Planning in a text-based environment continues to be a significant challenge for AI systems. Recent approaches have utilized language models to predict planning domain definitions (e.g., PDDL) but have only been evaluated in closed-domain simulated environments. To address this, we present Proc2PDDL, the first dataset containing open-domain procedural texts paired with expert-annotated PDDL representations. Using this dataset, we evaluate the task of predicting domain actions (parameters, preconditions, and effects). We experiment with various large language models (LLMs) and prompting mechanisms, including a novel instruction inspired by the zone of proximal development (ZPD), which reconstructs the task as incremental basic skills. Our results demonstrate that Proc2PDDL is highly challenging for end-to-end LLMs, with GPT-3.5’s success rate close to 0% and GPT-4o’s 38%. With ZPD instructions, GPT-4o’s success rate increases to 45%, outperforming regular chain-of-thought prompting’s 34%. Our analysis systematically examines both syntactic and semantic errors, providing insights into the strengths and weaknesses of language models in generating domain-specific programs.
How-to procedures, such as how to plant a garden, are now used by millions of users, but sometimes need customizing to meet a user’s specific needs, e.g., planting a garden without pesticides. Our goal is to measure and improve an LLM’s ability to perform such customization. Our approach is to test several simple multi-LLM-agent architectures for customization, as well as an end-to-end LLM, using a new evaluation set, called CustomPlans, of over 200 WikiHow procedures each with a customization need. We find that a simple architecture with two LLM agents used sequentially performs best, one that edits a generic how-to procedure and one that verifies its executability, significantly outperforming (10.5% absolute) an end-to-end prompted LLM. This suggests that LLMs can be configured reasonably effectively for procedure customization. This also suggests that multi-agent editing architectures may be worth exploring further for other customization applications (e.g. coding, creative writing) in the future.
Pretrained Language Models (PLMs) have advanced Natural Language Processing (NLP) tasks significantly, but finetuning PLMs on low-resource datasets poses significant challenges such as instability and overfitting. Previous methods tackle these issues by finetuning a strategically chosen subnetwork on a downstream task, while keeping the remaining weights fixed to the pretrained weights. However, they rely on a suboptimal criteria for sub-network selection, leading to suboptimal solutions. To address these limitations, we propose a regularization method based on attention-guided weight mixup for finetuning PLMs. Our approach represents each network weight as a mixup of task-specific weight and pretrained weight, controlled by a learnable attention parameter, providing finer control over sub-network selection. Furthermore, we employ a bi-level optimization (BLO) based framework on two separate splits of the training dataset, improving generalization and combating overfitting. We validate the efficacy of our proposed method through extensive experiments, demonstrating its superiority over previous methods, particularly in the context of finetuning PLMs on low-resource datasets. Our code is available at https://github.com/Sai-Ashish/Attention_guided_weight_mixup_BLO.
Planning in textual environments have been shown to be a long-standing challenge even for current models. A recent, promising line of work uses LLMs to generate a formal representation of the environment that can be solved by a symbolic planner. However, existing methods rely on a fully-observed environment where all entity states are initially known, so a one-off representation can be constructed, leading to a complete plan. In contrast, we tackle partially-observed environments where there is initially no sufficient information to plan for the end-goal. We propose PDDLEGO that iteratively construct a planning representation that can lead to a partial plan for a given sub-goal. By accomplishing the sub-goal, more information is acquired to augment the representation, eventually achieving the end-goal. We show that plans produced by few-shot PDDLEGO are 43% more efficient than generating plans end-to-end on the Coin Collector simulation, with strong performance (98%) on the more complex Cooking World simulation where end-to-end LLMs fail to generate coherent plans (4%).
Much texts describe a changing world (e.g., procedures, stories, newswires), and understanding them requires tracking how entities change. An earlier dataset, OpenPI, provided crowdsourced annotations of entity state changes in text. However, a major limitation was that those annotations were free-form and did not identify salient changes, hampering model evaluation. To overcome these limitations, we present an improved dataset, OpenPI2.0, where entities and attributes are fully canonicalized and additional entity salience annotations are added. On our fairer evaluation setting, we find that current state-of-the-art language models are far from competent. We also show that using state changes of salient entities as a chain-of-thought prompt, downstream performance is improved on tasks such as question answering and classical planning, outperforming the setting involving all related entities indiscriminately. We offer OpenPI2.0 for the continued development of models that can understand the dynamics of entities in text.
