Multifarm is a well-known comprehensive dataset for multilingual ontology matching evaluation.We extend the Multifarm dataset in the eleventh language, i.e., Hindi ( 11). This Hindi component of 111 has been created by translating the entities using the Google translation service, validating manually, and then creating the reference alignments for the matching task. Work is in progress to determine the impact of 11 on diferent multilingual ontology alignment systems. The complexities introduced by the Hindi language will introduce novel challenges to the behavior of cross-lingual ontology matching systems.
1. Introduction and Motivation Hindi with an aim to include the same during the OAEI ontology matching workshops. After Chinese, Hindi has more characters than any other available language in the Multifarm track. For any Computer system to take contextually correct decisions and serve the vast population that speaks Hindi, we need more datasets in it that are helping one or more type of computer operations (in this case, Ontology matching). Many letters (and even words) in English has various mappings in Hindi. For example, ’S’ can be written as श, ष or स. Hindi has no capitalization, though have short vowels. Version and variation of word in Hindi is context dependent. Same word in English (or any other languages) can have diferent Hindi associated word, depending on the context. 2. The
11 Benchmark The Hindi Language Component is developed referring to the multifarm dataset of the OAEI campaign. The structure of the ontologies and reference alignments has been reused while enriching them with the contextually verified entities in Hindi language. After including Hindi, a total of 55 language pairs will be there for the evaluation of the matching systems. 1. Translation of Ontology Entities - A total of around 2500 terms were fetched from the seven ontologies of the dataset and enlisted, out of which around 1000 are unique.
Translations of the entities were done with the Google translation service. 2. Validation - Contextual verification was performed in line with conference domain. Errors like ’paper’, which was translated to ’कागज़’, which was contextually corrected to ’शोध पत्र . The task requires validators to have knowledge of conference domain and Hindi Language. The authors are well suited for the task, they all are aware of conference domain as they all are researchers and for Hindi, two of them are native Hindi speakers. 3. Generation of Reference Alignments - The reference alignments were created by reproducing the alignments based on the reference alignments available in the multifarm track. For example, dokument (of cmt ontology) in German is aligned to समम्ेलन दतस्ावेज़ in Hindi (of conference ontology).
Acknowledgements This work is supported by the IHUB-ANUBHUTI-IIITD FOUNDATION set up under the NMICPS scheme of the Department of Science and Technology, India