Institutional Research Information Service
UCL Logo
Please report any queries concerning the funding data grouped in the sections named "Externally Awarded" or "Internally Disbursed" (shown on the profile page) to your Research Finance Administrator. Your can find your Research Finance Administrator at https://www.ucl.ac.uk/finance/research/rs-contacts.php by entering your department
Please report any queries concerning the student data shown on the profile page to:

Email: portico-services@ucl.ac.uk

Help Desk: http://www.ucl.ac.uk/ras/portico/helpdesk
Publication Detail
Training Datasets for Machine Reading Comprehension and Their Limitations
  • Publication Type:
  • Authors:
    Welbl J
  • Date awarded:
  • Awarding institution:
    UCL (University College London)
  • Language:
Neural networks are a powerful model class to learn machine Reading Comprehen- sion (RC), yet they crucially depend on the availability of suitable training datasets. In this thesis we describe methods for data collection, evaluate the performance of established models, and examine a number of model behaviours and dataset limita- tions. We first describe the creation of a data resource for the science exam QA do- main, and compare existing models on the resulting dataset. The collected ques- tions are plausible – non-experts can distinguish them from real exam questions with 55% accuracy – and using them as additional training data leads to improved model scores on real science exam questions. Second, we describe and apply a distant supervision dataset construction method for multi-hop RC across documents. We identify and mitigate several dataset assembly pitfalls – a lack of unanswerable candidates, label imbalance, and spurious correlations between documents and particular candidates – which often leave shallow predictive cues for the answer. Furthermore we demonstrate that se- lecting relevant document combinations is a critical performance bottleneck on the datasets created. We thus investigate Pseudo-Relevance Feedback, which leads to improvements compared to TF-IDF-based document combination selection both in retrieval metrics and answer accuracy. Third, we investigate model undersensitivity: model predictions do not change when given adversarially altered questions in SQUAD2.0 and NEWSQA, even though they should. We characterise affected samples, and show that the phe- nomenon is related to a lack of structurally similar but unanswerable samples during training: data augmentation reduces the adversarial error rate, e.g. from 51.7% to 20.7% for a BERT model on SQUAD2.0, and improves robustness also in other settings. Finally we explore efficient formal model verification via Interval Bound Propagation (IBP) to measure and address model undersensitivity, and show that using an IBP-derived auxiliary loss can improve verification rates, e.g. from 2.8% to 18.4% on the SNLI test set.
Publication data is maintained in RPS. Visit https://rps.ucl.ac.uk
 More search options
UCL Researchers
Dept of Computer Science
University College London - Gower Street - London - WC1E 6BT Tel:+44 (0)20 7679 2000

© UCL 1999–2011

Search by