Yoav Artzi

Varsha Kishore, Chao Wan, Justin Lovelace et al. · cmu

Differentiable Search Index is a recently proposed paradigm for document retrieval, that encodes information about a corpus of documents within the parameters of a neural network and directly maps queries to corresponding documents. These models have achieved state-of-the-art performances for document retrieval across many benchmarks. These kinds of models have a significant limitation: it is not easy to add new documents after a model is trained. We propose IncDSI, a method to add documents in real time (about 20-50ms per document), without retraining the model on the entire dataset (or even parts thereof). Instead we formulate the addition of documents as a constrained optimization problem that makes minimal changes to the network parameters. Although orders of magnitude faster, our approach is competitive with re-training the model on the whole dataset and enables the development of document retrieval systems that can be updated with new information in real-time. Our code for IncDSI is available at https://github.com/varshakishore/IncDSI.

Felix Wu, Kwangyoun Kim, Shinji Watanabe et al. · deepmind

We introduce Wav2Seq, the first self-supervised approach to pre-train both parts of encoder-decoder models for speech data. We induce a pseudo language as a compact discrete representation, and formulate a self-supervised pseudo speech recognition task -- transcribing audio inputs into pseudo subword sequences. This process stands on its own, or can be applied as low-cost second-stage pre-training. We experiment with automatic speech recognition (ASR), spoken named entity recognition, and speech-to-text translation. We set new state-of-the-art results for end-to-end spoken named entity recognition, and show consistent improvements on 20 language pairs for speech-to-text translation, even when competing methods use additional text data for training. Finally, on ASR, our approach enables encoder-decoder methods to benefit from pre-training for all parts of the network, and shows comparable performance to highly optimized recent methods.

Anya Ji, Noriyuki Kojima, Noah Rush et al. · berkeley

We introduce KiloGram, a resource for studying abstract visual reasoning in humans and machines. Drawing on the history of tangram puzzles as stimuli in cognitive science, we build a richly annotated dataset that, with >1k distinct stimuli, is orders of magnitude larger and more diverse than prior resources. It is both visually and linguistically richer, moving beyond whole shape descriptions to include segmentation maps and part labels. We use this resource to evaluate the abstract visual reasoning capacities of recent multi-modal models. We observe that pre-trained weights demonstrate limited abstract reasoning, which dramatically improves with fine-tuning. We also observe that explicitly describing parts aids abstract reasoning for both humans and models, especially when jointly encoding the linguistic and visual inputs. KiloGram is available at https://lil.nlp.cornell.edu/kilogram .

Alane Suhr, Yoav Artzi · berkeley

We propose and deploy an approach to continually train an instruction-following agent from feedback provided by users during collaborative interactions. During interaction, human users instruct an agent using natural language, and provide realtime binary feedback as they observe the agent following their instructions. We design a contextual bandit learning approach, converting user feedback to immediate reward. We evaluate through thousands of human-agent interactions, demonstrating 15.4% absolute improvement in instruction execution accuracy over time. We also show our approach is robust to several design variations, and that the feedback signal is roughly equivalent to the learning signal of supervised demonstration data.

Yilun Hua, Yoav Artzi

Humans spontaneously use increasingly efficient language as interactions progress, by adapting and forming ad-hoc conventions. This phenomenon has been studied extensively using reference games, showing properties of human language that go beyond relaying intents. It remains unexplored whether multimodal large language models (MLLMs) similarly increase communication efficiency during interactions, and what mechanisms they may adopt for this purpose. We introduce ICCA, an automated framework to evaluate such conversational adaptation as an in-context behavior in MLLMs. We evaluate several state-of-the-art MLLMs, and observe that while they may understand the increasingly efficient language of their interlocutor, they do not spontaneously make their own language more efficient over time. This latter ability can only be elicited in some models (e.g., GPT-4) with heavy-handed prompting. This shows that this property of linguistic interaction does not arise from current training regimes, even though it is a common hallmark of human language. ICCA is available at https://github.com/lil-lab/ICCA.

Noriyuki Kojima, Hadar Averbuch-Elor, Yoav Artzi

Key to tasks that require reasoning about natural language in visual contexts is grounding words and phrases to image regions. However, observing this grounding in contemporary models is complex, even if it is generally expected to take place if the task is addressed in a way that is conductive to generalization. We propose a framework to jointly study task performance and phrase grounding, and propose three benchmarks to study the relation between the two. Our results show that contemporary models demonstrate inconsistency between their ability to ground phrases and solve tasks. We show how this can be addressed through brute-force training on ground phrasing annotations, and analyze the dynamics it creates. Code and at available at https://github.com/lil-lab/phrase_grounding.

