LAMBERT: Layout-Aware (Language) Modeling for information extraction

We introduce a new simple approach to the problem of understanding documents where non-trivial layout influences the local semantics. To this end, we modify the Transformer encoder architecture in a way that allows it to use layout features obtained from an OCR system, without the need to re-learn the language semantics from scratch. We augment the input of the model only with the coordinates of token bounding boxes, avoiding the use of raw images. This leads to a layout-aware language model which can be then fine-tuned on downstream tasks. The model is evaluated on an end-to-end information extraction task using four publicly available datasets: Kleister NDA, Kleister Charity, SROIE and CORD. We show that it achieves superior performance on datasets consisting of visually rich documents, at the same time outperforming the baseline RoBERTa on documents with flat layout (NDA F1 increase from 78.50 to 80.42). Our solution ranked 1st on the public leaderboard for the Key Information Extraction from the SROIE dataset, improving the SOTA F1-score from 97.81 to 98.17.