Automated directory assistance system - from theory to practice

Semantic Frame‐Based Spoken Language Understanding

“…The contents of the field, such as business or product names, are often unstructured text. For example, directory assistance (DA) (Bacchiani et al 2008;Yu et al 2007) is one of the most popular voice search applications, in which users issue a spoken query and an automated system returns the phone number and address information of a business or an individual. The applications include both telephone only services and multi-modal services on mobile devices.…”

“…While automating residential DA is important in reducing the operational cost, it is only a small portion (19%) of the total received calls compared to the 61% of business DA calls (Billi et al 1998). Therefore there have been increasing interests in business DA recently, with the commercial deployments from Tellme (acquired by Microsoft), Jingle Networks, AT&T, Google (Bacchiani et al 2008), Verizon and Cingular (merged with AT&T Wireless now) and Microsoft (Yu et al 2007). Because the level of linguistic variance is much higher in business DA queries, SLU/search aiming at correctly interpreting user's intent becomes an important research topic.…”

“…However, the technology is often applicable to other applications as well. For example, the product/business rating systems (Zweig et al 2007) directly used the technology developed in a DA application (Yu et al 2007).…”

“…Another approach to improved the robustness is via statistical n-gram models (Bacchiani et al 2008;Natarajan et al 2002;Yu et al 2007). An n-gram model is more robust because it does not require a user's utterance to match a rule exactly; because it provides a statistical framework for fair comparison between different hypotheses; and because it has well-studied smoothing algorithms to estimate the likelihood of unseen events more accurately.…”

Benchmarking Deep Reinforcement Learning for Continuous Control

“…Benchmarks have played a significant role in other areas such as computer vision and speech recognition. Examples include MNIST (Le-Cun et al, 1998), Caltech101 (Fei-Fei et al, 2006, CI-FAR (Krizhevsky & Hinton, 2009), ImageNet (Deng et al, 2009), PASCAL VOC (Everingham et al, 2010), BSDS500 (Martin et al, 2001), SWITCHBOARD (Godfrey et al, 1992), TIMIT (Garofolo et al, 1993), Aurora (Hirsch & Pearce, 2000), and VoiceSearch (Yu et al, 2007). The lack of a standardized and challenging testbed for reinforcement learning and continuous control makes it difficult to quantify scientific progress.…”