Ge Gao, Eunsol Choi, Yoav Artzi

We study learning from user feedback for extractive question answering by simulating feedback using supervised data. We cast the problem as contextual bandit learning, and analyze the characteristics of several learning scenarios with focus on reducing data annotation. We show that systems initially trained on a small number of examples can dramatically improve given feedback from users on model-predicted answers, and that one can use existing datasets to deploy systems in new domains without any annotation, but instead improving the system on-the-fly via user feedback.

Mustafa Omer Gul, Yoav Artzi

Systems with both language comprehension and generation capabilities can benefit from the tight connection between the two. This work studies coupling comprehension and generation with focus on continually learning from interaction with users. We propose techniques to tightly integrate the two capabilities for both learning and inference. We situate our studies in two-player reference games, and deploy various models for thousands of interactions with human users, while learning from interaction feedback signals. We show dramatic improvements in performance over time, with comprehension-generation coupling leading to performance improvements up to 26% in absolute terms and up to 17% higher accuracies compared to a non-coupled system. Our analysis also shows coupling has substantial qualitative impact on the system's language, making it significantly more human-like.

Nathan Godey, Yoav Artzi

The last layer of neural language models (LMs) projects output features of dimension $D$ to logits in dimension $V$, the size of the vocabulary, where usually $D \ll V$. This mismatch is known to raise risks of limited expressivity in neural LMs, creating a so-called softmax bottleneck. We show the softmax bottleneck is not only an expressivity bottleneck but also an optimization bottleneck. Backpropagating $V$-dimensional gradients through a rank-$D$ linear layer induces unavoidable compression, which alters the training feedback provided to the vast majority of the parameters. We present a theoretical analysis of this phenomenon and measure empirically that 95-99% of the gradient norm is suppressed by the output layer, resulting in vastly suboptimal update directions. We conduct controlled pretraining experiments showing that the gradient bottleneck makes trivial patterns unlearnable, and drastically affects the training dynamics of LLMs. We argue that this inherent flaw contributes to training inefficiencies at scale independently of the model architecture, and raises the need for new LM head designs.

Anne Wu, Kianté Brantley, Noriyuki Kojima et al.

We present lilGym, a new benchmark for language-conditioned reinforcement learning in visual environments. lilGym is based on 2,661 highly-compositional human-written natural language statements grounded in an interactive visual environment. We introduce a new approach for exact reward computation in every possible world state by annotating all statements with executable Python programs. Each statement is paired with multiple start states and reward functions to form thousands of distinct Markov Decision Processes of varying difficulty. We experiment with lilGym with different models and learning regimes. Our results and analysis show that while existing methods are able to achieve non-trivial performance, lilGym forms a challenging open problem. lilGym is available at https://lil.nlp.cornell.edu/lilgym/.

Paloma Sodhi, S. R. K. Branavan, Yoav Artzi et al.

Performing tasks on the web presents fundamental challenges to large language models (LLMs), including combinatorially large open-world tasks and variations across web interfaces. Simply specifying a large prompt to handle all possible behaviors and states is extremely complex, and results in behavior leaks between unrelated behaviors. Decomposition to distinct policies can address this challenge, but requires carefully handing off control between policies. We propose Stacked LLM Policies for Web Actions (SteP), an approach to dynamically compose policies to solve a diverse set of web tasks. SteP defines a Markov Decision Process where the state is a stack of policies representing the control state, i.e., the chain of policy calls. Unlike traditional methods that are restricted to static hierarchies, SteP enables dynamic control that adapts to the complexity of the task. We evaluate SteP against multiple baselines and web environments including WebArena, MiniWoB++, and a CRM. On WebArena, SteP improves (14.9\% to 33.5\%) over SOTA that use GPT-4 policies, while on MiniWob++, SteP is competitive with prior works while using significantly less data. Our code and data are available at https://asappresearch.github.io/webagents-step.

Jacob Sharf, Mustafa Omer Gul, Yoav Artzi

CB2 is a multi-agent platform to study collaborative natural language interaction in a grounded task-oriented scenario. It includes a 3D game environment, a backend server designed to serve trained models to human agents, and various tools and processes to enable scalable studies. We deploy CB2 at https://cb2.ai as a system demonstration with a learned instruction following model.

