Abstract 3D scene synthesis using natural language instructions has become a popular direction in computer graphics, with significant progress made by data‐driven generative models recently. However, previous methods have mainly focused on one‐time scene generation, lacking the interactive capability to generate, update, or correct scenes according to user instructions. To overcome this limitation, this paper focuses on text‐guided interactive scene synthesis. First, we introduce the SceneMod dataset, which comprises 168k paired scenes with textual descriptions of the modifications. To support the interactive scene synthesis task, we propose a two‐stage diffusion generative model that integrates scene‐prior guidance into the denoising process to explicitly enforce physical constraints and foster more realistic scenes. Experimental results demonstrate that our approach outperforms baseline methods in text‐guided scene synthesis tasks. Our system expands the scope of data‐driven scene synthesis tasks and provides a novel, more flexible tool for users and designers in 3D scene generation. Code and dataset are available at https://github.com/bshfang/SceneMod .

In this paper, we examine how large language models (LLMs) solve multi-step problems under a language agent framework with three components: a generator, a discriminator, and a planning method.We investigate the practical utility of two advanced planning methods, iterative correction and tree search.We present a comprehensive analysis of how discrimination accuracy affects the overall performance of agents when using these two methods or a simpler method, re-ranking.Experiments on two tasks, text-to-SQL parsing and mathematical reasoning, show that: (1) advanced planning methods demand discriminators with at least 90% accuracy to achieve significant improvements over re-ranking; (2) current LLMs' discrimination abilities have not met the needs of advanced planning methods to achieve such improvements; (3) with LLM-based discriminators, advanced planning methods may not adequately balance accuracy and efficiency.For example, compared to the other two methods, tree search is at least 10-20 times slower but leads to negligible performance gains, which hinders its real-world applications.1

We introduce Semantic Parsing in Contextual Environments (SPICE), a task designed to enhance artificial agents' contextual awareness by integrating multimodal inputs with prior contexts.SPICE goes beyond traditional semantic parsing by offering a structured, interpretable framework for dynamically updating an agent's knowledge with new information, mirroring the complexity of human communication.We develop the VG-SPICE dataset, crafted to challenge agents with visual scene graph construction from spoken conversational exchanges, highlighting speech and visual data integration.We also present the Audio-Vision Dialogue Scene Parser (AViD-SP) developed for use on VG-SPICE.These innovations aim to improve multimodal information processing and integration.Both the VG-SPICE dataset and the AViD-SP model are publicly available.

Understanding the abilities of LLMs to reason about natural language plans, such as instructional text and recipes, is critical to reliably using them in decision-making systems.A fundamental aspect of plans is the temporal order in which their steps need to be executed, which reflects the underlying causal dependencies between them.We introduce CAT-BENCH, a benchmark of Step Order Prediction questions, which test whether a step must necessarily occur before or after another in cooking recipe plans.We use this to evaluate how well frontier LLMs understand causal and temporal dependencies.We find that SOTA LLMs are underwhelming (best zero-shot is only 0.59 in F1), and are biased towards predicting dependence more often, perhaps relying on temporal order of steps as a heuristic.While prompting for explanations and using few-shot examples improve performance, the best F1 result is only 0.73.Further, human evaluation of explanations along with answer correctness show that, on average, humans do not agree with model reasoning.Surprisingly, we also find that explaining after answering leads to better performance than normal chain-of-thought prompting, and LLM answers are not consistent across questions about the same step pairs.Overall, results show that LLMs' ability to detect dependence between steps has significant room for improvement. * Equal ContributionAlmond Flour Chocolate Cake … Step 6: Stir in ground almonds.Step 7: Add half flour and half milk.Step 8: Use wooden spoon to stir.… Step 12: Whip cream till stiff peaks … Q: Must Step 6 happen before Step 8? Questions about dependent steps Q: Must Step 7 happen after Step 6? Questions about non-dependent steps A: Yes, all ingredients have to be in bowl before stirring A: No, almonds can be added after flour and milk Parallel Steps Preconditions CAT-Bench Dependent Steps

Extracting knowledge and reasoning from large language models (LLMs) offers a path to designing intelligent robots. Common approaches that leverage LLMs for planning are unable to recover when actions fail and resort to retrying failed actions without resolving the underlying cause. We propose a novel approach (CAPE) that generates corrective actions to resolve precondition errors during planning. CAPE improves the quality of generated plans through few-shot reasoning on action preconditions. Our approach enables embodied agents to execute more tasks than baseline methods while maintaining semantic correctness and minimizing re-prompting. In VirtualHome, CAPE improves a human-annotated plan correctness metric from 28.89% to 49.63% over SayCan, whilst achieving competitive executability. Our improvements transfer to a Boston Dynamics Spot robot initialized with a set of skills (specified in language) and associated preconditions, where CAPE improves correctness by 76.49% with higher executability compared to SayCan. Our approach enables embodied agents to follow natural language commands and robustly recover from failures.

Audio-visual pre-trained models have gained substantial attention recently and demonstrated superior performance on various audio-visual tasks. This study investigates whether pre-trained audio-visual models demonstrate non-arbitrary associations between sounds and visual representations-known as sound symbolism-which is also observed in humans. We developed a specialized dataset with synthesized images and audio samples and assessed these models using a non-parametric approach in a zero-shot setting. Our findings reveal a significant correlation between the models’ outputs and established patterns of sound symbolism, particularly in models trained on speech data. These results suggest that such models can capture sound-meaning connections akin to human language processing, providing insights into both cognitive architectures and machine learning strategies.

