Collecting and annotating task-oriented dialogues is time-consuming and costly. Thus, few-shot learning for dialogue tasks presents an exciting opportunity. In this work, we propose an in-context (IC) learning framework for few-shot dialogue state tracking (DST), where a large pre-trained language model (LM) takes a test instance and a few annotated examples as input, and directly decodes the dialogue states without any parameter updates. This makes the LM more flexible and scalable compared to prior few-shot DST work when adapting to new domains and scenarios. We study ways to formulate dialogue context into prompts for LMs and propose an efficient approach to retrieve dialogues as exemplars given a test instance and a selection pool of few-shot examples. To better leverage the pre-trained LMs, we also reformulate DST into a text-to-SQL problem. Empirical results on MultiWOZ 2.1 and 2.4 show that our method IC-DST outperforms previous fine-tuned state-of-the-art models in few-shot settings.
Language models increasingly rely on massive web dumps for diverse text data. However, these sources are rife with undesirable content. As such, resources like Wikipedia, books, and newswire often serve as anchors for automatically selecting web text most suitable for language modeling, a process typically referred to as quality filtering. Using a new dataset of U.S. high school newspaper articles -- written by students from across the country -- we investigate whose language is preferred by the quality filter used for GPT-3. We find that newspapers from larger schools, located in wealthier, educated, and urban ZIP codes are more likely to be classified as high quality. We then demonstrate that the filter's measurement of quality is unaligned with other sensible metrics, such as factuality or literary acclaim. We argue that privileging any corpus as high quality entails a language ideology, and more care is needed to construct training corpora for language models, with better transparency and justification for the inclusion or exclusion of various texts.
Structured knowledge grounding (SKG) leverages structured knowledge to complete user requests, such as semantic parsing over databases and question answering over knowledge bases. Since the inputs and outputs of SKG tasks are heterogeneous, they have been studied separately by different communities, which limits systematic and compatible research on SKG. In this paper, we overcome this limitation by proposing the SKG framework, which unifies 21 SKG tasks into a text-to-text format, aiming to promote systematic SKG research, instead of being exclusive to a single task, domain, or dataset. We use UnifiedSKG to benchmark T5 with different sizes and show that T5, with simple modifications when necessary, achieves state-of-the-art performance on almost all of the 21 tasks. We further demonstrate that multi-task prefix-tuning improves the performance on most tasks, largely improving the overall performance. UnifiedSKG also facilitates the investigation of zero-shot and few-shot learning, and we show that T0, GPT-3, and Codex struggle in zero-shot and few-shot learning for SKG. We also use UnifiedSKG to conduct a series of controlled experiments on structured knowledge encoding variants across SKG tasks. UnifiedSKG is easily extensible to more tasks, and it is open-sourced at https://github.com/hkunlp/unifiedskg Latest collections at https://unifiedskg.com.
A recurring challenge of crowdsourcing NLP datasets at scale is that human writers often rely on repetitive patterns when crafting examples, leading to a lack of linguistic diversity. We introduce a novel paradigm for dataset creation based on human and machine collaboration, which brings together the generative strength of language models and the evaluative strength of humans. Starting with an existing dataset, MultiNLI, our approach uses dataset cartography to automatically identify examples that demonstrate challenging reasoning patterns, and instructs GPT-3 to compose new examples with similar patterns. Machine generated examples are then automatically filtered, and finally revised and labeled by human crowdworkers to ensure quality. The resulting dataset, WANLI, consists of 108,357 natural language inference (NLI) examples that present unique empirical strengths over existing NLI datasets. Remarkably, training a model on WANLI instead of MNLI (which is 4 times larger) improves performance on seven out-of-domain test sets we consider, including by 11% on HANS and 9% on Adversarial NLI. Moreover, combining MNLI with WANLI is more effective than combining with other augmentation sets that have been introduced. Our results demonstrate the potential of natural language generation techniques to curate NLP datasets of enhanced quality and diversity.
