Ruling-based table analysis for noisy handwritten documents

Chen, Jin; Lopresti, Daniel

doi:10.1145/2505377.2505392

Cited by 7 publications

(3 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…2) Datasets like PC32 (Lin et al, 2020) fail to accurately reflect the real-world distribution of data, with high-resource languages like French and German disproportionately represented by 40 million and 4 million sentences, respectively. 3) Linguistic diversity, a critical factor, is often overlooked in datasets such as Europarl (Koehn, 2005) and Mul-tiUN (Chen and Eisele, 2012). 4) Lastly, systematic zero-shot NMT evaluations are rarely found in existing MNMT datasets, either missing entirely or covering less than 1% of possible zero-shot combinations (Aharoni et al, 2019;Pan et al, 2021;Tang et al, 2021).…”

Section: Ec40 Datasetmentioning

confidence: 99%

Towards a Better Understanding of Variations in Zero-Shot Neural Machine Translation Performance

Tan,

Monz

2023

Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing

View full text Add to dashboard Cite

Multilingual Neural Machine Translation (MNMT) facilitates knowledge sharing but often suffers from poor zero-shot (ZS) translation qualities. While prior work has explored the causes of overall low zero-shot translation qualities, our work introduces a fresh perspective: the presence of significant variations in zeroshot performance. This suggests that MNMT does not uniformly exhibit poor zero-shot capability; instead, certain translation directions yield reasonable results. Through systematic experimentation, spanning 1,560 language directions across 40 languages, we identify three key factors contributing to high variations in ZS NMT performance: 1) target-side translation quality, 2) vocabulary overlap, and 3) linguistic properties. Our findings highlight that the target side translation quality is the most influential factor, with vocabulary overlap consistently impacting zero-shot capabilities. Additionally, linguistic properties, such as language family and writing system, play a role, particularly with smaller models. Furthermore, we suggest that the off-target issue is a symptom of inadequate performance, emphasizing that zero-shot translation challenges extend beyond addressing the off-target problem. To support future research, we release the data and models as a benchmark for the study of ZS NMT. 1

show abstract

Section: Ec40 Datasetmentioning

confidence: 99%

Towards a Better Understanding of Variations in Zero-Shot Neural Machine Translation Performance

Tan,

Monz

2023

Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing

View full text Add to dashboard Cite

show abstract

“…Ziemski et al (2016) created the United Nations Parallel Corpus, which consists of over 2 million words of parallel texts in 6 official languages, including English and Arabic. Another work that includes Arabic is the multilingual parallel corpus MultiUN (Chen and Eisele, 2012). It extends the United Nations Parallel Corpus by including texts from various sources such as the United Nations and other international organisations.…”

Section: Related Workmentioning

confidence: 99%

AlphaMWE-Arabic: Arabic Edition of Multilingual Parallel Corpora with Multiword Expression Annotations

Hadj Mohamed,

Rassem,

Han

et al. 2023

Proceedings of the Conference Recent Advances in Natural Language Processing - Large Language Models for Natural Language Proce

View full text Add to dashboard Cite

Multiword Expressions (MWEs) have been a bottleneck for Natural Language Understanding (NLU) and Natural Language Generation (NLG) tasks due to their idiomaticity, ambiguity, and noncompositionality. Bilingual parallel corpora introducing MWE annotations are very scarce which set another challenge for current Natural Language Processing (NLP) systems, especially in a multilingual setting.This work presents AlphaMWE-Arabic, an Arabic edition of the AlphaMWE parallel corpus with MWE annotations. We introduce how we created this corpus including machine translation (MT), post-editing, and annotations for both standard and dialectal varieties, i.e. Tunisian and Egyptian Arabic. We analyse the MT errors when they meet MWEs-related content, both quantitatively using the human-in-the-loop metric HOPE and qualitatively. We report the current state-of-the-art MT systems are far from reaching human parity performances. We expect our bilingual English-Arabic corpus will be an asset for multilingual research on MWEs such as translation and localisation, as well as for monolingual settings including the study of Arabicspecific lexicography and phrasal verbs on MWEs. Our corpus and experimental data are available at https://github. com/aaronlifenghan/AlphaMWE

show abstract

“…Chen and Lopresti [2] use a probabilistic alternative of Hough transform to detect lines in the document. In order to ensure high recall of table rulings, some lines are excluded based on the fact that the table ruling lines are parallel or orthogonal.…”

Section: Introductionmentioning

confidence: 99%

Table Detection in Handwritten Chemistry Documents Using Conditional Random Fields

Ghanmi

Belaïd²

2014

2014 14th International Conference on Frontiers in Handwriting Recognition

View full text Add to dashboard Cite

In this paper, we present a new approach using conditional random fields (CRFs) to localize tabular components in an unconstrained handwritten compound document. Given a line-segmented document, the extraction of table is considered as a labeling task that consists in assigning a label to each line: TableRow label for a line which belongs to a table and LineText label for a line which belongs to a text block. To perform the labeling task, we use a CRF model to combine two classifiers: a local classifier which assigns a label to the line based on local features and a contextual classifier which uses features taking into account the neighborhood. The CRF model gives the global conditional probability of a given labeling of the line considering the outputs of the two classifiers. A set of chemistry documents is used for the evaluation of this approach. The obtained results are around 88% of table lines correctly detected.

show abstract

Ruling-based table analysis for noisy handwritten documents

Abstract: ABSTRACT

Cited by 7 publications

References 21 publications

Towards a Better Understanding of Variations in Zero-Shot Neural Machine Translation Performance

Towards a Better Understanding of Variations in Zero-Shot Neural Machine Translation Performance

AlphaMWE-Arabic: Arabic Edition of Multilingual Parallel Corpora with Multiword Expression Annotations

Table Detection in Handwritten Chemistry Documents Using Conditional Random Fields

Contact Info

Product

Resources

About