Writing tests is a time-consuming yet essential task during software development. We propose to leverage recent advances in deep learning for text and code generation to assist developers in writing tests. We formalize the novel task of test completion to automatically complete the next statement in a test method based on the context of prior statements and the code under test. We develop TECo-a deep learning model using code semantics for test completion. The key insight underlying TECO is that predicting the next statement in a test method requires reasoning about code execution, which is hard to do with only syntax-level data that existing code completion models use. Teco extracts and uses six kinds of code semantics data, including the execution result of prior statements and the execution context of the test method. To provide a testbed for this new task, as well as to evaluate TECO, we collect a corpus of 130,934 test methods from 1,270 open-source Java projects. Our results show that Teco achieves an exact-match accuracy of 18, which is 29% higher than the best baseline using syntax-level data only. When measuring functional correctness of generated next statement, Teco can generate runnable code in 29% of the cases compared to 18% obtained by the best baseline. Moreover, Teco is sianificantly better than prior work on test oracle generation.

There is growing interest in generating skeleton-based human motions from natural language descriptions. While most efforts have focused on developing better neural architectures for this task, there has been no significant work on determining the proper evaluation metric. Human evaluation is the ultimate accuracy measure for this task, and automated metrics should correlate well with human quality judgments. Since descriptions are compatible with many motions, determining the right metric is critical for evaluating and designing effective generative models. This paper systematically studies which metrics best align with human evaluations and proposes new metrics that align even better. Our findings indicate that none of the metrics currently used for this task show even a moderate correlation with human judgments on a sample level. However, for assessing average model performance, commonly used metrics such as R-Precision and less-used coordinate errors show strong correlations. Additionally, several recently developed metrics are not recommended due to their low correlation compared to alternatives. We also introduce a novel metric based on a multimodal BERT-like model, MoBERT, which offers strongly human-correlated sample-level evaluations while maintaining near-perfect model-level correlation. Our results demonstrate that this new metric exhibits extensive benefits over all current alternatives.

There has been a growing interest in developing machine learning (ML) models for code summarization tasks, e.g., comment generation and method naming.Despite substantial increase in the effectiveness of ML models, the evaluation methodologies, i.e., the way people split datasets into training, validation, and test sets, were not well studied.Specifically, no prior work on code summarization considered the timestamps of code and comments during evaluation.This may lead to evaluations that are inconsistent with the intended use cases.In this paper, we introduce the time-segmented evaluation methodology, which is novel to the code summarization research community, and compare it with the mixed-project and cross-project methodologies that have been commonly used.Each methodology can be mapped to some use cases, and the time-segmented methodology should be adopted in the evaluation of ML models for code summarization.To assess the impact of methodologies, we collect a dataset of (code, comment) pairs with timestamps to train and evaluate several recent ML models for code summarization.Our experiments show that different methodologies lead to conflicting evaluation results.We invite the community to expand the set of methodologies used in evaluations.

Yash Kumar Lal, Niket Tandon, Tanvi Aggarwal, Horace Liu, Nathanael Chambers, Raymond Mooney, Niranjan Balasubramanian. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

Most Outside-Knowledge Visual Question Answering (OK-VQA) systems employ a twostage framework that first retrieves external knowledge given the visual question and then predicts the answer based on the retrieved content.However, the retrieved knowledge is often inadequate.Retrievals are frequently too general and fail to cover specific knowledge needed to answer the question.Also, the naturally available supervision (whether the passage contains the correct answer) is weak and does not guarantee question relevancy.To address these issues, we propose an Entity-Focused Retrieval (EnFoRe) model that provides stronger supervision during training and recognizes questionrelevant entities to help retrieve more specific knowledge.Experiments show that our En-FoRe model achieves superior retrieval performance on OK-VQA, the currently largest outside-knowledge VQA dataset.We also combine the retrieved knowledge with state-of-theart VQA models, and achieve a new state-ofthe-art performance on OK-VQA.Q: What holiday is this?A: Thanksgiving.

With robotics rapidly advancing, more effective human–robot interaction is increasingly needed to realize the full potential of robots for society. While spoken language must be part of the solution, our ability to provide spoken language interaction capabilities is still very limited. In this article, based on the report of an interdisciplinary workshop convened by the National Science Foundation, we identify key scientific and engineering advances needed to enable effective spoken language interaction with robotics. We make 25 recommendations, involving eight general themes: putting human needs first, better modeling the social and interactive aspects of language, improving robustness, creating new methods for rapid adaptation, better integrating speech and language with other communication modalities, giving speech and language components access to rich representations of the robot’s current knowledge and state, making all components operate in real time, and improving research infrastructure and resources. Research and development that prioritizes these topics will, we believe, provide a solid foundation for the creation of speech-capable robots that are easy and effective for humans to work with.

Intelligent systems need to be able to recover from mistakes, resolve uncertainty, and adapt to novel concepts not seen during training. Dialog interaction can enable this by the use of clarifications for correction and resolving uncertainty, and active learning queries to learn new concepts encountered during operation. Prior work on dialog systems has either focused on exclusively learning how to perform clarification/ information seeking, or to perform active learning. In this work, we train a hierarchical dialog policy to jointly perform {\it both} clarification and active learning in the context of an interactive language-based image retrieval task motivated by an online shopping application, and demonstrate that jointly learning dialog policies for clarification and active learning is more effective than the use of static dialog policies for one or both of these functions.

Natural language comments convey key aspects of source code such as implementation, usage, and pre- and post-conditions. Failure to update comments accordingly when the corresponding code is modified introduces inconsistencies, which is known to lead to confusion and software bugs. In this paper, we aim to detect whether a comment becomes inconsistent as a result of changes to the corresponding body of code, in order to catch potential inconsistencies just-in-time, i.e., before they are committed to a code base. To achieve this, we develop a deep-learning approach that learns to correlate a comment with code changes. By evaluating on a large corpus of comment/code pairs spanning various comment types, we show that our model outperforms multiple baselines by significant margins. For extrinsic evaluation, we show the usefulness of our approach by combining it with a comment update model to build a more comprehensive automatic comment maintenance system which can both detect and resolve inconsistent comments based on code changes.

