The task of document classification, particularly multi-label classification, presents a significant challenge due to the complexity of assigning multiple relevant labels to each document. This complexity is further amplified in multi-property multi-label classification tasks, where documents must be categorized across various sets of labels. In this research, we introduce an innovative encoder embedding-driven approach to multi-property multi-label document classification that leverages semantic-text similarity and the reuse of pre-existing annotated data to enhance the efficiency and accuracy of the document annotation process. Our method requires only a single model for text similarity, eliminating the need for multiple property-specific classifiers and thereby reducing computational demands and simplifying deployment. We evaluate our approach through a prototype deployed for daily operations, which demonstrates superior performance over existing classification systems. Our contributions include improved accuracy without additional training, increased efficiency, and demonstrated effectiveness in practical applications. The results of our study indicate the potential of our approach to be applied across various domains requiring multi-property multi-label document classification, offering a scalable and adaptable solution for metadata annotation tasks.

Mathematical expressions (ME) are widely used in scholar documents. In this paper we analyze characteristics of textual and visual MEs characteristics for the image-to-LaTeX translation task. While there are open data-sets of LaTeX files with MEs included it is very complicated to extract these MEs from a document and to compile the list of MEs. Therefore we release a corpus of open-access scholar documents with PDF and JATS-XML parallel files. The MEs in these documents are LaTeX encoded and are document independent. The data contains more than 1.2 million distinct annotated formulae and more than 80 million raw tokens of LaTeX MEs in more than 8 thousand documents. While the variety of textual lengths and visual sizes of MEs are not well defined we found that the task of analyzing MEs in scholar documents can be reduced to the subtask of a particular text length, image width and height bounds, and display MEs can be processed as arrays of partial MEs.

We describe the VTeX Language Editing Dataset of Academic Texts (LEDAT), a dataset of text extracts from scientific papers that were edited by professional native English language editors at VTeX. The goal of the LEDAT is to provide a large data resource for the development of language evaluation and grammar error correction systems for the scientific community. We describe the data collection and the compilation process of the LEDAT. The new dataset can be used in many NLP studies and applications where deeper knowledge of the academic language and language editing is required. The dataset can be used also as a knowledge base of English academic language to support many writers of scientific papers.

This paper describes the UPC-BMIC-VMU participation in the IWSLT 2010 evaluation campaign. The SMT system is a standard phrase-based enriched with novel segmentations. These novel segmentations are computed using statistical measures such as Log-likelihood, T-score, Chi-squared, Dice, Mutual Information or Gravity-Counts. The analysis of translation results allows to divide measures into three groups. First, Log-likelihood, Chi-squared and T-score tend to combine high frequency words and collocation segments are very short. They improve the SMT system by adding new translation units. Second, Mutual Information and Dice tend to combine low frequency words and collocation segments are short. They improve the SMT system by smoothing the translation units. And third, GravityCounts tends to combine high and low frequency words and collocation segments are long. However, in this case, the SMT system is not improved. Thus, the road-map for translation system improvement is to introduce new phrases with either low frequency or high frequency words. It is hard to introduce new phrases with low and high frequency words in order to improve translation quality. Experimental results are reported in the French-to-English IWSLT 2010 evaluation where our system was ranked 3rd out of nine systems.