Developing internal world models for artificial agents opens an efficient channel for humans to communicate with and control them. In addition to updating policies, humans can modify the world models of these agents in order to influence their decisions.The challenge, however, is that currently existing world models are difficult for humans to adapt because they lack a natural communication interface. Aimed at addressing this shortcoming, we develop *Language-Guided World Models* (LWMs), which can capture environment dynamics by reading language descriptions. These models enhance agent communication efficiency, allowing humans to simultaneously alter their behavior on multiple tasks with concise language feedback. They also enable agents to self-learn from texts originally written to instruct humans. To facilitate the development of LWMs, we design a challenging benchmark based on the game of MESSENGER (Hanjie et al., 2021), requiring compositional generalization to new language descriptions and environment dynamics. Our experiments reveal that the current state-of-the-art Transformer architecture performs poorly on this benchmark, motivating us to design a more robust architecture. To showcase the practicality of our proposed LWMs, we simulate a scenario where these models augment the interpretability and safety of an agent by enabling it to generate and discuss plans with a human before execution. By effectively incorporating language feedback on the plan, the models boost the agent performance in the real environment by up to three times without collecting any interactive experiences in this environment.

Quantitative evaluation metrics have been pivotal in gauging the advancements of AI systems like large language models (LLMs).However, due to the intricate nature of real-world tasks, a single scalar to quantify and compare performance trivializes the fine-grained nuances of model behavior. Additionally, metrics do not yield actionable diagnostics for model improvement, thus requiring extensive manual efforts of scientists, involving sifting through vast datasets and attempting hit-or-miss adjustments to training data or setups. In this work, we address the shortcomings of quantitative metrics by proposing QualEval, which uses automated qualitative evaluation as a vehicle for model improvement. QualEval uses a powerful LLM reasoner and our novel flexible linear programming solver to generate human-readable insights that when applied, accelerate model improvement. The insights are supported by a dashboard report with fine-grained visualizations and human-interpretable analyses. We corroborate the faithfulness of QualEval by demonstrating that leveraging its insights, for example, improves the absolute performance of the Llama 2 model by up to 15% points relative on a challenging dialogue task (DialogSum) when compared to baselines. QualEval successfully increases the pace and quality of model development by eliminating the need of arduous manual analysis, thus serving as a data-scientist-in-a-box.

In-context learning (ICL) performs tasks by prompting a large language model (LLM) using an instruction and a small set of annotated examples called demonstrations. Recent work has shown that precise details of the inputs used in the ICL prompt significantly impact performance, which has incentivized instruction selection algorithms. The effect of instruction-choice however is severely underexplored, with existing analyses restricted to shallow subsets of models and tasks, limiting the generalizability of their insights. We develop InstructEval, an ICL evaluation suite to conduct a thorough assessment of these techniques. The suite includes 13 open-sourced LLMs of varying scales from four model families, and covers nine tasks across three categories. Using the suite, we evaluate the relative performance of seven popular instruction selection methods over five metrics relevant to ICL. Our experiments reveal that using curated manually-written instructions or simple instructions without any task-specific descriptions often elicits superior ICL performance overall than that of automatic instruction-induction methods, pointing to a lack of generalizability among the latter. We release our evaluation suite (at https://github.com/princeton-nlp/InstructEval) for benchmarking instruction selection approaches and enabling more generalizable methods in this space.

We propose Referral-Augmented Retrieval (RAR), a simple technique that concatenates document indices with referrals: text from other documents that cite or link to the given document. We find that RAR provides significant performance gains for tasks across paper retrieval, entity retrieval, and open-domain question-answering in both zero-shot and in-domain (e.g., fine-tuned) settings. We examine how RAR provides especially strong improvements on more structured tasks, and can greatly outperform generative text expansion techniques such as DocT5Query and Query2Doc, with a 37% and 21% absolute improvement on ACL paper retrieval, respectively. We also compare three ways to aggregate referrals for RAR. Overall, we believe RAR can help revive and re-contextualize the classic information retrieval idea of using anchor texts to improve the representations of documents in a wide variety of corpuses in the age of neural retrieval.