Anne Wu, Kianté Brantley, Yoav Artzi

This study evaluates three state-of-the-art MLLMs -- GPT-4V, Gemini Pro, and the open-source model IDEFICS -- on the compositional natural language vision reasoning task NLVR. Given a human-written sentence paired with a synthetic image, this task requires the model to determine the truth value of the sentence with respect to the image. Despite the strong performance demonstrated by these models, we observe they perform poorly on NLVR, which was constructed to require compositional and spatial reasoning, and to be robust for semantic and systematic biases.

Rachit Saluja, Jacob Rosenthal, Yoav Artzi et al.

Large Language Models (LLMs) have shown significant promise across various natural language processing tasks. However, their application in the field of pathology, particularly for extracting meaningful insights from unstructured medical texts such as pathology reports, remains underexplored and not well quantified. In this project, we leverage state-of-the-art language models, including the GPT family, Mistral models, and the open-source Llama models, to evaluate their performance in comprehensively analyzing pathology reports. Specifically, we assess their performance in cancer type identification, AJCC stage determination, and prognosis assessment, encompassing both information extraction and higher-order reasoning tasks. Based on a detailed analysis of their performance metrics in a zero-shot setting, we developed two instruction-tuned models: Path-llama3.1-8B and Path-GPT-4o-mini-FT. These models demonstrated superior performance in zero-shot cancer type identification, staging, and prognosis assessment compared to the other models evaluated.

William Huey, Huaxiaoyue Wang, Anne Wu et al.

We examine the problem of learning sequential tasks from a single visual demonstration. A key challenge arises when demonstrations are temporally misaligned due to variations in timing, differences in embodiment, or inconsistencies in execution. Existing approaches treat imitation as a distribution-matching problem, aligning individual frames between the agent and the demonstration. However, we show that such frame-level matching fails to enforce temporal ordering or ensure consistent progress. Our key insight is that matching should instead be defined at the level of sequences. We propose that perfect matching occurs when one sequence successfully covers all the subgoals in the same order as the other sequence. We present ORCA (ORdered Coverage Alignment), a dense per-timestep reward function that measures the probability of the agent covering demonstration frames in the correct order. On temporally misaligned demonstrations, we show that agents trained with the ORCA reward achieve $4.5$x improvement ($0.11 \rightarrow 0.50$ average normalized returns) for Meta-world tasks and $6.6$x improvement ($6.55 \rightarrow 43.3$ average returns) for Humanoid-v4 tasks compared to the best frame-level matching algorithms. We also provide empirical analysis showing that ORCA is robust to varying levels of temporal misalignment. Our code is available at https://github.com/portal-cornell/orca/

Zizhao Chen, Yoav Artzi

We propose KnotGym, an interactive environment for complex, spatial reasoning and manipulation. KnotGym includes goal-oriented rope manipulation tasks with varying levels of complexity, all requiring acting from pure image observations. Tasks are defined along a clear and quantifiable axis of complexity based on the number of knot crossings, creating a natural generalization test. KnotGym has a simple observation space, allowing for scalable development, yet it highlights core challenges in integrating acute perception, spatial reasoning, and grounded manipulation. We evaluate methods of different classes, including model-based RL, model-predictive control, and chain-of-thought reasoning, and illustrate the challenges KnotGym presents. KnotGym is available at https://github.com/lil-lab/knotgym.

Suwon Shon, Ankita Pasad, Felix Wu et al.

Progress in speech processing has been facilitated by shared datasets and benchmarks. Historically these have focused on automatic speech recognition (ASR), speaker identification, or other lower-level tasks. Interest has been growing in higher-level spoken language understanding tasks, including using end-to-end models, but there are fewer annotated datasets for such tasks. At the same time, recent work shows the possibility of pre-training generic representations and then fine-tuning for several tasks using relatively little labeled data. We propose to create a suite of benchmark tasks for Spoken Language Understanding Evaluation (SLUE) consisting of limited-size labeled training sets and corresponding evaluation sets. This resource would allow the research community to track progress, evaluate pre-trained representations for higher-level tasks, and study open questions such as the utility of pipeline versus end-to-end approaches. We present the first phase of the SLUE benchmark suite, consisting of named entity recognition, sentiment analysis, and ASR on the corresponding datasets. We focus on naturally produced (not read or synthesized) speech, and freely available datasets. We provide new transcriptions and annotations on subsets of the VoxCeleb and VoxPopuli datasets, evaluation metrics and results for baseline models, and an open-source toolkit to reproduce the baselines and evaluate new models.

