Prior work shows that program-aided reasoning, in which large language models (LLMs) are combined with programs written in programming languages such as Python, can significantly improve accuracy on various reasoning tasks. However, while accuracy is essential, it is also important for such reasoners to "know what they know", which can be quantified through the calibration of the model. In this paper, we compare the calibration of Program Aided Language Models (PAL) and text-based Chain-of-thought (COT) prompting techniques over 5 datasets and 2 model types: LLaMA models and OpenAI models. Our results indicate that PAL leads to improved calibration in 75% of the instances. Our analysis uncovers that prompting styles that produce lesser diversity in generations also have more calibrated results, and thus we also experiment with inducing lower generation diversity using temperature scaling and find that for certain temperatures, PAL is not only more accurate but is also more calibrated than COT. Overall, we demonstrate that, in the majority of cases, program-aided reasoners better know what they know than text-based counterparts.
Do machines and humans process language in similar ways? A recent line of research has hinted in the affirmative, demonstrating that human brain signals can be effectively predicted using the internal representations of language models (LMs). This is thought to reflect shared computational principles between LMs and human language processing. However, there are also clear differences in how LMs and humans acquire and use language, even if the final task they are performing is the same. Despite this, there is little work exploring systematic differences between human and machine language processing using brain data. To address this question, we examine the differences between LM representations and the human brain's responses to language, specifically by examining a dataset of Magnetoencephalography (MEG) responses to a written narrative. In doing so we identify three phenomena that, in prior work, LMs have been found to not capture well: emotional understanding, figurative language processing, and physical commonsense. By fine-tuning LMs on datasets related to these phenomena, we observe that fine-tuned LMs show improved alignment with human brain responses across these tasks. Our study implies that the observed divergences between LMs and human brains may stem from LMs' inadequate representation of these specific types of knowledge.
On-the-fly retrieval of relevant knowledge has proven an essential element of reliable systems for tasks such as open-domain question answering and fact verification. However, because retrieval systems are not perfect, generation models are required to generate outputs given partially or entirely irrelevant passages. This can cause over- or under-reliance on context, and result in problems in the generated output such as hallucinations. To alleviate these problems, we propose FILCO, a method that improves the quality of the context provided to the generator by (1) identifying useful context based on lexical and information-theoretic approaches, and (2) training context filtering models that can filter retrieved contexts at test time. We experiment on six knowledge-intensive tasks with FLAN-T5 and LLaMa2, and demonstrate that our method outperforms existing approaches on extractive question answering (QA), complex multi-hop and long-form QA, fact verification, and dialog generation tasks. FILCO effectively improves the quality of context, whether or not it supports the canonical output.
We consider the task of optimally fine-tuning pre-trained multilingual models, given small amounts of unlabelled target data and an annotation budget. In this paper, we introduce DEMUX, a framework that prescribes the exact data-points to label from vast amounts of unlabelled multilingual data, having unknown degrees of overlap with the target set. Unlike most prior works, our end-to-end framework is language-agnostic, accounts for model representations, and supports multilingual target configurations. Our active learning strategies rely upon distance and uncertainty measures to select task-specific neighbors that are most informative to label, given a model. DeMuX outperforms strong baselines in 84% of the test cases, in the zero-shot setting of disjoint source and target language sets (including multilingual target pools), across three models and four tasks. Notably, in low-budget settings (5-100 examples), we observe gains of up to 8-11 F1 points for token-level tasks, and 2-5 F1 for complex tasks. Our code is released here: https://github.com/simran-khanuja/demux.