Comments are an integral part of software development; they are natural language descriptions associated with source code elements. Understanding explicit associations can be useful in improving code comprehensibility and maintaining the consistency between code and comments. As an initial step towards this larger goal, we address the task of associating entities in Javadoc comments with elements in Java source code. We propose an approach for automatically extracting supervised data using revision histories of open source projects and present a manually annotated evaluation dataset for this task. We develop a binary classifier and a sequence labeling model by crafting a rich feature set which encompasses various aspects of code, comments, and the relationships between them. Experiments show that our systems outperform several baselines learning from the proposed supervision.

We formulate the novel task of automatically updating an existing natural language comment based on changes in the body of code it accompanies.We propose an approach that learns to correlate changes across two distinct language representations, to generate a sequence of edits that are applied to the existing comment to reflect the source code modifications.We train and evaluate our model using a dataset that we collected from commit histories of open-source software projects, with each example consisting of a concurrent update to a method and its corresponding comment.We compare our approach against multiple baselines using both automatic metrics and human evaluation.Results reflect the challenge of this task and that our model outperforms baselines with respect to making edits.

In this work, we present methods for using human-robot dialog to improve language understanding for a mobile robot agent. The agent parses natural language to underlying semantic meanings and uses robotic sensors to create multi-modal models of perceptual concepts like red and heavy. The agent can be used for showing navigation routes, delivering objects to people, and relocating objects from one location to another. We use dialog clari_cation questions both to understand commands and to generate additional parsing training data. The agent employs opportunistic active learning to select questions about how words relate to objects, improving its understanding of perceptual concepts. We evaluated this agent on Amazon Mechanical Turk. After training on data induced from conversations, the agent reduced the number of dialog questions it asked while receiving higher usability ratings. Additionally, we demonstrated the agent on a robotic platform, where it learned new perceptual concepts on the y while completing a real-world task.

Visual question answering (VQA) and image captioning require a shared body of general knowledge connecting language and vision.We present a novel approach to improve VQA performance that exploits this connection by jointly generating captions that are targeted to help answer a specific visual question.The model is trained using an existing caption dataset by automatically determining question-relevant captions using an online gradient-based method.Experimental results on the VQA v2 challenge demonstrates that our approach obtains state-of-the-art VQA performance (e.g.68.4% on the Test-standard set using a single model) by simultaneously generating question-relevant captions.

Natural language understanding for robotics can require substantial domain- and platform-specific engineering. For example, for mobile robots to pick-and-place objects in an environment to satisfy human commands, we can specify the language humans use to issue such commands, and connect concept words like red can to physical object properties. One way to alleviate this engineering for a new domain is to enable robots in human environments to adapt dynamically-continually learning new language constructions and perceptual concepts. In this work, we present an end-to-end pipeline for translating natural language commands to discrete robot actions, and use clarification dialogs to jointly improve language parsing and concept grounding. We train and evaluate this agent in a virtual setting on Amazon Mechanical Turk, and we transfer the learned agent to a physical robot platform to demonstrate it in the real world.

Recent reinforcement learning (RL) approaches have shown strong performance in complex domains, such as Atari games, but are highly sample inefficient. A common approach to reduce interaction time with the environment is to use reward shaping, which involves carefully designing reward functions that provide the agent intermediate rewards for progress towards the goal. Designing such rewards remains a challenge, though. In this work, we use natural language instructions to perform reward shaping. We propose a framework that maps free-form natural language instructions to intermediate rewards, that can seamlessly be integrated into any standard reinforcement learning algorithm. We experiment with Montezuma's Revenge from the Atari video games domain, a popular benchmark in RL. Our experiments on a diverse set of 15 tasks demonstrate that for the same number of interactions with the environment, using language-based rewards can successfully complete the task 60% more often, averaged across all tasks, compared to learning without language.

Visual Question Answering (VQA) deep-learning systems tend to capture superficial statistical correlations in the training data because of strong language priors and fail to generalize to test data with a significantly different question-answer (QA) distribution. To address this issue, we introduce a self-critical training objective that ensures that visual explanations of correct answers match the most influential image regions more than other competitive answer candidates. The influential regions are either determined from human visual/textual explanations or automatically from just significant words in the question and answer. We evaluate our approach on the VQA generalization task using the VQA-CP dataset, achieving a new state-of-the-art i.e. 49.5\% using textual explanations and 48.5\% using automatically

Natural language elements, e.g., todo comments, are frequently used to communicate among developers and to describe tasks that need to be performed (actions) when specific conditions hold on artifacts related to the code repository (triggers), e.g., from the Apache Struts project: “remove expectedJDK15 and if() after switching to Java 1.6”. As projects evolve, development processes change, and development teams reorganize, these comments, because of their informal nature, frequently become irrelevant or forgotten. We present the first framework, dubbed TrigIt, to specify trigger-action todo comments in executable format. Thus, actions are executed automatically when triggers evaluate to true. TrigIt specifications are written in the host language (e.g., Java) and are evaluated as part of the build process. The triggers are specified as query statements over abstract syntax trees, abstract representation of build configuration scripts, issue tracking systems, and system clock time. The actions are either notifications to developers or code transformation steps. We implemented TrigIt for the Java programming language and migrated 44 existing trigger-action comments from several popular open-source projects. Evaluation of TrigIt, via a user study, showed that users find TrigIt easy to learn and use. TrigIt has the potential to enforce more discipline in writing and maintaining comments in large code repositories.

