Document authors commonly use tables to support arguments presented in the text. But, because tables are usually separate from the main body text, readers must split their attention between different parts of the document. We present an interactive document reader that automatically links document text with corresponding table cells. Readers can select a sentence (or tables cells) and our reader highlights the relevant table cells (or sentences). We provide an automatic pipeline for extracting such references between sentence text and table cells for existing PDF documents that combines structural analysis of tables with natural language processing and rule-based matching. On a test corpus of 330 (sentence, table) pairs, our pipeline correctly extracts 48.8% of the references. An additional 30.5% contain only false negatives (FN) errors -- the reference is missing table cells. The remaining 20.7% contain false positives (FP) errors -- the reference includes extraneous table cells and could therefore mislead readers. A user study finds that despite such errors, our interactive document reader helps readers match sentences with corresponding table cells more accurately and quickly than a baseline document reader.
more »
« less
TABBIE: Pretrained Representations of Tabular Data
Existing work on tabular representation-learning jointly models tables and associated text using self-supervised objective functions derived from pretrained language models such as BERT. While this joint pretraining improves tasks involving paired tables and text (e.g., answering questions about tables), we show that it underperforms on tasks that operate over tables without any associated text (e.g., populating missing cells). We devise a simple pretraining objective (corrupt cell detection) that learns exclusively from tabular data and reaches the state-of-the-art on a suite of table-based prediction tasks. Unlike competing approaches, our model (TABBIE) provides embeddings of all table substructures (cells, rows, and columns), and it also requires far less compute to train. A qualitative analysis of our model’s learned cell, column, and row representations shows that it understands complex table semantics and numerical trends.
more »
« less
- Award ID(s):
- 1955567
- NSF-PAR ID:
- 10254049
- Date Published:
- Journal Name:
- Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies
- Page Range / eLocation ID:
- 3446 to 3456
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Controlled table-to-text generation seeks to generate natural language descriptions for highlighted subparts of a table. Previous SOTA systems still employ a sequence-to-sequence generation method, which merely captures the table as a linear structure and is brittle when table layouts change. We seek to go beyond this paradigm by (1) effectively expressing the relations of content pieces in the table, and (2) making our model robust to content-invariant structural transformations. Accordingly, we propose an equivariance learning framework, which encodes tables with a structure-aware self-attention mechanism. This prunes the full self-attention structure into an order-invariant graph attention that captures the connected graph structure of cells belonging to the same row or column, and it differentiates between relevant cells and irrelevant cells from the structural perspective. Our framework also modifies the positional encoding mechanism to preserve the relative position of tokens in the same cell but enforce position invariance among different cells. Our technology is free to be plugged into existing table-to-text generation models, and has improved T5-based models to offer better performance on ToTTo and HiTab. Moreover, on a harder version of ToTTo, we preserve promising performance, while previous SOTA systems, even with transformation-based data augmentation, have seen significant performance drops.more » « less
-
A real-world text corpus sometimes comprises not only text documents, but also semantic links between them (e.g., academic papers in a bibliographic network are linked by citations and co-authorships). Text documents and semantic connections form a text-rich network, which empowers a wide range of downstream tasks such as classification and retrieval. However, pretraining methods for such structures are still lacking, making it difficult to build one generic model that can be adapted to various tasks on text-rich networks. Current pretraining objectives, such as masked language modeling, purely model texts and do not take inter-document structure information into consideration. To this end, we propose our PretrAining on TexT-Rich NetwOrk framework PATTON. PATTON1 includes two pretraining strategies: network-contextualized masked language modeling and masked node prediction, to capture the inherent dependency between textual attributes and network structure. We conduct experiments on four downstream tasks in five datasets from both academic and e-commerce domains, where PATTON outperforms baselines significantly and consistently.more » « less
-
Product catalogs, conceptually in the form of text-rich tables, are self-reported by individual retailers and thus inevitably contain noisy facts. Verifying such textual attributes in product catalogs is essential to improve their reliability. However, popular methods for processing free-text content, such as pre-trained language models, are not particularly effective on structured tabular data since they are typically trained on free-form natural language texts. In this paper, we present Tab-Cleaner, a model designed to handle error detection over text-rich tabular data following a pre-training / fine-tuning paradigm. We train Tab-Cleaner on a real-world Amazon Product Catalog table w.r.t millions of products and show improvements over state-of-the-art methods by 16% on PR AUC over attribute applicability classification task and by 11% on PR AUC over attribute value validation task.more » « less
-
Synthesizing information from collections of tables embedded within scientific and technical documents is increasingly critical to emerging knowledge-driven applications. Given their structural heterogeneity, highly domain-specific content, and diffuse context, inferring a precise semantic understanding of such tables is traditionally better accomplished through linking tabular content to concepts and entities in reference knowledge graphs. However, existing tabular data discovery systems are not designed to adequately exploit these explicit, human-interpretable semantic linkages. Moreover, given the prevalence of misinformation, the level of confidence in the reliability of tabular information has become an important, often overlooked, factor in the discovery over open datasets. We describe a preliminary implementation of a discovery engine that enables table-based semantic search and retrieval of tabular information from a linked knowledge graph of scientific tables. We discuss the viability of semantics-guided tabular data analysis operations, including on-the-fly table generation under reliability constraints, within discovery scenarios motivated by intelligence production from documents.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.18653/v1/2021.naacl-main.270