Lifelong experiences and learned knowledge lead to shared expectations about how common situations tend to unfold. Such knowledge enables people to interpret story narratives and identify salient events effortlessly. We study differences in the narrative flow of events in autobiographical versus imagined stories using GPT-3, one of the largest neural language models created to date. The diary-like stories were written by crowdworkers about either a recently experienced event or an imagined event on the same topic. To analyze the narrative flow of events of these stories, we measured sentence *sequentiality*, which compares the probability of a sentence with and without its preceding story context. We found that imagined stories have higher sequentiality than autobiographical stories, and that the sequentiality of autobiographical stories is higher when they are retold than when freshly recalled. Through an annotation of events in story sentences, we found that the story types contain similar proportions of major salient events, but that the autobiographical stories are denser in factual minor events. Furthermore, in comparison to imagined stories, autobiographical stories contain more concrete words and words related to the first person, cognitive processes, time, space, numbers, social words, and core drives and needs. Our findings highlight the opportunity to investigate memory and cognition with large-scale statistical language models.
The dominant paradigm for neural text generation is left-to-right decoding from autoregressive language models. Constrained or controllable generation under complex lexical constraints, however, requires foresight to plan ahead feasible future paths. Drawing inspiration from the A* search algorithm, we propose NeuroLogic A*esque, a decoding algorithm that incorporates heuristic estimates of future cost. We develop efficient lookahead heuristics that are efficient for large-scale language models, making our method a drop-in replacement for common techniques such as beam search and top-k sampling. To enable constrained generation, we build on NeuroLogic decoding (Lu et al., 2021), combining its flexibility in incorporating logical constraints with A*esque estimates of future constraint satisfaction. Our approach outperforms competitive baselines on five generation tasks, and achieves new state-of-the-art performance on table-to-text generation, constrained machine translation, and keyword-constrained generation. The improvements are particularly notable on tasks that require complex constraint satisfaction or in few-shot or zero-shot settings. NeuroLogic A*esque illustrates the power of decoding for improving and enabling new capabilities of large-scale language models.
Natural language processing researchers have identified limitations of evaluation methodology for generation tasks, with new questions raised about the validity of automatic metrics and of crowdworker judgments. Meanwhile, efforts to improve generation models tend to focus on simple n-gram overlap metrics (e.g., BLEU, ROUGE). We argue that new advances on models and metrics should each more directly benefit and inform the other. We therefore propose a generalization of leaderboards, bidimensional leaderboards (Billboards), that simultaneously tracks progress in language generation tasks and metrics for their evaluation. Unlike conventional unidimensional leaderboards that sort submitted systems by predetermined metrics, a Billboard accepts both generators and evaluation metrics as competing entries. A Billboard automatically creates an ensemble metric that selects and linearly combines a few metrics based on a global analysis across generators. Further, metrics are ranked based on their correlations with human judgments. We release four Billboards for machine translation, summarization, and image captioning. We demonstrate that a linear ensemble of a few diverse metrics sometimes substantially outperforms existing metrics in isolation. Our mixed-effects model analysis shows that most automatic metrics, especially the reference-based ones, overrate machine over human generation, demonstrating the importance of updating metrics as generation models become stronger (and perhaps more similar to humans) in the future.
We establish a rubric-based human evaluation protocol for image captioning models. Our scoring rubrics and their definitions are carefully developed based on machine- and human-generated captions on the MSCOCO dataset. Each caption is evaluated along two main dimensions in a tradeoff (precision and recall) as well as other aspects that measure the text quality (fluency, conciseness, and inclusive language). Our evaluations demonstrate several critical problems of the current evaluation practice. Human-generated captions show substantially higher quality than machine-generated ones, especially in coverage of salient information (i.e., recall), while all automatic metrics say the opposite. Our rubric-based results reveal that CLIPScore, a recent metric that uses image features, better correlates with human judgments than conventional text-only metrics because it is more sensitive to recall. We hope that this work will promote a more transparent evaluation protocol for image captioning and its automatic metrics.