As language models increase in size by the day, methods for efficient inference are critical to leveraging their capabilities for various applications. Prior work has investigated techniques like model pruning, knowledge distillation, and data multiplexing to increase model throughput without sacrificing accuracy. In this paper, we combine two such methods – structured pruning and data multiplexing – to compound the speedup gains obtained by either method. Our approach, PruMUX, obtains up to 7.5-29.5X throughput improvement over BERT-base model with accuracy threshold from 80% to 74%. We further study various combinations of parameters (such as sparsity and multiplexing factor) in the two techniques to provide a comprehensive analysis of the tradeoff between accuracy and throughput in the resulting models. We then propose Auto-PruMUX, a meta-level model that can predict the high-performance parameters for pruning and multiplexing given a desired accuracy loss budget, providing a practical method to leverage the combination effectively.

Large language models (LLMs) have shown incredible capabilities and transcended the natural language processing (NLP) community, with adoption throughout many services like healthcare, therapy, education, and customer service. Since users include people with critical information needs like students or patients engaging with chatbots, the safety of these systems is of prime importance. Legislation has recognized its significance and recently drafted a “Blueprint For An AI Bill Of Rights” which calls for domain experts to identify risks and potential impact of AI systems. To this end, we systematically evaluate toxicity in over half a million generations of ChatGPT, a popular dialogue-based LLM. We find that setting the system parameter of ChatGPT by assigning it a persona, say that of the boxer Muhammad Ali, significantly increases the toxicity of generations. Depending on the persona assigned to ChatGPT, its toxicity can increase up to 6×, with outputs engaging in incorrect stereotypes, harmful dialogue, and hurtful opinions. Furthermore, we find concerning patterns where specific entities (e.g., certain races) are targeted more than others (3× more) irrespective of the assigned persona, reflecting discriminatory biases in the model. Our findings show that multiple provisions in the legislative blueprint are being violated, and we hope that the broader AI community rethinks the efficacy of current safety guardrails and develops better techniques that lead to robust, safe, and trustworthy AI.

The widespread adoption of large language models such as ChatGPT and Bard has led to unprecedented demand for these technologies. The burgeoning cost of inference for ever-increasing model sizes coupled with hardware shortages has limited affordable access and poses a pressing need for efficiency approaches geared towards high throughput and performance. Multi-input multi-output (MIMO) algorithms such as data multiplexing, offer a promising solution with a many-fold increase in throughput by performing inference for multiple inputs at the cost of a single input. Yet these approaches are not currently performant enough to be deployed in modern systems. We change that by developing MUX-PLMs, a class of high throughput pre-trained language models (PLMs) trained with data multiplexing, that can be fine-tuned for any downstream task to yield high-throughput high-performance. Our novel multiplexing and demultiplexing modules proficiently entangle and disentangle inputs, and enable high-performance high throughput MUX-PLMs that are competitive with vanilla PLMs while achieving 2x/5x inference speedup with only a 1-4 % drop on a broad suite of tasks.

Anthropomorphization is the tendency to attribute human-like traits to non-human entities. It is prevalent in many social contexts – children anthropomorphize toys, adults do so with brands, and it is a literary device. It is also a versatile tool in science, with behavioral psychology and evolutionary biology meticulously documenting its consequences. With widespread adoption of AI systems, and the push from stakeholders to make it human-like through alignment techniques, human voice, and pictorial avatars, the tendency for users to anthropomorphize it increases significantly. We take a dyadic approach to understanding this phenomenon with large language models (LLMs) by studying (1) the objective legal implications, as analyzed through the lens of the recent blueprint of AI bill of rights and the (2) subtle psychological aspects customization and anthropomorphization. We find that anthropomorphized LLMs customized for different user bases violate multiple provisions in the legislative blueprint. In addition, we point out that anthropomorphization of LLMs affects the influence they can have on their users, thus having the potential to fundamentally change the nature of human-AI interaction, with potential for manipulation and negative influence. With LLMs being hyper-personalized for vulnerable groups like children and patients among others, our work is a timely and important contribution. We propose a conservative strategy for the cautious use of anthropomorphization to improve trustworthiness of AI systems.