A major goal of grounded language learning research is to enable robots to connect language predicates to a robot's physical interactive perception of the world. Coupling object exploratory behaviors such as grasping, lifting, and looking with multiple sensory modalities (e.g., audio, haptics, and vision) enables a robot to ground non-visual words like ``heavy'' as well as visual words like ``red''. A major limitation of existing approaches to multi-modal language grounding is that a robot has to exhaustively explore training objects with a variety of actions when learning a new such language predicate. This paper proposes a method for guiding a robot's behavioral exploration policy when learning a novel predicate based on known grounded predicates and the novel predicate's linguistic relationship to them. We demonstrate our approach on two datasets in which a robot explored large sets of objects and was tasked with learning to recognize whether novel words applied to those objects.

Active learning identifies data points to label that are expected to be the most useful in improving a supervised model.Opportunistic active learning incorporates active learning into interactive tasks that constrain possible queries during interactions.Prior work has shown that opportunistic active learning can be used to improve grounding of natural language descriptions in an interactive object retrieval task.In this work, we use reinforcement learning for such an object retrieval task, to learn a policy that effectively trades off task completion with model improvement that would benefit future tasks.

Nazneen Fatema Rajani, Raymond Mooney. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018.

Recent captioning models are limited in their ability to scale and describe concepts unseen in paired image-text corpora. We propose the Novel Object Captioner (NOC), a deep visual semantic captioning model that can describe a large number of object categories not present in existing image-caption datasets. Our model takes advantage of external sources - labeled images from object recognition datasets, and semantic knowledge extracted from unannotated text. We propose minimizing a joint objective which can learn from these diverse data sources and leverage distributional semantic embeddings, enabling the model to generalize and describe novel objects outside of image-caption datasets. We demonstrate that our model exploits semantic information to generate captions for hundreds of object categories in the ImageNet object recognition dataset that are not observed in MSCOCO image-caption training data, as well as many categories that are observed very rarely. Both automatic evaluations and human judgements show that our model considerably outperforms prior work in being able to describe many more categories of objects.

Natural language understanding and dialog management are two integral components of interactive dialog systems.Previous research has used machine learning techniques to individually optimize these components, with different forms of direct and indirect supervision.We present an approach to integrate the learning of both a dialog strategy using reinforcement learning, and a semantic parser for robust natural language understanding, using only natural dialog interaction for supervision.Experimental results on a simulated task of robot instruction demonstrate that joint learning of both components improves dialog performance over learning either of these components alone.

Ensembling methods are well known for improving prediction accuracy. However, they are limited in the sense that they cannot effectively discriminate among component models. In this paper, we propose stacking with auxiliary features that learns to fuse additional relevant information from multiple component systems as well as input instances to improve performance. We use two types of auxiliary features -- instance features and provenance features. The instance features enable the stacker to discriminate across input instances and the provenance features enable the stacker to discriminate across component systems. When combined together, our algorithm learns to rely on systems that not just agree on an output but also the provenance of this output in conjunction with the properties of the input instance. We demonstrate the success of our approach on three very different and challenging natural language and vision problems: Slot Filling, Entity Discovery and Linking, and ImageNet Object Detection. We obtain new state-of-the-art results on the first two tasks and significant improvements on the ImageNet task, thus verifying the power and generality of our approach.

A word in natural language can be polysemous, having multiple meanings, as well as synonymous, meaning the same thing as other words. Word sense induction attempts to find the senses of polysemous words. Synonymy detection attempts to find when two words are interchangeable. We combine these tasks, first inducing word senses and then detecting similar senses to form word-sense synonym sets (synsets) in an unsupervised fashion. Given pairs of images and text with noun phrase labels, we perform synset induction to produce collections of underlying concepts described by one or more noun phrases. We find that considering multi-modal features from both visual and textual context yields better induced synsets than using either context alone. Human evaluations show that our unsupervised, multi-modally induced synsets are comparable in quality to annotation-assisted ImageNet synsets, achieving about 84% of ImageNet synsets' approval.

We explore techniques to maximize the effectiveness of discourse information in the task of authorship attribution. We present a novel method to embed discourse features in a Convolutional Neural Network text classifier, which achieves a state-of-the-art result by a significant margin. We empirically investigate several featurization methods to understand the conditions under which discourse features contribute non-trivial performance gains, and analyze discourse embeddings.

Scripts encode knowledge of prototypical sequences of events. We describe a Recurrent Neural Network model for statistical script learning using Long Short-Term Memory, an architecture which has been demonstrated to work well on a range of Artificial Intelligence tasks. We evaluate our system on two tasks, inferring held-out events from text and inferring novel events from text, substantially outperforming prior approaches on both tasks.

There is a small but growing body of research on statistical scripts, models of event sequences that allow probabilistic inference of implicit events from documents.These systems operate on structured verb-argument events produced by an NLP pipeline.We compare these systems with recent Recurrent Neural Net models that directly operate on raw tokens to predict sentences, finding the latter to be roughly comparable to the former in terms of predicting missing events in documents.

While recent deep neural network models have achieved promising results on the image captioning task, they rely largely on the availability of corpora with paired image and sentence captions to describe objects in context. In this work, we propose the Deep Compositional Captioner (DCC) to address the task of generating descriptions of novel objects which are not present in paired imagesentence datasets. Our method achieves this by leveraging large object recognition datasets and external text corpora and by transferring knowledge between semantically similar concepts. Current deep caption models can only describe objects contained in paired image-sentence corpora, despite the fact that they are pre-trained with large object recognition datasets, namely ImageNet. In contrast, our model can compose sentences that describe novel objects and their interactions with other objects. We demonstrate our model's ability to describe novel concepts by empirically evaluating its performance on MSCOCO and show qualitative results on ImageNet images of objects for which no paired image-sentence data exist. Further, we extend our approach to generate descriptions of objects in video clips. Our results show that DCC has distinct advantages over existing image and video captioning approaches for generating descriptions of new objects in context.