Zizhao Chen, Mustafa Omer Gul, Yiwei Chen et al.

Multi-turn interactions between large language models (LLMs) and users naturally include implicit feedback signals. If an LLM responds in an unexpected way to an instruction, the user is likely to signal it by rephrasing the request, expressing frustration, or pivoting to an alternative task. Such signals are task-independent and occupy a relatively constrained subspace of language, allowing the LLM to identify them even if it fails on the actual task. We introduce ReSpect, a method to learn from such signals in past interactions via retrospection without additional annotations. We deploy ReSpect in a new multimodal interaction scenario, where humans instruct a multimodal LLM to solve an abstract reasoning task with a combinatorial solution space. Through thousands of interactions with humans, we show how ReSpect gradually improves task completion rate from 31% to 82%, all without any external annotation.

Yilun Hua, Evan Wang, Yoav Artzi

Humans communicate with increasing efficiency in multi-turn interactions, by adapting their language and forming ad-hoc conventions. In contrast, prior work shows that LLMs do not naturally show this behavior. We develop a post-training process to develop this ability through targeted fine-tuning on heuristically identified demonstrations of convention formation. We evaluate with two new benchmarks focused on this capability. First, we design a focused, cognitively-motivated interaction benchmark that consistently elicits strong convention formation trends in humans. Second, we create a new document-grounded reference completion task that reflects in-the-wild convention formation behavior. Our studies show significantly improved convention formation abilities in post-trained LLMs across the two evaluation methods.

Linxi Zhao, Sofian Zalouk, Christian K. Belardi et al. · cmu

Neural language models are black-boxes--both linguistic patterns and factual knowledge are distributed across billions of opaque parameters. This entangled encoding makes it difficult to reliably inspect, verify, or update specific facts. We introduce Limited Memory Language Models (LMLM), a new class of language models that externalizes factual knowledge to external database during pre-training rather than memorizing them. Our pre-training approach strategically masks externally retrieved factual values from the training loss, thereby teaching the model to perform targeted lookups rather than relying on memorization in model weights. Our experiments demonstrate that LMLMs achieve competitive performance compared to significantly larger LLMs on standard benchmarks, while offering the advantages of explicit, editable, and verifiable knowledge bases.

Saujas Vaduguru, Yilun Hua, Yoav Artzi et al.

Humans leverage shared conversational context to become increasingly successful and efficient at communicating over time. One manifestation of this is the formation of ad hoc linguistic conventions, which allow people to coordinate on short, less costly utterances that are understood using shared conversational context. We present a method to train large multimodal models to form conventions, enabling efficient communication. Our approach uses simulated reference games between models, and requires no additional human-produced data. In repeated reference games involving photographs and tangram images, our method enables models to communicate efficiently with people: reducing the message length by up to 41% while increasing success by 15% over the course of the interaction. Human listeners respond faster when interacting with our model that forms conventions. We also show that training based on success or cost alone is insufficient - both are necessary to elicit convention formation.

Yair Feldman, Yoav Artzi

A common strategy to reduce the computational costs of using long contexts in retrieval-augmented generation (RAG) with large language models (LLMs) is soft context compression, where the input sequence is transformed into a shorter continuous representation. We develop a lightweight and simple mean-pooling approach that consistently outperforms the widely used compression-tokens architecture, and study training the same compressor to output multiple compression ratios. We conduct extensive experiments across in-domain and out-of-domain QA datasets, as well as across model families, scales, and compression ratios. Overall, our simple mean-pooling approach achieves the strongest performance, with a relatively small drop when training for multiple compression ratios. More broadly though, across architectures and training regimes the trade-offs are more nuanced, illustrating the complex landscape of compression methods.

Giovanni Monea, Yair Feldman, Shankar Padmanabhan et al.

The scalability of large language models for long-context reasoning is severely constrained by the linear growth of their Transformer key-value cache, which incurs significant memory and computational costs. We posit that as a model generates reasoning tokens, the informational value of past generated tokens diminishes, creating an opportunity for compression. In this work, we propose to periodically compress the generation KV cache with a learned, special-purpose token and evict compressed entries. We train the model to perform this compression via a modified joint distillation and reinforcement learning (RL) framework. Our training method minimizes overhead over the conventional RL process, as it leverages RL outputs for distillation. Empirically, our method achieves a superior memory-accuracy Pareto frontier compared to both the model without cache compression and training-free compression techniques.

Anya Ji, Claire Augusta Bergey, Ron Eliav et al.