Semantic textual similarity (STS) has been a cornerstone task in NLP that measures the degree of similarity between a pair of sentences, with applications in information retrieval, question answering, and embedding methods. However, it is an inherently ambiguous task, with the sentence similarity depending on the specific aspect of interest. We resolve this ambiguity by proposing a novel task called conditional STS (C-STS) which measures similarity conditioned on an aspect elucidated in natural language (hereon, condition). As an example, the similarity between the sentences “The NBA player shoots a three-pointer.” and “A man throws a tennis ball into the air to serve.” is higher for the condition “The motion of the ball.” (both upward) and lower for “The size of the ball.” (one large and one small). C-STS’s advantages are two-fold: (1) it reduces the subjectivity and ambiguity of STS, and (2) enables fine-grained similarity evaluation using diverse conditions. C-STS contains almost 20,000 instances from diverse domains and we evaluate several state-of-the-art models to demonstrate that even the most performant fine-tuning and in-context learning models (GPT-4, Flan, SimCSE) find it challenging, with Spearman correlation scores of <50. We encourage the community to evaluate their models on C-STS to provide a more holistic view of semantic similarity and natural language understanding.

The widespread adoption of large language models such as ChatGPT and Bard has led to unprecedented demand for these technologies. The burgeoning cost of inference for ever-increasing model sizes coupled with hardware shortages has limited affordable access and poses a pressing need for efficiency approaches geared towards high throughput and performance. Multi-input multi-output (MIMO) algorithms such as data multiplexing, offer a promising solution with a many-fold increase in throughput by performing inference for multiple inputs at the cost of a single input. Yet these approaches are not currently performant enough to be deployed in modern systems. We change that by developing MUX-PLMs, a class of high throughput pre-trained language models (PLMs) trained with data multiplexing, that can be fine-tuned for any downstream task to yield high-throughput high-performance. Our novel multiplexing and demultiplexing modules proficiently entangle and disentangle inputs, and enable high-performance high throughput that are competitive with vanilla PLMs while achieving 2x/5x inference speedup with only a 1−4% drop on a broad suite of tasks.

We introduce CARETS, a systematic test suite to measure consistency and robustness of modern VQA models through a series of six fine-grained capability tests. In contrast to existing VQA test sets, CARETS features balanced question generation to create pairs of instances to test models, with each pair focusing on a specific capability such as rephrasing, logical symmetry or image obfuscation. We evaluate six modern VQA systems on CARETS and identify several actionable weaknesses in model comprehension, especially with concepts such as negation, disjunction, or hypernym invariance. Interestingly, even the most sophisticated models are sensitive to aspects such as swapping the order of terms in a conjunction or varying the number of answer choices mentioned in the question. We release CARETS to be used as an extensible tool for evaluating multi-modal model robustness.

While recent work on multilingual language models has demonstrated their capacity for cross-lingual zero-shot transfer on downstream tasks, there is a lack of consensus in the community as to what shared properties between languages enable such transfer. Analyses involving pairs of natural languages are often inconclusive and contradictory since languages simultaneously differ in many linguistic aspects. In this paper, we perform a large-scale empirical study to isolate the effects of various linguistic properties by measuring zero-shot transfer between four diverse natural languages and their counterparts constructed by modifying aspects such as the script, word order, and syntax. Among other things, our experiments show that the absence of sub-word overlap significantly affects zero-shot transfer when languages differ in their word order, and there is a strong correlation between transfer performance and word embedding alignment between languages (e.g., 𝜌_s=0.94 on the task of NLI). Our results call for focus in multilingual models on explicitly improving word embedding alignment between languages rather than relying on its implicit emergence.