As large language models (LLMs) become more capable, there is growing excitement about the possibility of using LLMs as proxies for humans in real-world tasks where subjective labels are desired, such as in surveys and opinion polling. One widely-cited barrier to the adoption of LLMs is their sensitivity to prompt wording -- but interestingly, humans also display sensitivities to instruction changes in the form of response biases. As such, we argue that if LLMs are going to be used to approximate human opinions, it is necessary to investigate the extent to which LLMs also reflect human response biases, if at all. In this work, we use survey design as a case study, where human response biases caused by permutations in wordings of ``prompts'' have been extensively studied. Drawing from prior work in social psychology, we design a dataset and propose a framework to evaluate whether LLMs exhibit human-like response biases in survey questionnaires. Our comprehensive evaluation of nine models shows that popular open and commercial LLMs generally fail to reflect human-like behavior. These inconsistencies tend to be more prominent in models that have been instruction fine-tuned. Furthermore, even if a model shows a significant change in the same direction as humans, we find that perturbations that are not meant to elicit significant changes in humans may also result in a similar change, suggesting that such a result could be partially due to other spurious correlations. These results highlight the potential pitfalls of using LLMs to substitute humans in parts of the annotation pipeline, and further underscore the importance of finer-grained characterizations of model behavior. Our code, dataset, and collected samples are available at https://github.com/lindiatjuatja/BiasMonkey
One of the challenges in language teaching is how best to organize rules regarding syntax, semantics, or phonology in a meaningful manner. This not only requires content creators to have pedagogical skills, but also have that language's deep understanding. While comprehensive materials to develop such curricula are available in English and some broadly spoken languages, for many other languages, teachers need to manually create them in response to their students' needs. This is challenging because i) it requires that such experts be accessible and have the necessary resources, and ii) describing all the intricacies of a language is time-consuming and prone to omission. In this work, we aim to facilitate this process by automatically discovering and visualizing grammar descriptions. We extract descriptions from a natural text corpus that answer questions about morphosyntax (learning of word order, agreement, case marking, or word formation) and semantics (learning of vocabulary). We apply this method for teaching two Indian languages, Kannada and Marathi, which, unlike English, do not have well-developed resources for second language learning. To assess the perceived utility of the extracted material, we enlist the help of language educators from schools in North America to perform a manual evaluation, who find the materials have potential to be used for their lesson preparation and learner evaluation.
Idioms are common in everyday language, but often pose a challenge to translators because their meanings do not follow from the meanings of their parts. Despite significant advances, machine translation systems still struggle to translate idiomatic expressions. We provide a simple characterization of idiomatic translation and related issues. This allows us to conduct a synthetic experiment revealing a tipping point at which transformer-based machine translation models correctly default to idiomatic translations. To expand multilingual resources, we compile a dataset of ~4k natural sentences containing idiomatic expressions in French, Finnish, and Japanese. To improve translation of natural idioms, we introduce two straightforward yet effective techniques: the strategic upweighting of training loss on potentially idiomatic sentences, and using retrieval-augmented models. This not only improves the accuracy of a strong pretrained MT model on idiomatic sentences by up to 13% in absolute accuracy, but also holds potential benefits for non-idiomatic sentences.
Humans are social beings; we pursue social goals in our daily interactions, which is a crucial aspect of social intelligence. Yet, AI systems' abilities in this realm remain elusive. We present SOTOPIA, an open-ended environment to simulate complex social interactions between artificial agents and evaluate their social intelligence. In our environment, agents role-play and interact under a wide variety of scenarios; they coordinate, collaborate, exchange, and compete with each other to achieve complex social goals. We simulate the role-play interaction between LLM-based agents and humans within this task space and evaluate their performance with a holistic evaluation framework called SOTOPIA-Eval. With SOTOPIA, we find significant differences between these models in terms of their social intelligence, and we identify a subset of SOTOPIA scenarios, SOTOPIA-hard, that is generally challenging for all models. We find that on this subset, GPT-4 achieves a significantly lower goal completion rate than humans and struggles to exhibit social commonsense reasoning and strategic communication skills. These findings demonstrate SOTOPIA's promise as a general platform for research on evaluating and improving social intelligence in artificial agents.
Minimum Bayes Risk (MBR) decoding is a method for choosing the outputs of a machine learning system based not on the output with the highest probability, but the output with the lowest risk (expected error) among multiple candidates. It is a simple but powerful method: for an additional cost at inference time, MBR provides reliable several-point improvements across metrics for a wide variety of tasks without any additional data or training. Despite this, MBR is not frequently applied in NLP works, and knowledge of the method itself is limited. We first provide an introduction to the method and the recent literature. We show that several recent methods that do not reference MBR can be written as special cases of MBR; this reformulation provides additional theoretical justification for the performance of these methods, explaining some results that were previously only empirical. We provide theoretical and empirical results about the effectiveness of various MBR variants and make concrete recommendations for the application of MBR in NLP models, including future directions in this area.