We propose an algorithm that combines supervised and unsupervised methods to ensemble multiple systems for two popular Knowledge Base Population (KBP) tasks, Cold Start Slot Filling (CSSF) and Tri-lingual Entity Discovery and Linking (TEDL).We demonstrate that it outperforms the best system for both tasks in the 2015 competition, several ensembling baselines, as well as a state-of-the-art stacking approach.The success of our technique on two different and challenging problems demonstrates the power and generality of our combined approach to ensembling.

This paper investigates how linguistic knowledge mined from large text corpora can aid the generation of natural language descriptions of videos.Specifically, we integrate both a neural language model and distributional semantics trained on large text corpora into a recent LSTM-based architecture for video description.We evaluate our approach on a collection of Youtube videos as well as two large movie description datasets showing significant improvements in grammaticality while modestly improving descriptive quality.

NLP tasks differ in the semantic information they require, and at this time no single semantic representation fulfills all requirements. Logic-based representations characterize sentence structure, but do not capture the graded aspect of meaning. Distributional models give graded similarity ratings for words and phrases, but do not capture sentence structure in the same detail as logic-based approaches. It has therefore been argued that the two are complementary. We adopt a hybrid approach that combines logical and distributional semantics using probabilistic logic, specifically Markov Logic Networks. In this article, we focus on the three components of a practical system:11) Logical representation focuses on representing the input problems in probabilistic logic; 2) knowledge base construction creates weighted inference rules by integrating distributional information with other sources; and 3) probabilistic inference involves solving the resulting MLN inference problems efficiently. To evaluate our approach, we use the task of textual entailment, which can utilize the strengths of both logic-based and distributional representations. In particular we focus on the SICK data set, where we achieve state-of-the-art results. We also release a lexical entailment data set of 10,213 rules extracted from the SICK data set, which is a valuable resource for evaluating lexical entailment systems.2

Chris Quirk, Raymond Mooney, Michel Galley. 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.

Vidhoon Viswanathan, Nazneen Fatema Rajani, Yinon Bentor, Raymond Mooney. 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.

Real-world videos often have complex dynamics, methods for generating open-domain video descriptions should be sensitive to temporal structure and allow both input (sequence of frames) and output (sequence of words) of variable length. To approach this problem we propose a novel end-to-end sequence-to-sequence model to generate captions for videos. For this we exploit recurrent neural networks, specifically LSTMs, which have demonstrated state-of-the-art performance in image caption generation. Our LSTM model is trained on video-sentence pairs and learns to associate a sequence of video frames to a sequence of words in order to generate a description of the event in the video clip. Our model naturally is able to learn the temporal structure of the sequence of frames as well as the sequence model of the generated sentences, i.e. a language model. We evaluate several variants of our model that exploit different visual features on a standard set of YouTube videos and two movie description datasets (M-VAD and MPII-MD).

Subhashini Venugopalan, Huijuan Xu, Jeff Donahue, Marcus Rohrbach, Raymond Mooney, Kate Saenko. Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2015.

Probabilistic Soft Logic (PSL) is a re-cently developed framework for proba-bilistic logic. We use PSL to combine logical and distributional representations of natural-language meaning, where distri-butional information is represented in the form of weighted inference rules. We ap-ply this framework to the task of Seman-tic Textual Similarity (STS) (i.e. judg-ing the semantic similarity of natural-language sentences), and show that PSL gives improved results compared to a pre-vious approach based on Markov Logic Networks (MLNs) and a purely distribu-tional approach. 1

Scripts represent knowledge of stereotypical event sequences that can aid text understanding.Initial statistical methods have been developed to learn probabilistic scripts from raw text corpora; however, they utilize a very impoverished representation of events, consisting of a verb and one dependent argument.We present a script learning approach that employs events with multiple arguments.Unlike previous work, we model the interactions between multiple entities in a script.Experiments on a large corpus using the task of inferring held-out events (the "narrative cloze evaluation") demonstrate that modeling multi-argument events improves predictive accuracy.

Multitask learning (MTL) via a shared representation has been adopted to alleviate problems with sparsity of labeled data across different learning tasks. Active learning, on the other hand, reduces the cost of labeling examples by making informative queries over an unlabeled pool of data. Therefore, a unification of both of these approaches can potentially be useful in settings where labeled information is expensive to obtain but the learning tasks or domains have some common characteristics. This paper introduces two such models – Active Doubly Supervised Latent Dirichlet Allocation (Act-DSLDA) and its non-parametric variation (Act-NPDSLDA) that integrate MTL and active learning in the same framework. These models make use of both latent and supervised shared topics to accomplish multitask learning. Experimental results on both document and image classification show that integrating MTL and active learning along with shared latent and supervised topics is superior to other methods which do not employ all of these components.

We present a holistic data-driven technique that generates natural-language descriptions for videos. We combine the output of state-of-the-art object and activity detectors with "real-world' knowledge to select the most probable subject-verb-object triplet for describing a video. We show that this knowledge, automatically mined from web-scale text corpora, enhances the triplet selection algorithm by providing it contextual information and leads to a four-fold increase in activity identification. Unlike previous methods, our approach can annotate arbitrary videos without requiring the expensive collection and annotation of a similar training video corpus. We evaluate our technique against a baseline that does not use text-mined knowledge and show that humans prefer our descriptions 61% of the time.

This paper presents an approach for detecting promotional content in Wikipedia.By incorporating stylometric features, including features based on n-gram and PCFG language models, we demonstrate improved accuracy at identifying promotional articles, compared to using only lexical information and metafeatures.