How do people talk about things they've never talked about before? One view suggests that a new shared naming system establishes an arbitrary link to a specific target, like proper names that cannot extend beyond their bearers. An alternative view proposes that forming a shared way of describing objects involves broader conceptual alignment, reshaping each individual's semantic space in ways that should generalize to new referents. We test these competing accounts in a dyadic communication study (N=302) leveraging the recently-released KiloGram dataset containing over 1,000 abstract tangram images. After pairs of participants coordinated on referential conventions for one set of images through repeated communication, we measured the extent to which their descriptions aligned for undiscussed images. We found strong evidence for generalization: partners showed increased alignment relative to their pre-test labels. Generalization also decayed nonlinearly with visual similarity (consistent with Shepard's law) and was robust across levels of the images' nameability. These findings suggest that ad hoc conventions are not arbitrary labels but reflect genuine conceptual coordination, with implications for theories of reference and the design of more adaptive language agents.

Ge Gao, Hung-Ting Chen, Yoav Artzi et al.

We study continually improving an extractive question answering (QA) system via human user feedback. We design and deploy an iterative approach, where information-seeking users ask questions, receive model-predicted answers, and provide feedback. We conduct experiments involving thousands of user interactions under diverse setups to broaden the understanding of learning from feedback over time. Our experiments show effective improvement from user feedback of extractive QA models over time across different data regimes, including significant potential for domain adaptation.

Ron Eliav, Anya Ji, Yoav Artzi et al.

A long tradition of studies in psycholinguistics has examined the formation and generalization of ad hoc conventions in reference games, showing how newly acquired conventions for a given target transfer to new referential contexts. However, another axis of generalization remains understudied: how do conventions formed for one target transfer to completely distinct targets, when specific lexical choices are unlikely to repeat? This paper presents two dyadic studies (N = 240) that address this axis of generalization, focusing on the role of nameability -- the a priori likelihood that two individuals will share the same label. We leverage the recently-released KiloGram dataset, a collection of abstract tangram images that is orders of magnitude larger than previously available, exhibiting high diversity of properties like nameability. Our first study asks how nameability shapes convention formation, while the second asks how new conventions generalize to entirely new targets of reference. Our results raise new questions about how ad hoc conventions extend beyond target-specific re-use of specific lexical choices.

Menglin Jia, Austin Reiter, Ser-Nam Lim et al.

We introduce Classification with Alternating Normalization (CAN), a non-parametric post-processing step for classification. CAN improves classification accuracy for challenging examples by re-adjusting their predicted class probability distribution using the predicted class distributions of high-confidence validation examples. CAN is easily applicable to any probabilistic classifier, with minimal computation overhead. We analyze the properties of CAN using simulated experiments, and empirically demonstrate its effectiveness across a diverse set of classification tasks.

Felix Wu, Kwangyoun Kim, Jing Pan et al.

This paper is a study of performance-efficiency trade-offs in pre-trained models for automatic speech recognition (ASR). We focus on wav2vec 2.0, and formalize several architecture designs that influence both the model performance and its efficiency. Putting together all our observations, we introduce SEW (Squeezed and Efficient Wav2vec), a pre-trained model architecture with significant improvements along both performance and efficiency dimensions across a variety of training setups. For example, under the 100h-960h semi-supervised setup on LibriSpeech, SEW achieves a 1.9x inference speedup compared to wav2vec 2.0, with a 13.5% relative reduction in word error rate. With a similar inference time, SEW reduces word error rate by 25-50% across different model sizes.

Anna Effenberger, Eva Yan, Rhia Singh et al.

We analyze language change over time in a collaborative, goal-oriented instructional task, where utility-maximizing participants form conventions and increase their expertise. Prior work studied such scenarios mostly in the context of reference games, and consistently found that language complexity is reduced along multiple dimensions, such as utterance length, as conventions are formed. In contrast, we find that, given the ability to increase instruction utility, instructors increase language complexity along these previously studied dimensions to better collaborate with increasingly skilled instruction followers.

Claire Yuqing Cui, Apoorv Khandelwal, Yoav Artzi et al.

We present a task and benchmark dataset for person-centric visual grounding, the problem of linking between people named in a caption and people pictured in an image. In contrast to prior work in visual grounding, which is predominantly object-based, our new task masks out the names of people in captions in order to encourage methods trained on such image-caption pairs to focus on contextual cues (such as rich interactions between multiple people), rather than learning associations between names and appearances. To facilitate this task, we introduce a new dataset, Who's Waldo, mined automatically from image-caption data on Wikimedia Commons. We propose a Transformer-based method that outperforms several strong baselines on this task, and are releasing our data to the research community to spur work on contextual models that consider both vision and language.

