Dyer, Dani Yogatama, Sven Gowal, Krishnamurthy Dvijotham, Pushmeet Kohli. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

We introduce a lifelong language learning setup where a model needs to learn from a stream of text examples without any dataset identifier. We propose an episodic memory model that performs sparse experience replay and local adaptation to mitigate catastrophic forgetting in this setup. Experiments on text classification and question answering demonstrate the complementary benefits of sparse experience replay and local adaptation to allow the model to continuously learn from new datasets. We also show that the space complexity of the episodic memory module can be reduced significantly (~50-90%) by randomly choosing which examples to store in memory with a minimal decrease in performance. We consider an episodic memory component as a crucial building block of general linguistic intelligence and see our model as a first step in that direction.

Abstract We present two studies on neural network architectures that learn to represent sentences by composing their words according to automatically induced binary trees, without ever being shown a correct parse tree. We use Tree-Long Short-Term Memories (LSTMs) as our composition function, applied along a tree structure found by a differentiable natural language chart parser. The models simultaneously optimise both the composition function and the parser, thus eliminating the need for externally provided parse trees, which are normally required for Tree-LSTMs. They can therefore be seen as tree-based recurrent neural networks that are unsupervised with respect to the parse trees. Due to being fully differentiable, the models are easily trained with an off-the-shelf gradient descent method and backpropagation. In the first part of this paper, we introduce a model based on the CKY chart parser, and evaluate its downstream performance on a natural language inference task and a reverse dictionary task. Further, we show how its performance can be improved with an attention mechanism which fully exploits the parse chart, by attending over all possible subspans of the sentence. We find that our approach is competitive against similar models of comparable size and outperforms Tree-LSTMs that use trees produced by a parser. Finally, we present an alternative architecture based on a shift-reduce parser. We perform an analysis of the trees induced by both our models, to investigate whether they are consistent with each other and across re-runs, and whether they resemble the trees produced by a standard parser.

Adhiguna Kuncoro, Chris Dyer, John Hale, Dani Yogatama, Stephen Clark, Phil Blunsom. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2018.

Solving algebraic word problems requires executing a series of arithmetic operations-a program-to obtain a final answer.However, since programs can be arbitrarily complicated, inducing them directly from question-answer pairs is a formidable challenge.To make this task more feasible, we solve these problems by generating answer rationales, sequences of natural language and human-readable mathematical expressions that derive the final answer through a series of small steps.Although rationales do not explicitly specify programs, they provide a scaffolding for their structure via intermediate milestones.To evaluate our approach, we have created a new 100,000-sample dataset of questions, answers and rationales.Experimental results show that indirect supervision of program learning via answer rationales is a promising strategy for inducing arithmetic programs.

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech–two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, enabling experiments that previously took weeks to now run in days. This allows us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

Manaal Faruqui, Yulia Tsvetkov, Dani Yogatama, Chris Dyer, Noah A. Smith. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2015.

The most successful approaches to extrac-tive text summarization seek to maximize bigram coverage subject to a budget con-straint. In this work, we propose instead to maximize semantic volume. We em-bed each sentence in a semantic space and construct a summary by choosing a sub-set of sentences whose convex hull max-imizes volume in that space. We provide a greedy algorithm based on the Gram-Schmidt process to efficiently perform volume maximization. Our method out-performs the state-of-the-art summariza-tion approaches on benchmark datasets. 1

When applying machine learning to problems in NLP, there are many choices to make about how to represent input texts.They can have a big effect on performance, but they are often uninteresting to researchers or practitioners who simply need a module that performs well.We apply sequential model-based optimization over this space of choices and show that it makes standard linear models competitive with more sophisticated, expensive state-ofthe-art methods based on latent variables or neural networks on various topic classification and sentiment analysis problems.Our approach is a first step towards black-box NLP systems that work with raw text and do not require manual tuning.

We propose a new method for learning word representations using hierarchical regularization in sparse coding inspired by the linguistic study of word meanings. We show an efficient learning algorithm based on stochastic proximal methods that is significantly faster than previous approaches, making it possible to perform hierarchical sparse coding on a corpus of billions of word tokens. Experiments on various benchmark tasks—word similarity ranking, syntactic and semantic analogies, sentence completion, and sentiment analysis—demonstrate that the method outperforms or is competitive with state-of-the-art methods.

We introduce three linguistically motivated structured regularizers based on parse trees, topics, and hierarchical word clusters for text categorization.These regularizers impose linguistic bias in feature weights, enabling us to incorporate prior knowledge into conventional bagof-words models.We show that our structured regularizers consistently improve classification accuracies compared to standard regularizers that penalize features in isolation (such as lasso, ridge, and elastic net regularizers) on a range of datasets for various text prediction problems: topic classification, sentiment analysis, and forecasting.

We propose a fast and effective algorithm for automatic hyperparameter tuning that can generalize across datasets. Our method is an instance of sequential model-based optimization (SMBO) that transfers information by constructing a common response surface for all datasets, similar to Bardenet et al. (2013). The time complexity of reconstructing the response surface at every SMBO iteration in our method is linear in the number of trials (significantly less than previous work with comparable performance), allowing the method to realistically scale to many more datasets. Specifically, we use deviations from the per-dataset mean as the response values. We empirically show the superiority of our method on a large number of synthetic and real-world datasets for tuning hyperparameters of logistic regression and ensembles of classifiers.

In many high-dimensional learning problems, only some parts of an observation are important to the prediction task; for example, the cues to correctly categorizing a document may lie in a handful of its sentences. We introduce a learning algorithm that exploits this intuition by encoding it in a regularizer. Specifically, we apply the sparse overlapping group lasso with one group for every bundle of features occurring together in a training-data sentence, leading to thousands to millions of overlapping groups. We show how to efficiently solve the resulting optimization challenge using the alternating directions method of multipliers. We find that the resulting method significantly outperforms competitive baselines (standard ridge, lasso, and elastic net regularizers) on a suite of real-world text categorization problems.

We present a probabilistic language model that captures temporal dynamics and conditions on arbitrary non-linguistic context features. These context features serve as important indicators of language changes that are otherwise difficult to capture using text data by itself. We learn our model in an efficient online fashion that is scalable for large, streaming data. With five streaming datasets from two different genres—economics news articles and social media—we evaluate our model on the task of sequential language modeling. Our model consistently outperforms competing models.