Script learning studies how daily events unfold. It enables machines to reason about narratives with implicit information. Previous works mainly consider a script as a linear sequence of events while ignoring the potential branches that arise due to people’s circumstantial choices. We hence propose Choice-75, the first benchmark that challenges intelligent systems to make decisions given descriptive scenarios, containing 75 scripts and more than 600 scenarios. We also present preliminary results with current large language models (LLM). Although they demonstrate overall decent performances, there is still notable headroom in hard scenarios.
The rise of social media platforms has led to an increase in polarised online discussions, especially on political and socio-cultural topics such as elections and climate change. We propose a simple and entirely novel unsupervised method to better predict whether the authors of two posts agree or disagree, leveraging user stances about named entities obtained from their posts. We present STEntConv, a model which builds a graph of users and named entities weighted by stance and trains a Signed Graph Convolutional Network (SGCN) to detect disagreement between comment and reply posts. We run experiments and ablation studies and show that including this information improves disagreement detection performance on a dataset of Reddit posts for a range of controversial subreddit topics, without the need for platform-specific features or user history
Schema induction builds a graph representation explaining how events unfold in a scenario. Existing approaches have been based on information retrieval (IR) and information extraction (IE), often with limited human curation. We demonstrate a human-in-the-loop schema induction system powered by GPT-3. We first describe the different modules of our system, including prompting to generate schematic elements, manual edit of those elements, and conversion of those into a schema graph. By qualitatively comparing our system to previous ones, we show that our system not only transfers to new domains more easily than previous approaches, but also reduces efforts of human curation thanks to our interactive interface.
Entities and events are crucial to natural language reasoning and common in procedural texts. Existing work has focused either exclusively on entity state tracking (e.g., whether a pan is hot) or on event reasoning (e.g., whether one would burn themselves by touching the pan), while these two tasks are often causally related. We propose CREPE, the first benchmark on causal reasoning of event plausibility and entity states. We show that most language models, including GPT-3, perform close to chance at .35 F1, lagging far behind human at .87 F1. We boost model performance to .59 F1 by creatively representing events as programming languages while prompting language models pretrained on code. By injecting the causal relations between entities and events as intermediate reasoning steps in our representation, we further boost the performance to .67 F1. Our findings indicate not only the challenge that CREPE brings for language models, but also the efficacy of code-like prompting combined with chain-of-thought prompting for multihop event reasoning.
Recent work has shown that prompting language models with code-like representations of natural language leads to performance improvements on structured reasoning tasks. However, such tasks comprise only a small subset of all natural language tasks. In our work, we seek to answer whether or not code-prompting is the preferred way of interacting with language models in general. We compare code and text prompts across three popular GPT models (davinci, code-davinci-002, and text-davinci-002) on a broader selection of tasks (e.g., QA, sentiment, summarization) and find that with few exceptions, code prompts do not consistently outperform text prompts. Furthermore, we show that the style of code prompt has a large effect on performance for some (but not all) tasks and that fine-tuning on text instructions leads to better relative performance of code prompts.
Recursive noun phrases (NPs) have interesting semantic properties. For example, “my favorite new movie” is not necessarily my favorite movie, whereas “my new favorite movie” is. This is common sense to humans, yet it is unknown whether language models have such knowledge. We introduce the Recursive Noun Phrase Challenge (RNPC), a dataset of three textual inference tasks involving textual entailment and event plausibility comparison, precisely targeting the understanding of recursive NPs. When evaluated on RNPC, state-of-the-art Transformer models only perform around chance. Still, we show that such knowledge is learnable with appropriate data. We further probe the models for relevant linguistic features that can be learned from our tasks, including modifier semantic category and modifier scope. Finally, models trained on RNPC achieve strong zero-shot performance on an extrinsic Harm Detection evaluation task, showing the usefulness of the understanding of recursive NPs in downstream applications.
Argument classification is at the core of Semantic Role Labeling. Given a sentence and the predicate, a semantic role label is assigned to each argument of the predicate. While semantic roles come with meaningful definitions, existing work has treated them as symbolic. Learning symbolic labels usually requires ample training data, which is frequently unavailable due to the cost of annotation. We instead propose to retrieve and leverage the definitions of these labels from the annotation guidelines. For example, the verb predicate “work” has arguments defined as “worker”, “job”, “employer”, etc. Our model achieves state-of-the-art performance on the CoNLL09 dataset injected with label definitions given the predicate senses. The performance improvement is even more pronounced in low-resource settings when training data is scarce.