Noriyuki Kojima, Alane Suhr, Yoav Artzi

We study continual learning for natural language instruction generation, by observing human users' instruction execution. We focus on a collaborative scenario, where the system both acts and delegates tasks to human users using natural language. We compare user execution of generated instructions to the original system intent as an indication to the system's success communicating its intent. We show how to use this signal to improve the system's ability to generate instructions via contextual bandit learning. In interaction with real users, our system demonstrates dramatic improvements in its ability to generate language over time.

Valts Blukis, Chris Paxton, Dieter Fox et al.

Natural language provides an accessible and expressive interface to specify long-term tasks for robotic agents. However, non-experts are likely to specify such tasks with high-level instructions, which abstract over specific robot actions through several layers of abstraction. We propose that key to bridging this gap between language and robot actions over long execution horizons are persistent representations. We propose a persistent spatial semantic representation method, and show how it enables building an agent that performs hierarchical reasoning to effectively execute long-term tasks. We evaluate our approach on the ALFRED benchmark and achieve state-of-the-art results, despite completely avoiding the commonly used step-by-step instructions.

Valts Blukis, Ross A. Knepper, Yoav Artzi

We study the problem of learning a robot policy to follow natural language instructions that can be easily extended to reason about new objects. We introduce a few-shot language-conditioned object grounding method trained from augmented reality data that uses exemplars to identify objects and align them to their mentions in instructions. We present a learned map representation that encodes object locations and their instructed use, and construct it from our few-shot grounding output. We integrate this mapping approach into an instruction-following policy, thereby allowing it to reason about previously unseen objects at test-time by simply adding exemplars. We evaluate on the task of learning to map raw observations and instructions to continuous control of a physical quadcopter. Our approach significantly outperforms the prior state of the art in the presence of new objects, even when the prior approach observes all objects during training.

Tianyi Zhang, Felix Wu, Arzoo Katiyar et al.

This paper is a study of fine-tuning of BERT contextual representations, with focus on commonly observed instabilities in few-sample scenarios. We identify several factors that cause this instability: the common use of a non-standard optimization method with biased gradient estimation; the limited applicability of significant parts of the BERT network for down-stream tasks; and the prevalent practice of using a pre-determined, and small number of training iterations. We empirically test the impact of these factors, and identify alternative practices that resolve the commonly observed instability of the process. In light of these observations, we re-visit recently proposed methods to improve few-sample fine-tuning with BERT and re-evaluate their effectiveness. Generally, we observe the impact of these methods diminishes significantly with our modified process.

Noriyuki Kojima, Hadar Averbuch-Elor, Alexander M. Rush et al.

Visual features are a promising signal for learning bootstrap textual models. However, blackbox learning models make it difficult to isolate the specific contribution of visual components. In this analysis, we consider the case study of the Visually Grounded Neural Syntax Learner (Shi et al., 2019), a recent approach for learning syntax from a visual training signal. By constructing simplified versions of the model, we isolate the core factors that yield the model's strong performance. Contrary to what the model might be capable of learning, we find significantly less expressive versions produce similar predictions and perform just as well, or even better. We also find that a simple lexical signal of noun concreteness plays the main role in the model's predictions as opposed to more complex syntactic reasoning.

Matt Gardner, Yoav Artzi, Victoria Basmova et al.

Standard test sets for supervised learning evaluate in-distribution generalization. Unfortunately, when a dataset has systematic gaps (e.g., annotation artifacts), these evaluations are misleading: a model can learn simple decision rules that perform well on the test set but do not capture a dataset's intended capabilities. We propose a new annotation paradigm for NLP that helps to close systematic gaps in the test data. In particular, after a dataset is constructed, we recommend that the dataset authors manually perturb the test instances in small but meaningful ways that (typically) change the gold label, creating contrast sets. Contrast sets provide a local view of a model's decision boundary, which can be used to more accurately evaluate a model's true linguistic capabilities. We demonstrate the efficacy of contrast sets by creating them for 10 diverse NLP datasets (e.g., DROP reading comprehension, UD parsing, IMDb sentiment analysis). Although our contrast sets are not explicitly adversarial, model performance is significantly lower on them than on the original test sets---up to 25\% in some cases. We release our contrast sets as new evaluation benchmarks and encourage future dataset construction efforts to follow similar annotation processes.