A growing line of work has investigated the development of neural NLP models that can produce rationales–subsets of input that can explain their model predictions. In this paper, we ask whether such rationale models can provide robustness to adversarial attacks in addition to their interpretable nature. Since these models need to first generate rationales (“rationalizer”) before making predictions (“predictor”), they have the potential to ignore noise or adversarially added text by simply masking it out of the generated rationale. To this end, we systematically generate various types of ‘AddText’ attacks for both token and sentence-level rationalization tasks and perform an extensive empirical evaluation of state-of-the-art rationale models across five different tasks. Our experiments reveal that the rationale models promise to improve robustness over AddText attacks while they struggle in certain scenarios–when the rationalizer is sensitive to position bias or lexical choices of attack text. Further, leveraging human rationale as supervision does not always translate to better performance. Our study is a first step towards exploring the interplay between interpretability and robustness in the rationalize-then-predict framework.

Text-based games simulate worlds and interact with players using natural language. Recent work has used them as a testbed for autonomous language-understanding agents, with the motivation being that understanding the meanings of words or semantics is a key component of how humans understand, reason, and act in these worlds. However, it remains unclear to what extent artificial agents utilize semantic understanding of the text. To this end, we perform experiments to systematically reduce the amount of semantic information available to a learning agent. Surprisingly, we find that an agent is capable of achieving high scores even in the complete absence of language semantics, indicating that the currently popular experimental setup and models may be poorly designed to understand and leverage game texts. To remedy this deficiency, we propose an inverse dynamics decoder to regularize the representation space and encourage exploration, which shows improved performance on several games including Zork I. We discuss the implications of our findings for designing future agents with stronger semantic understanding.

Recent work has demonstrated the vulnerability of modern text classifiers to universal adversarial attacks, which are input-agnostic sequences of words added to text processed by classifiers. Despite being successful, the word sequences produced in such attacks are often ungrammatical and can be easily distinguished from natural text. We develop adversarial attacks that appear closer to natural English phrases and yet confuse classification systems when added to benign inputs. We leverage an adversarially regularized autoencoder (ARAE) to generate triggers and propose a gradient-based search that aims to maximize the downstream classifier’s prediction loss. Our attacks effectively reduce model accuracy on classification tasks while being less identifiable than prior models as per automatic detection metrics and human-subject studies. Our aim is to demonstrate that adversarial attacks can be made harder to detect than previously thought and to enable the development of appropriate defenses.

In this paper, we explore the ability to model and infer personality types of opponents, predict their responses, and use this information to adapt a dialog agent’s high-level strategy in negotiation tasks. Inspired by the idea of incorporating a theory of mind (ToM) into machines, we introduce a probabilistic formulation to encapsulate the opponent’s personality type during both learning and inference. We test our approach on the CraigslistBargain dataset (He et al. 2018) and show that our method using ToM inference achieves a 20% higher dialog agreement rate compared to baselines on a mixed population of opponents. We also demonstrate that our model displays diverse negotiation behavior with different types of opponents.

Despite their impressive performance in NLP, self-attention networks were recently proved to be limited for processing formal languages with hierarchical structure, such as Dyck-k, the language consisting of well-nested parentheses of k types. This suggested that natural language can be approximated well with models that are too weak for formal languages, or that the role of hierarchy and recursion in natural language might be limited. We qualify this implication by proving that self-attention networks can process Dyck-(k, D), the subset of Dyck-k with depth bounded by D, which arguably better captures the bounded hierarchical structure of natural language. Specifically, we construct a hard-attention network with D+1 layers and O(log k) memory size (per token per layer) that recognizes Dyck-(k, D), and a soft-attention network with two layers and O(log k) memory size that generates Dyck-(k, D). Experiments show that self-attention networks trained on Dyck-(k, D) generalize to longer inputs with near-perfect accuracy, and also verify the theoretical memory advantage of self-attention networks over recurrent networks.

