Search

In this paper, a public dataset for Offline Handwriting Recognition, along with an appropriate evaluation method to provide benchmark indicators at sentence level, is presented. This dataset, called SPA-Sentences, consists of offline handwritten Spanish sentences extracted from 1617 forms produced by the same number of writers. A total of 13,691 sentences comprising around 100,000 word instances out of a vocabulary of 3288 words occur in the collection. Careful attention has been paid to make the baseline experiments both reproducible and competitive. To this end, experiments are based on state-of-the-art recognition techniques combining convolutional blocks with one-dimensional Bidirectional Long Short Term Memory (LSTM) networks using Connectionist Temporal Classification (CTC) decoding. The scripts with the entire experimental setting have been made available. The SPA-Sentences dataset and its baseline evaluation are freely available for research purposes via the institutional University repository. We expect the research community to include this corpus, as is usually done with English IAM and French RIMES datasets, in their battery of experiments when reporting novel handwriting recognition techniques. [ABSTRACT FROM AUTHOR]

The purpose of this paper is to provide guidelines for building a word alignment evaluation scheme. The notion of word alignment quality depends on the application: here we review standard scoring metrics for full text alignment and give explanations on how to use them better. We discuss strategies to build a reference corpus, and show that the ratio between ambiguous and unambiguous links in the reference has a great impact on scores measured with these metrics. In particular, automatically computed alignments with higher precision or higher recall can be favoured depending on the value of this ratio. Finally, we suggest a strategy to build a reference corpus particularly adapted to applications where recall plays a significant role, like in machine translation. The manually aligned corpus we built for the Spanish-English European Parliament corpus is also described. This corpus is freely available. [ABSTRACT FROM AUTHOR]