Harsh Mehta, Yoav Artzi, Jason Baldridge et al.

The Touchdown dataset (Chen et al., 2019) provides instructions by human annotators for navigation through New York City streets and for resolving spatial descriptions at a given location. To enable the wider research community to work effectively with the Touchdown tasks, we are publicly releasing the 29k raw Street View panoramas needed for Touchdown. We follow the process used for the StreetLearn data release (Mirowski et al., 2019) to check panoramas for personally identifiable information and blur them as necessary. These have been added to the StreetLearn dataset and can be obtained via the same process as used previously for StreetLearn. We also provide a reference implementation for both of the Touchdown tasks: vision and language navigation (VLN) and spatial description resolution (SDR). We compare our model results to those given in Chen et al. (2019) and show that the panoramas we have added to StreetLearn fully support both Touchdown tasks and can be used effectively for further research and comparison.

Lili Yu, Howard Chen, Sida Wang et al.

We study the potential for interaction in natural language classification. We add a limited form of interaction for intent classification, where users provide an initial query using natural language, and the system asks for additional information using binary or multi-choice questions. At each turn, our system decides between asking the most informative question or making the final classification prediction.The simplicity of the model allows for bootstrapping of the system without interaction data, instead relying on simple crowdsourcing tasks. We evaluate our approach on two domains, showing the benefit of interaction and the advantage of learning to balance between asking additional questions and making the final prediction.

Valts Blukis, Yannick Terme, Eyvind Niklasson et al.

We propose a joint simulation and real-world learning framework for mapping navigation instructions and raw first-person observations to continuous control. Our model estimates the need for environment exploration, predicts the likelihood of visiting environment positions during execution, and controls the agent to both explore and visit high-likelihood positions. We introduce Supervised Reinforcement Asynchronous Learning (SuReAL). Learning uses both simulation and real environments without requiring autonomous flight in the physical environment during training, and combines supervised learning for predicting positions to visit and reinforcement learning for continuous control. We evaluate our approach on a natural language instruction-following task with a physical quadcopter, and demonstrate effective execution and exploration behavior.

Alane Suhr, Claudia Yan, Charlotte Schluger et al.

We study a collaborative scenario where a user not only instructs a system to complete tasks, but also acts alongside it. This allows the user to adapt to the system abilities by changing their language or deciding to simply accomplish some tasks themselves, and requires the system to effectively recover from errors as the user strategically assigns it new goals. We build a game environment to study this scenario, and learn to map user instructions to system actions. We introduce a learning approach focused on recovery from cascading errors between instructions, and modeling methods to explicitly reason about instructions with multiple goals. We evaluate with a new evaluation protocol using recorded interactions and online games with human users, and observe how users adapt to the system abilities.

Alane Suhr, Yoav Artzi

NLVR2 (Suhr et al., 2019) was designed to be robust for language bias through a data collection process that resulted in each natural language sentence appearing with both true and false labels. The process did not provide a similar measure of control for visual bias. This technical report analyzes the potential for visual bias in NLVR2. We show that some amount of visual bias likely exists. Finally, we identify a subset of the test data that allows to test for model performance in a way that is robust to such potential biases. We show that the performance of existing models (Li et al., 2019; Tan and Bansal 2019) is relatively robust to this potential bias. We propose to add the evaluation on this subset of the data to the NLVR2 evaluation protocol, and update the official release to include it. A notebook including an implementation of the code used to replicate this analysis is available at http://nlvr.ai/NLVR2BiasAnalysis.html.

Tianyi Zhang, Varsha Kishore, Felix Wu et al.

We propose BERTScore, an automatic evaluation metric for text generation. Analogously to common metrics, BERTScore computes a similarity score for each token in the candidate sentence with each token in the reference sentence. However, instead of exact matches, we compute token similarity using contextual embeddings. We evaluate using the outputs of 363 machine translation and image captioning systems. BERTScore correlates better with human judgments and provides stronger model selection performance than existing metrics. Finally, we use an adversarial paraphrase detection task to show that BERTScore is more robust to challenging examples when compared to existing metrics.

Howard Chen, Alane Suhr, Dipendra Misra et al.

We study the problem of jointly reasoning about language and vision through a navigation and spatial reasoning task. We introduce the Touchdown task and dataset, where an agent must first follow navigation instructions in a real-life visual urban environment, and then identify a location described in natural language to find a hidden object at the goal position. The data contains 9,326 examples of English instructions and spatial descriptions paired with demonstrations. Empirical analysis shows the data presents an open challenge to existing methods, and qualitative linguistic analysis shows that the data displays richer use of spatial reasoning compared to related resources.