Procedures are inherently hierarchical. To “make videos”, one may need to “purchase a camera”, which in turn may require one to “set a budget”. While such hierarchical knowledge is critical for reasoning about complex procedures, most existing work has treated procedures as shallow structures without modeling the parent-child relation. In this work, we attempt to construct an open-domain hierarchical knowledge-base (KB) of procedures based on wikiHow, a website containing more than 110k instructional articles, each documenting the steps to carry out a complex procedure. To this end, we develop a simple and efficient method that links steps (e.g., “purchase a camera”) in an article to other articles with similar goals (e.g., “how to choose a camera”), recursively constructing the KB. Our method significantly outperforms several strong baselines according to automatic evaluation, human judgment, and application to downstream tasks such as instructional video retrieval.
Entity linking, the task of linking potentially ambiguous mentions in texts to corresponding knowledge-base entities, is an important component for language understanding. We address two challenge in entity linking: how to leverage wider contexts surrounding a mention, and how to deal with limited training data. We propose a fully unsupervised model called SumMC that first generates a guided summary of the contexts conditioning on the mention, and then casts the task to a multiple-choice problem where the model chooses an entity from a list of candidates. In addition to evaluating our model on existing datasets that focus on named entities, we create a new dataset that links noun phrases from WikiHow to Wikidata. We show that our SumMC model achieves state-of-the-art unsupervised performance on our new dataset and on exiting datasets.
Evaluations in machine learning rarely use the latest metrics, datasets, or human evaluation in favor of remaining compatible with prior work. The compatibility, often facilitated through leaderboards, thus leads to outdated but standardized evaluation practices. We pose that the standardization is taking place in the wrong spot. Evaluation infrastructure should enable researchers to use the latest methods and what should be standardized instead is how to incorporate these new evaluation advances. We introduce GEMv2, the new version of the Generation, Evaluation, and Metrics Benchmark which uses a modular infrastructure for dataset, model, and metric developers to benefit from each other’s work. GEMv2 supports 40 documented datasets in 51 languages, ongoing online evaluation for all datasets, and our interactive tools make it easier to add new datasets to the living benchmark.
We present a multi-level geocoding model (MLG) that learns to associate texts to geographic coordinates. The Earth’s surface is represented using space-filling curves that decompose the sphere into a hierarchical grid. MLG balances classification granularity and accuracy by combining losses across multiple levels and jointly predicting cells at different levels simultaneously. It obtains large gains without any gazetteer metadata, demonstrating that it can effectively learn the connection between text spans and coordinates—and thus makes it a gazetteer-free geocoder. Furthermore, MLG obtains state-of-the-art results for toponym resolution on three English datasets without any dataset-specific tuning.
This paper proposes a new problem of complementary evidence identification for open-domain question answering (QA). The problem aims to efficiently find a small set of passages that covers full evidence from multiple aspects as to answer a complex question. To this end, we proposes a method that learns vector representations of passages and models the sufficiency and diversity within the selected set, in addition to the relevance between the question and passages. Our experiments demonstrate that our method considers the dependence within the supporting evidence and significantly improves the accuracy of complementary evidence selection in QA domain.
The knowledge of scripts, common chains of events in stereotypical scenarios, is a valuable asset for task-oriented natural language understanding systems. We propose the Goal-Oriented Script Construction task, where a model produces a sequence of steps to accomplish a given goal. We pilot our task on the first multilingual script learning dataset supporting 18 languages collected from wikiHow, a website containing half a million how-to articles. For baselines, we consider both a generation-based approach using a language model and a retrieval-based approach by first retrieving the relevant steps from a large candidate pool and then ordering them. We show that our task is practical, feasible but challenging for state-of-the-art Transformer models, and that our methods can be readily deployed for various other datasets and domains with decent zero-shot performance.
Understanding what sequence of steps are needed to complete a goal can help artificial intelligence systems reason about human activities. Past work in NLP has examined the task of goal-step inference for text. We introduce the visual analogue. We propose the Visual Goal-Step Inference (VGSI) task, where a model is given a textual goal and must choose which of four images represents a plausible step towards that goal. With a new dataset harvested from wikiHow consisting of 772,277 images representing human actions, we show that our task is challenging for state-of-the-art multimodal models. Moreover, the multimodal representation learned from our data can be effectively transferred to other datasets like HowTo100m, increasing the VGSI accuracy by 15 - 20%. Our task will facilitate multimodal reasoning about procedural events.