Despite a recent push towards large-scale object recognition, activity recognition remains limited to narrow domains and small vocabularies of actions. In this paper, we tackle the challenge of recognizing and describing activities ``in-the-wild''. We present a solution that takes a short video clip and outputs a brief sentence that sums up the main activity in the video, such as the actor, the action and its object. Unlike previous work, our approach works on out-of-domain actions: it does not require training videos of the exact activity. If it cannot find an accurate prediction for a pre-trained model, it finds a less specific answer that is also plausible from a pragmatic standpoint. We use semantic hierarchies learned from the data to help to choose an appropriate level of generalization, and priors learned from Web-scale natural language corpora to penalize unlikely combinations of actors/actions/objects, we also use a Web-scale language model to ``fill in'' novel verbs, i.e. when the verb does not appear in the training set. We evaluate our method on a large YouTube corpus and demonstrate it is able to generate short sentence descriptions of video clips better than baseline approaches.

The ability to understand natural-language instructions is critical to building intelligent agents that interact with humans. We present a system that learns to transform natural-language navigation instructions into executable formal plans. Given no prior linguistic knowledge, the system learns by simply observing how humans follow navigation instructions. The system is evaluated in three complex virtual indoor environments with numerous objects and landmarks. A previously collected realistic corpus of complex English navigation instructions for these environments is used for training and testing data. By using a learned lexicon to refine inferred plans and a supervised learner to induce a semantic parser, the system is able to automatically learnto correctly interpret a reasonable fraction of the complex instructions in this corpus.

Plan recognition is a form of abductive reasoning that involves inferring plans that best explain sets of observed actions. Most existing approaches to plan recognition and other abductive tasks employ either purely logical methods that donot handle uncertainty, or purely probabilistic methods thatdo not handle structured representations. To overcome these limitations, this paper introduces an approach to abductive reasoning using a first-order probabilistic logic, specifically Markov Logic Networks (MLNs). It introduces several novel techniques for making MLNs efficient and effective for abduction. Experiments on three plan recognition datasets showthe benefit of our approach over existing methods.

Most of the existing weight-learning algorithms for Markov Logic Networks (MLNs) use batch training which becomes computationally expensive and even infeasible for very large datasets since the training examples may not fit in main memory. To overcome this problem, previous work has used online learning algorithms to learn weights for MLNs. However, this prior work has only applied existing online algorithms, and there is no comprehensive study of online weight learning for MLNs. In this paper, we derive a new online algorithm for structured prediction using the primal-dual framework, apply it to learn weights for MLNs, and compare against existing online algorithms on three large, real-world datasets. The experimental results show that our new algorithm generally achieves better accuracy than existing methods, especially on noisy datasets.

Recognizing activities in real-world videos is a difficult problem exacerbated by background clutter, changes in camera angle & zoom, and rapid camera movements. Large corpora of labeled videos can be used to train automated activity recognition systems, but this requires expensive human labor and time. This paper explores how closed captions that naturally accompany many videos can act as weak supervision that allows automatically collecting "labeled" data for activity recognition. We show that such an approach can improve activity retrieval in soccer videos. Our system requires no manual labeling of video clips and needs minimal human supervision. We also present a novel caption classifier that uses additional linguistic information to determine whether a specific comment refers to an ongoing activity. We demonstrate that combining linguistic analysis and automatically trained activity recognizers can significantly improve the precision of video retrieval.

We present a novel framework for learning to interpret and generate language using only perceptual context as supervision. We demonstrate its capabilities by developing a system that learns to sportscast simulated robot soccer games in both English and Korean without any language-specific prior knowledge. Training employs only ambiguous supervision consisting of a stream of descriptive textual comments and a sequence of events extracted from the simulation trace. The system simultaneously establishes correspondences between individual comments and the events that they describe while building a translation model that supports both parsing and generation. We also present a novel algorithm for learning which events are worth describing. Human evaluations of the generated commentaries indicate they are of reasonable quality and in some cases even on par with those produced by humans for our limited domain.

We present a novel commentator system that learns language from sportscasts of simulated soccer games. The system learns to parse and generate commentaries without any engineered knowledge about the English language. Training is done using only ambiguous supervision in the form of textual human commentaries and simulation states of the soccer games. The system simultaneously tries to establish correspondences between the commentaries and the simulation states as well as build a translation model. We also present a novel algorithm, Iterative Generation Strategy Learning (IGSL), for deciding which events to comment on. Human evaluations of the generated commentaries indicate they are of reasonable quality compared to human commentaries.

Markov logic networks (MLNs) are an expressive representation for statistical relational learning that generalizes both first-order logic and graphical models. Existing methods for learning the logical structure of an MLN are not discriminative; however, many relational learning problems involve specific target predicates that must be inferred from given background information. We found that existing MLN methods perform very poorly on several such ILP benchmark problems, and we present improved discriminative methods for learning MLN clauses and weights that outperform existing MLN and traditional ILP methods.

We present a new approach to multiple instance learning (MIL) that is particularly effective when the positive bags are sparse (i.e. contain few positive instances). Unlike other SVM-based MIL methods, our approach more directly enforces the desired constraint that at least one of the instances in a positive bag is positive. Using both artificial and real-world data, we experimentally demonstrate that our approach achieves greater accuracy than state-of-the-art MIL methods when positive bags are sparse, and performs competitively when they are not. In particular, our approach is the best performing method for image region classification. 1.

Markov logic networks (MLNs) are a statistical relational model that consists of weighted firstorder clauses and generalizes first-order logic and Markov networks. The current state-of-the-art algorithm for learning MLN structure follows a top-down paradigm where many potential candidate structures are systematically generated without considering the data and then evaluated using a statistical measure of their fit to the data. Even though this existing algorithm outperforms an impressive array of benchmarks, its greedy search is susceptible to local maxima or plateaus. We present a novel algorithm for learning MLN structure that follows a more bottom-up approach to address this problem. Our algorithm uses a "propositional" Markov network learning method to construct "template" networks that guide the construction of candidate clauses. Our algorithm significantly improves accuracy and learning time over the existing topdown approach in three real-world domains.