Learning representations of spatial references in natural language is a key challenge in tasks like autonomous navigation and robotic manipulation. Recent work has investigated various neural architectures for learning multi-modal representations for spatial concepts. However, the lack of explicit reasoning over entities makes such approaches vulnerable to noise in input text or state observations. In this paper, we develop effective models for understanding spatial references in text that are robust and interpretable, without sacrificing performance. We design a text-conditioned relation network whose parameters are dynamically computed with a cross-modal attention module to capture fine-grained spatial relations between entities. This design choice provides interpretability of learned intermediate outputs. Experiments across three tasks demonstrate that our model achieves superior performance, with a 17% improvement in predicting goal locations and a 15% improvement in robustness compared to state-of-the-art systems.

In this paper, we propose a simple and effective technique to allow for efficient self-supervised learning with bi-directional Transformers. Our approach is motivated by recent studies demonstrating that self-attention patterns in trained models contain a majority of non-linguistic regularities. We propose a computationally efficient auxiliary loss function to guide attention heads to conform to such patterns. Our method is agnostic to the actual pre-training objective and results in faster convergence of models as well as better performance on downstream tasks compared to the baselines, achieving state of the art results in low-resource settings. Surprisingly, we also find that linguistic properties of attention heads are not necessarily correlated with language modeling performance.

Text-based games present a unique challenge for autonomous agents to operate in natural language and handle enormous action spaces. In this paper, we propose the Contextual Action Language Model (CALM) to generate a compact set of action candidates at each game state. Our key insight is to train language models on human gameplay, where people demonstrate linguistic priors and a general game sense for promising actions conditioned on game history. We combine CALM with a reinforcement learning agent which re-ranks the generated action candidates to maximize in-game rewards. We evaluate our approach using the Jericho benchmark, on games unseen by CALM during training. Our method obtains a 69% relative improvement in average game score over the previous state-of-the-art model. Surprisingly, on half of these games, CALM is competitive with or better than other models that have access to ground truth admissible actions. Code and data are available at https://github.com/princeton-nlp/calm-textgame.

The interpretation of spatial references is highly contextual, requiring joint inference over both language and the environment. We consider the task of spatial reasoning in a simulated environment, where an agent can act and receive rewards. The proposed model learns a representation of the world steered by instruction text. This design allows for precise alignment of local neighborhoods with corresponding verbalizations, while also handling global references in the instructions. We train our model with reinforcement learning using a variant of generalized value iteration. The model outperforms state-of-the-art approaches on several metrics, yielding a 45% reduction in goal localization error.

This paper focuses on unsupervised modeling of morphological families, collectively comprising a forest over the language vocabulary. This formulation enables us to capture edge-wise properties reflecting single-step morphological derivations, along with global distributional properties of the entire forest. These global properties constrain the size of the affix set and encourage formation of tight morphological families. The resulting objective is solved using Integer Linear Programming (ILP) paired with contrastive estimation. We train the model by alternating between optimizing the local log-linear model and the global ILP objective. We evaluate our system on three tasks: root detection, clustering of morphological families, and segmentation. Our experiments demonstrate that our model yields consistent gains in all three tasks compared with the best published results.

Most state-of-the-art systems today produce morphological analysis based only on orthographic patterns. In contrast, we propose a model for unsupervised morphological analysis that integrates orthographic and semantic views of words. We model word formation in terms of morphological chains, from base words to the observed words, breaking the chains into parent-child relations. We use log-linear models with morpheme and word-level features to predict possible parents, including their modifications, for each word. The limited set of candidate parents for each word render contrastive estimation feasible. Our model consistently matches or outperforms five state-of-the-art systems on Arabic, English and Turkish.