Modern task-oriented dialog systems need to reliably understand users’ intents. Intent detection is even more challenging when moving to new domains or new languages, since there is little annotated data. To address this challenge, we present a suite of pretrained intent detection models which can predict a broad range of intended goals from many actions because they are trained on wikiHow, a comprehensive instructional website. Our models achieve state-of-the-art results on the Snips dataset, the Schema-Guided Dialogue dataset, and all 3 languages of the Facebook multilingual dialog datasets. Our models also demonstrate strong zero- and few-shot performance, reaching over 75% accuracy using only 100 training examples in all datasets.
We introduce SmartCiteCon (SCC), a Java API for extracting both explicit and implicit citation context from academic literature in English. The tool is built on a Support Vector Machine (SVM) model trained on a set of 7,058 manually annotated citation context sentences, curated from 34,000 papers from the ACL Anthology. The model with 19 features achieves F1=85.6%. SCC supports PDF, XML, and JSON files out-of-box, provided that they are conformed to certain schemas. The API supports single document processing and batch processing in parallel. It takes about 12–45 seconds on average depending on the format to process a document on a dedicated server with 6 multithreaded cores. Using SCC, we extracted 11.8 million citation context sentences from ~33.3k PMC papers in the CORD-19 dataset, released on June 13, 2020. We will provide continuous supplementary data contribution to the CORD-19 and other datasets. The source code is released at https://gitee.com/irlab/SmartCiteCon.
Split and Rephrase is a text simplification task of rewriting a complex sentence into simpler ones. As a relatively new task, it is paramount to ensure the soundness of its evaluation benchmark and metric. We find that the widely used benchmark dataset universally contains easily exploitable syntactic cues caused by its automatic generation process. Taking advantage of such cues, we show that even a simple rule-based model can perform on par with the state-of-the-art model. To remedy such limitations, we collect and release two crowdsourced benchmark datasets. We not only make sure that they contain significantly more diverse syntax, but also carefully control for their quality according to a well-defined set of criteria. While no satisfactory automatic metric exists, we apply fine-grained manual evaluation based on these criteria using crowdsourcing, showing that our datasets better represent the task and are significantly more challenging for the models.
We propose a suite of reasoning tasks on two types of relations between procedural events: goal-step relations (“learn poses” is a step in the larger goal of “doing yoga”) and step-step temporal relations (“buy a yoga mat” typically precedes “learn poses”). We introduce a dataset targeting these two relations based on wikiHow, a website of instructional how-to articles. Our human-validated test set serves as a reliable benchmark for common-sense inference, with a gap of about 10% to 20% between the performance of state-of-the-art transformer models and human performance. Our automatically-generated training set allows models to effectively transfer to out-of-domain tasks requiring knowledge of procedural events, with greatly improved performances on SWAG, Snips, and Story Cloze Test in zero- and few-shot settings.
Multi-relational semantic similarity datasets define the semantic relations between two short texts in multiple ways, e.g., similarity, relatedness, and so on. Yet, all the systems to date designed to capture such relations target one relation at a time. We propose a multi-label transfer learning approach based on LSTM to make predictions for several relations simultaneously and aggregate the losses to update the parameters. This multi-label regression approach jointly learns the information provided by the multiple relations, rather than treating them as separate tasks. Not only does this approach outperform the single-task approach and the traditional multi-task learning approach, but it also achieves state-of-the-art performance on all but one relation of the Human Activity Phrase dataset.
Aspect level sentiment classification is a fine-grained sentiment analysis task. To detect the sentiment towards a particular aspect in a sentence, previous studies have developed various attention-based methods for generating aspect-specific sentence representations. However, the attention may inherently introduce noise and downgrade the performance. In this paper, we propose constrained attention networks (CAN), a simple yet effective solution, to regularize the attention for multi-aspect sentiment analysis, which alleviates the drawback of the attention mechanism. Specifically, we introduce orthogonal regularization on multiple aspects and sparse regularization on each single aspect. Experimental results on two public datasets demonstrate the effectiveness of our approach. We further extend our approach to multi-task settings and outperform the state-of-the-art methods.
To be informative, an evaluation must measure how well systems generalize to realistic unseen data. We identify limitations of and propose improvements to current evaluations of text-to-SQL systems. First, we compare human-generated and automatically generated questions, characterizing properties of queries necessary for real-world applications. To facilitate evaluation on multiple datasets, we release standardized and improved versions of seven existing datasets and one new text-to-SQL dataset. Second, we show that the current division of data into training and test sets measures robustness to variations in the way questions are asked, but only partially tests how well systems generalize to new queries; therefore, we propose a complementary dataset split for evaluation of future work. Finally, we demonstrate how the common practice of anonymizing variables during evaluation removes an important challenge of the task. Our observations highlight key difficulties, and our methodology enables effective measurement of future development.