We present a new approach for mapping natural language sentences to their formal meaning representations using string-kernel-based classifiers. Our system learns these classifiers for every production in the formal language grammar. Meaning representations for novel natural language sentences are obtained by finding the most probable semantic parse using these string classifiers. Our experiments on two real-world data sets show that this approach compares favorably to other existing systems and is particularly robust to noise.

Many data mining tasks require computing similarity between pairs of objects. Pairwise similarity computations are particularly important in record linkage systems, as well as in clustering and schema mapping algorithms. Because the number of object pairs grows quadratically with the size of the dataset, computing similarity between all pairs is impractical and becomes prohibitive for large datasets and complex similarity functions. Blocking methods alleviate this problem by efficiently selecting approximately similar object pairs for subsequent distance computations, leaving out the remaining pairs as dissimilar. Previously proposed blocking methods require manually constructing an index- based similarity function or selecting a set of predicates, followed by hand-tuning of parameters. In this paper, we introduce an adaptive framework for automatically learning blocking functions that are efficient and accurate. We describe two predicate-based formulations of learnable blocking functions and provide learning algorithms for training them. The effectiveness of the proposed techniques is demonstrated on real and simulated datasets, on which they prove to be more accurate than non-adaptive blocking methods.

In many classification tasks, training data have missing feature values that can be acquired at a cost. For building accurate predictive models, acquiring all missing values is often prohibitively expensive or unnecessary, while acquiring a random subset of feature values may not be most effective. The goal of active feature-value acquisition is to incrementally select feature values that are most cost-effective for improving the model's accuracy. We present an approach that acquires feature values for inducing a classification model based on an estimation of the expected improvement in model accuracy per unit cost. Experimental results demonstrate that our approach consistently reduces the cost of producing a model of a desired accuracy compared to random feature acquisitions.

Semi-supervised clustering algorithms aim to improve clustering results using limited supervision. The supervision is generally given as pairwise constraints; such constraints are natural for graphs, yet most semi-supervised clustering algorithms are designed for data represented as vectors. In this paper, we unify vector-based and graph-based approaches. We show that a recently-proposed objective function for semi-supervised clustering based on Hidden Markov Random Fields, with squared Euclidean distance and a certain class of constraint penalty functions, can be expressed as a special case of the weighted kernel k-means objective. A recent theoretical connection between kernel k-means and several graph clustering objectives enables us to perform semi-supervised clustering of data given either as vectors or as a graph. For vector data, the kernel approach also enables us to find clusters with non-linear boundaries in the input data space. Furthermore, we show that recent work on spectral learning (Kamvar et al., 2003) may be viewed as a special case of our formulation. We empirically show that our algorithm is able to outperform current state-of-the-art semi-supervised algorithms on both vector-based and graph-based data sets.

While the vast majority of clustering algorithms are partitional, many real world datasets have inherently overlapping clusters. Several approaches to finding overlapping clusters have come from work on analysis of biological datasets. In this paper, we interpret an overlapping clustering model proposed by Segal et al. [23] as a generalization of Gaussian mixture models, and we extend it to an overlapping clustering model based on mixtures of any regular exponential family distribution and the corresponding Bregman divergence. We provide the necessary algorithm modifications for this extension, and present results on synthetic data as well as subsets of 20-Newsgroups and EachMovie datasets.

Most information extraction (IE) systems treat separate potential extractions as independent. However, in many cases, considering influences between different potential extractions could improve overall accuracy. Statistical methods based on undirected graphical models, such as conditional random fields (CRFs), have been shown to be an effective approach to learning accurate IE systems. We present a new IE method that employs Relational Markov Networks (a generalization of CRFs), which can represent arbitrary dependencies between extractions. This allows for "collective information extraction" that exploits the mutual influence between possible extractions. Experiments on learning to extract protein names from biomedical text demonstrate the advantages of this approach.

Many induction problems include missing data that can be acquired at a cost. For building accurate predictive models, acquiring complete information for all instances is often expensive or unnecessary, while acquiring information for a random subset of instances may not be most effective. Active feature-value acquisition tries to reduce the cost of achieving a desired model accuracy by identifying instances for which obtaining complete information is most informative. We present an approach in which instances are selected for acquisition based on the current model's accuracy and its confidence in the prediction. Experimental results demonstrate that our approach can induce accurate models using substantially fewer feature-value acquisitions as compared to alternative policies.

Semi-supervised clustering employs a small amount of labeled data to aid unsupervised learning. Previous work in the area has utilized supervised data in one of two approaches: 1) constraint-based methods that guide the clustering algorithm towards a better grouping of the data, and 2) distance-function learning methods that adapt the underlying similarity metric used by the clustering algorithm. This paper provides new methods for the two approaches as well as presents a new semi-supervised clustering algorithm that integrates both of these techniques in a uniform, principled framework. Experimental results demonstrate that the unified approach produces better clusters than both individual approaches as well as previously proposed semi-supervised clustering algorithms.

Query by Committee is an effective approach to selective sampling in which disagreement amongst an ensemble of hypotheses is used to select data for labeling. Query by Bagging and Query by Boosting are two practical implementations of this approach that use Bagging and Boosting, respectively, to build the committees. For effective active learning, it is critical that the committee be made up of consistent hypotheses that are very different from each other. DECORATE is a recently developed method that directly constructs such diverse committees using artificial training data. This paper introduces ACTIVE-DECORATE, which uses DECORATE committees to select good training examples. Extensive experimental results demonstrate that, in general, ACTIVE-DECORATE outperforms both Query by Bagging and Query by Boosting.

