How Diachronic Text Corpora Affect Context based Retrieval of OOV Proper Names for Audio News

How Diachronic Text Corpora Affect Context based Retrieval of OOV Proper Names for Audio News Imran Sheikh, Irina Illina, Dominique Fohr Universit´e de Lorraine, LORIA, UMR 7503, Vandoeuvre-l`es-Nancy, F-54506, France Inria, Villers-l`es-Nancy, F-54600, France CNRS, LORIA, UMR 7503, Vandoeuvre-l`es-Nancy, F-54506, France {imran.sheikh, irina.illina, dominique.fohr}@loria.fr Abstract Out-Of-Vocabulary (OOV) words missed by Large Vocabulary Continuous Speech Recognition (LVCSR) systems can be recovered with the help of topic and semantic context of the OOV words captured from a diachronic text corpus. In this paper we investigate how the choice of documents for the diachronic text corpora affects the retrieval of OOV Proper Names (PNs) relevant to an audio document. We first present our diachronic French broadcast news datasets, which highlight the motivation of our study on OOV PNs. Then the effect of using diachronic text data from different sources and a different time span is analysed. With OOV PN retrieval experiments on French broadcast news videos, we conclude that a diachronic corpus with text from different sources leads to better retrieval performance than one relying on text from single source or from a longer time span. Keywords: oov, proper names, lvcsr, diachronic corpus 1. Introduction News content is diachronic in nature and is characterised by different topics which change with time, leading to a change in the linguistic content and vocabulary. As a re- sult, a typical problem faced by Large Vocabulary Contin- uous Speech Recognition (LVCSR) systems processing di- achronic audio news is Out-Of-Vocabulary (OOV) words. In previous works it has been observed that the majority of OOV words are Proper Names (PNs). PN percentage in OOV words been reported as: 56% by Qin (2013), 66% by Parada (2011), 57.6% by Palmer (2005), 70% by Al- lauzen (2005), 72% by Bechet (2000). On the other hand, PNs in audio news are of prime importance for content based indexing and browsing applications. Our work is closely related to recognition and recovery of OOV PNs i.e., PNs which appear in diachronic audio but are not present in the LVCSR vocabulary and cannot be recognised by the LVCSR system. To recognise OOV PNs in a test audio document, we rely on new PNs extracted from collections of diachronic text news from the internet, referred to as a diachronic corpus. Instead of using all the new PNs to recover the target OOV PNs1, it would be efficient to incorporate only those new PNs which are contextually related. We refer to this task as retrieval of OOV PNs relevant to an audio document. To achieve this, we use the topic and lexical context of the au- dio document and OOV PNs. In our previous works, we have presented different methods exploiting topic context to retrieve OOV PNs (Sheikh et al., 2015a) and we have further discussed topic model variations and methods to handle the less frequent OOV PNs (Sheikh et al., 2015b). 1Ideally new PNs extracted from collections of diachronic text news are OOV PNs with respect to the LVCSR vocabulary. How- ever, all new PNs are …

Content truncated to 500 words