Aaron Walsman, Yonatan Bisk, Saadia Gabriel et al.

Building perceptual systems for robotics which perform well under tight computational budgets requires novel architectures which rethink the traditional computer vision pipeline. Modern vision architectures require the agent to build a summary representation of the entire scene, even if most of the input is irrelevant to the agent's current goal. In this work, we flip this paradigm, by introducing EarlyFusion vision models that condition on a goal to build custom representations for downstream tasks. We show that these goal specific representations can be learned more quickly, are substantially more parameter efficient, and more robust than existing attention mechanisms in our domain. We demonstrate the effectiveness of these methods on a simulated robotic item retrieval problem that is trained in a fully end-to-end manner via imitation learning.

Valts Blukis, Dipendra Misra, Ross A. Knepper et al.

We propose an approach for mapping natural language instructions and raw observations to continuous control of a quadcopter drone. Our model predicts interpretable position-visitation distributions indicating where the agent should go during execution and where it should stop, and uses the predicted distributions to select the actions to execute. This two-step model decomposition allows for simple and efficient training using a combination of supervised learning and imitation learning. We evaluate our approach with a realistic drone simulator, and demonstrate absolute task-completion accuracy improvements of 16.85% over two state-of-the-art instruction-following methods.

Alane Suhr, Stephanie Zhou, Ally Zhang et al.

We introduce a new dataset for joint reasoning about natural language and images, with a focus on semantic diversity, compositionality, and visual reasoning challenges. The data contains 107,292 examples of English sentences paired with web photographs. The task is to determine whether a natural language caption is true about a pair of photographs. We crowdsource the data using sets of visually rich images and a compare-and-contrast task to elicit linguistically diverse language. Qualitative analysis shows the data requires compositional joint reasoning, including about quantities, comparisons, and relations. Evaluation using state-of-the-art visual reasoning methods shows the data presents a strong challenge.

Dipendra Misra, Andrew Bennett, Valts Blukis et al.

We propose to decompose instruction execution to goal prediction and action generation. We design a model that maps raw visual observations to goals using LINGUNET, a language-conditioned image generation network, and then generates the actions required to complete them. Our model is trained from demonstration only without external resources. To evaluate our approach, we introduce two benchmarks for instruction following: LANI, a navigation task; and CHAI, where an agent executes household instructions. Our evaluation demonstrates the advantages of our model decomposition, and illustrates the challenges posed by our new benchmarks.

Valts Blukis, Nataly Brukhim, Andrew Bennett et al.

We introduce a method for following high-level navigation instructions by mapping directly from images, instructions and pose estimates to continuous low-level velocity commands for real-time control. The Grounded Semantic Mapping Network (GSMN) is a fully-differentiable neural network architecture that builds an explicit semantic map in the world reference frame by incorporating a pinhole camera projection model within the network. The information stored in the map is learned from experience, while the local-to-world transformation is computed explicitly. We train the model using DAggerFM, a modified variant of DAgger that trades tabular convergence guarantees for improved training speed and memory use. We test GSMN in virtual environments on a realistic quadcopter simulator and show that incorporating an explicit mapping and grounding modules allows GSMN to outperform strong neural baselines and almost reach an expert policy performance. Finally, we analyze the learned map representations and show that using an explicit map leads to an interpretable instruction-following model.

Alane Suhr, Yoav Artzi

We propose a learning approach for mapping context-dependent sequential instructions to actions. We address the problem of discourse and state dependencies with an attention-based model that considers both the history of the interaction and the state of the world. To train from start and goal states without access to demonstrations, we propose SESTRA, a learning algorithm that takes advantage of single-step reward observations and immediate expected reward maximization. We evaluate on the SCONE domains, and show absolute accuracy improvements of 9.8%-25.3% across the domains over approaches that use high-level logical representations.

Max Grusky, Mor Naaman, Yoav Artzi

We present NEWSROOM, a summarization dataset of 1.3 million articles and summaries written by authors and editors in newsrooms of 38 major news publications. Extracted from search and social media metadata between 1998 and 2017, these high-quality summaries demonstrate high diversity of summarization styles. In particular, the summaries combine abstractive and extractive strategies, borrowing words and phrases from articles at varying rates. We analyze the extraction strategies used in NEWSROOM summaries against other datasets to quantify the diversity and difficulty of our new data, and train existing methods on the data to evaluate its utility and challenges.