Unsupervised clustering can be significantly improved using supervision in the form of pairwise constraints, i.e., pairs of instances labeled as belonging to same or different clusters. In recent years, a number of algorithms have been proposed for enhancing clustering quality by employing such supervision. Such methods use the constraints to either modify the objective function, or to learn the distance measure. We propose a probabilistic model for semi-supervised clustering based on Hidden Markov Random Fields (HMRFs) that provides a principled framework for incorporating supervision into prototype-based clustering. The model generalizes a previous approach that combines constraints and Euclidean distance learning, and allows the use of a broad range of clustering distortion measures, including Bregman divergences (e.g., Euclidean distance and I-divergence) and directional similarity measures (e.g., cosine similarity). We present an algorithm that performs partitional semi-supervised clustering of data by minimizing an objective function derived from the posterior energy of the HMRF model. Experimental results on several text data sets demonstrate the advantages of the proposed framework.

Semi-supervised clustering uses a small amount of supervised data to aid unsupervised learning. One typical approach specifies a limited number of must-link and cannot-link constraints between pairs of examples. This paper presents a pairwise constrained clustering framework and a new method for actively selecting informative pairwise constraints to get improved clustering performance. The clustering and active learning methods are both easily scalable to large datasets, and can handle very high dimensional data. Experimental and theoretical results confirm that this active querying of pairwise constraints significantly improves the accuracy of clustering when given a relatively small amount of supervision.

The problem of identifying approximately duplicate records in databases is an essential step for data cleaning and data integration processes. Most existing approaches have relied on generic or manually tuned distance metrics for estimating the similarity of potential duplicates. In this paper, we present a framework for improving duplicate detection using trainable measures of textual similarity. We propose to employ learnable text distance functions for each database field, and show that such measures are capable of adapting to the specific notion of similarity that is appropriate for the field's domain. We present two learnable text similarity measures suitable for this task: an extended variant of learnable string edit distance, and a novel vector-space based measure that employs a Support Vector Machine (SVM) for training. Experimental results on a range of datasets show that our framework can improve duplicate detection accuracy over traditional techniques.

This paper focuses on a system, WOLFIE (WOrd Learning From Interpreted Examples), that acquires a semantic lexicon from a corpus of sentences paired with semantic representations. The lexicon learned consists of phrases paired with meaning representations. WOLFIE is part of an integrated system that learns to transform sentences into representations such as logical database queries. Experimental results are presented demonstrating WOLFIE's ability to learn useful lexicons for a database interface in four different natural languages. The usefulness of the lexicons learned by WOLFIE are compared to those acquired by a similar system, with results favorable to WOLFIE. A second set of experiments demonstrates WOLFIE's ability to scale to larger and more difficult, albeit artificially generated, corpora. In natural language acquisition, it is difficult to gather the annotated data needed for supervised learning; however, unannotated data is fairly plentiful. Active learning methods attempt to select for annotation and training only the most informative examples, and therefore are potentially very useful in natural language applications. However, most results to date for active learning have only considered standard classification tasks. To reduce annotation effort while maintaining accuracy, we apply active learning to semantic lexicons. We show that active learning can significantly reduce the number of annotated examples required to achieve a given level of performance.

Variation and noise in database entries can prevent data mining algorithms, such as association rule mining, from discovering important regularities. In particular, textual fields can exhibit variation due to typographical errors, mispellings, abbreviations, etc.. By allowing partial or "soft matching" of items based on a similarity metric such as edit-distance or cosine similarity, additional important patterns can be detected. This paper introduces an algorithm, SoftApriori that discovers soft-matching association rules given a user-supplied similarity metric for each field. Experimental results on several "noisy" datasets extracted from text demonstrate that SoftApriori discovers additional relationships that more accurately reflect regularities in the data.

In this paper, we present a new method of estimating the novelty of rules discovered by data-mining methods using WordNet, a lexical knowledge-base of English words. We assess the novelty of a rule by the average semantic distance in a knowledge hierarchy between the words in the antecedent and the consequent of the rule - the more the average distance, more is the novelty of the rule. The novelty of rules extracted by the DiscoTEX text-mining system on Amazon.com book descriptions were evaluated by both human subjects and by our algorithm. By computing correlation coefficients between pairs of human ratings and between human and automatic ratings, we found that the automatic scoring of rules based on our novelty measure correlates with human judgments about as well as human judgments correlate with one another. @Text mining

The development of natural language interfaces (NLI's) for databases has been a challenging problem in natural language processing (NLP) since the 1970's. The need for NLI's has become more pronounced due to the widespread access to complex databases now available through the Internet. A challenging problem for empirical NLP is the automated acquisition of NLI's from training examples. We present a method for integrating statistical and relational learning techniques for this task which exploits the strength of both approaches. Experimental results from three different domains suggest that such an approach is more robust than a previous purely logic-based approach.

We present a knowledge and context-based system for parsing and translating natural language and evaluate it on sentences from the Wall Street Journal. Applying machine learning techniques, the system uses parse action examples acquired under supervision to generate a deterministic shift-reduce parser in the form of a decision structure. It relies heavily on context, as encoded in features which describe the morphological, syntactic, semantic and other aspects of a given parse state.

This paper presents a method for inducing logic programs from examples that learns a new class of concepts called first-order decision lists, defined as ordered lists of clauses each ending in a cut. The method, called FOIDL, is based on FOIL (Quinlan, 1990) but employs intensional background knowledge and avoids the need for explicit negative examples. It is particularly useful for problems that involve rules with specific exceptions, such as learning the past-tense of English verbs, a task widely studied in the context of the symbolic/connectionist debate. FOIDL is able to learn concise, accurate programs for this problem from significantly fewer examples than previous methods (both connectionist and symbolic).