With the rapid improvement of large language models capabilities, there has been increasing interest in challenging constrained text generation problems. However, existing benchmarks for constrained generation usually focus on fixed constraint types (e.g. generate a sentence containing certain words) that have proved to be easy for state-of-the-art models like GPT-4. We present COLLIE, a grammar- based framework that allows the specification of rich, compositional constraints with diverse generation levels (word, sentence, paragraph, passage) and modeling challenges (e.g. language understanding, logical reasoning, counting, semantic planning). We also develop tools for automatic extraction of task instances given a constraint structure and a raw text corpus. Using COLLIE, we compile the COLLIE- v1 dataset with 2,080 instances comprising 13 constraint structures. We perform systematic experiments across five state-of-the-art instruction-tuned language mod- els and analyze their performances to reveal shortcomings. COLLIE is designed to be extensible and lightweight, and we hope the community finds it useful to develop more complex constraints and evaluations in the future.
more »
« less
Sentence-Permuted Paragraph Generation
Generating paragraphs of diverse contents is important in many applications. Existing generation models produce similar contents from homogenized contexts due to the fixed left-to-right sentence order. Our idea is permuting the sentence orders to improve the content diversity of multi-sentence paragraph. We propose a novel framework PermGen whose objective is to maximize the expected log-likelihood of output paragraph distributions with respect to all possible sentence orders. PermGen uses hierarchical positional embedding and designs new procedures for training, and decoding in the sentence-permuted generation. Experiments on three paragraph generation benchmarks demonstrate PermGen generates more diverse outputs with a higher quality than existing models.
more »
« less
- NSF-PAR ID:
- 10334389
- Date Published:
- Journal Name:
- Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing
- Page Range / eLocation ID:
- 5051 to 5062
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Verifiable generation requires large language models (LLMs) to cite source documents supporting their outputs, thereby improve output transparency and trustworthiness. Yet, previous work mainly targets the generation of sentencelevel citations, lacking specificity about which part of the sentence is backed by which cited source. This work studies verifiable generation with subsentence-level fine-grained citations to locate the generated content that is supported by the cited sources in a more precise way. We first present a dataset, SCIFI, comprising 10K Wikipedia paragraphs with subsentence-level citations.1 Each paragraph in SCIFI is paired with a set of candidate source documents for citation and a query that triggers the generation of the paragraph content. On SCIFI, we then evaluate the performance of state-of-the-a rt LLMs and strategies for processing long documents designed for these models. Our experiment results reveal key factors that can enhance the quality of citations, including the expansion of the source documents’ context to be accessible to the models and the implementation of specialized model tuning.more » « less
-
A model that maps the requisite skills, or knowledge components, to the contents of an online course is necessary to implement many adaptive learning technologies. However, developing a skill model and tagging courseware contents with individual skills can be expensive and error prone. We propose a technology to automatically identify latent skills from instructional text on existing online courseware called Smart (Skill Model mining with Automated detection of Resemblance among Texts). Smart is capable of mining, labeling, and mapping skills without using an existing skill model or student learning (aka response) data. The goal of our proposed approach is to mine latent skills from assessment items included in existing courseware, provide discovered skills with human-friendly labels, and map didactic paragraph texts with skills. This way, mapping between assessment items and paragraph texts is formed. In doing so, automated skill models produced by Smart will reduce the workload of courseware developers while enabling adaptive online content at the launch of the course. In our evaluation study, we applied Smart to two existing authentic online courses. We then compared machine-generated skill models and human-crafted skill models in terms of the accuracy of predicting students’ learning. We also evaluated the similarity between machine-generated and human-crafted skill models. The results show that student models based on Smart-generated skill models were equally predictive of students’ learning as those based on human-crafted skill models— as validated on two OLI (Open Learning Initiative) courses. Also, Smart can generate skill models that are highly similar to human-crafted models as evidenced by the normalized mutual information (NMI) values.more » « less
-
Building effective text generation systems requires three critical components: content selection, text planning, and surface realization, and traditionally they are tackled as separate problems. Recent all-in-one style neural generation models have made impressive progress, yet they often produce outputs that are incoherent and unfaithful to the input. To address these issues, we present an end-to-end trained two-step generation model, where a sentence-level content planner first decides on the keyphrases to cover as well as a desired language style, followed by a surface realization decoder that generates relevant and coherent text. For experiments, we consider three tasks from domains with diverse topics and varying language styles: persuasive argument construction from Reddit, paragraph generation for normal and simple versions of Wikipedia, and abstract generation for scientific articles. Automatic evaluation shows that our system can significantly outperform competitive comparisons. Human judges further rate our system generated text as more fluent and correct, compared to the generations by its variants that do not consider language style.more » « less
-
Deep neural networks have achieved great successes on the image captioning task. However, most of the existing models depend heavily on paired image-sentence datasets, which are very expensive to acquire. In this paper, we make the first attempt to train an image captioning model in an unsupervised manner. Instead of relying on manually labeled image-sentence pairs, our proposed model merely requires an image set, a sentence corpus, and an existing visual concept detector. The sentence corpus is used to teach the captioning model how to generate plausible sentences. Meanwhile, the knowledge in the visual concept detector is distilled into the captioning model to guide the model to recognize the visual concepts in an image. In order to further encourage the generated captions to be semantically consistent with the image, the image and caption are projected into a common latent space so that they can reconstruct each other. Given that the existing sentence corpora are mainly designed for linguistic research and are thus with little reference to image contents, we crawl a large-scale image description corpus of two million natural sentences to facilitate the unsupervised image captioning scenario. Experimental results show that our proposed model is able to produce quite promising results without any caption annotations.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.18653/v1/2021.emnlp-main.412
