Transcripts of spontaneous human speech present a significant obstacle for traditional NER models. The lack of grammatical structure of spoken utterances and word errors introduced by the ASR make downstream NLP tasks challenging. In this paper, we examine in detail the complex relationship between ASR and NER errors which limit the ability of NER models to recover entity mentions from spontaneous speech transcripts. Using publicly available benchmark datasets (SWNE, Earnings-21, OntoNotes), we present the full taxonomy of ASR-NER errors and measure their true impact on entity recognition. We find that NER models fail spectacularly even if no word errors are introduced by the ASR. We also show why the F1 score is inadequate to evaluate NER models on conversational transcripts.

Recent work raises concerns about the use of standard splits to compare natural language processing models. We propose a Bayesian statistical model comparison technique which uses k-fold cross-validation across multiple data sets to estimate the likelihood that one model will outperform the other, or that the two will produce practically equivalent results. We use this technique to rank six English part-of-speech taggers across two data sets and three evaluation metrics.

Natural language processing of conversational speech requires the availability of high-quality transcripts. In this paper, we express our skepticism towards the recent reports of very low Word Error Rates (WERs) achieved by modern Automatic Speech Recognition (ASR) systems on benchmark datasets. We outline several problems with popular benchmarks and compare three state-of-the-art commercial ASR systems on an internal dataset of real-life spontaneous human conversations and HUB’05 public benchmark. We show that WERs are significantly higher than the best reported results. We formulate a set of guidelines which may aid in the creation of real-life, multi-domain datasets with high quality annotations for training and testing of robust ASR systems.