Crowdsourcing-based Annotation of the Accounting Registers of the Italian Comedy

Crowdsourcing-based Annotation of the Accounting Registers of the Italian Comedy Adeline GRANET1, Benjamin HERVY1, Geoffrey ROMAN-JIMENEZ1, Marouane HACHICHA1, Emmanuel MORIN1, Harold MOUCH `ERE1, Solen QUINIOU1, Guillaume RASCHIA1, Franc¸oise RUBELLIN2, Christian VIARD-GAUDIN1 1LS2N, UMR CNRS 6004, Universit´e de Nantes, France 2L’AMO, EA 4276, Nantes Universit´e de Nantes, France {firstname.lastname}@univ-nantes.fr Abstract In this paper, we present a double annotation system for new handwritten historical documents. We have 25,250 pages of registers of the Italian Comedy of the 18th century containing a great variety and amount of information. A crowdsourcing platform has been set up in order to perform labeling and transcription of the documents. The main purpose is to grasp budget data from the all 18th century and to create a dedicated database for the domain’s experts. In order to improve, help and accelerate the process, a parallel system has been designed to automatically process information. We focus on the titles field, segmenting them into lines and checking candidate transcripts. We have collected a base of 971 title lines. Keywords: handwriting recognition, crowdsourcing, historical data 1. Introduction The CIRESFI project1 sets out to reassess a theatrical her- itage that has often been considered as inferior to that of the two major, royally-privileged theaters (the Opera and the French Comedy). By studying the theater that was ex- cluded from the system of privileges, the Italian Theater, we analyze the questions of acculturation and institution- alization including the fusion of the Opera-Comique with the Italian Comedy. By employing a mass of untapped and unpublished resources (27,544 pages of registers available at the BnF2), this program will take a decidedly fresh look at emerging forms of creation and the changes in the enter- tainment economy. To this end, CIRESFI takes up a tech- nological challenge such as creating a tool for handwriting recognition and create an interactive database. Information retrieval is tedious in old handwritten docu- ments for humanities and social science researchers. The digitization of collections facilitates their consultation but it is necessary to extract information to make them fully exploitable. Sometimes, handwriting recognition systems could be set up but they require to be trained on an existing ground truth. Within the CIRESFI project, this assumption is not fulfilled since the corpus is very large, diverse and quite new w.r.t. the information extraction task. Crowd- sourcing allows the use of volunteers to annotate historical documents according to our needs: (i) indicate the type of a page; (ii) detect the different areas and identify informa- tion type; (iii) transcribe as required; (iv) validate or correct previous transcriptions. In order to support and boost this label-and-transcribe process, we have implemented an au- tomatic annotation enrichment method. The detected areas are segmented into lines. Then, the transcripts associated with these zones, if any, are reworked and validated manu- 1French National ANR-14-CE31-0017 program. 2Biblioth`eque Nationale de France. Digital library: http: //gallica.bnf.fr/accueil/ ally. Figure 1 illustrates this approach. 2. Registers of Italian Comedy The studied documents are registers of the Italian Comedy which record the daily, monthly and annual …

Content truncated to 500 words