Timothy Miller's work in the field of clinical natural language processing (NLP) has covered a broad array of applications, from clinical research-enabling phenotyping applications as part of the i2b2 center for biomedical computing, to semantic processing of clinical texts, to core contributions to NLP and machine learning. A major thread that ties all this work together is an interest in the value of syntax. He has been responsible for syntactic contributions in temporal relation extraction (Lin etal, 2014, Miller et al, 2013 and Miller et al, in preparation), UMLS relation extraction (Dligach et al, 2013), coreference resolution (Miller et al, 2012, Zheng et al, 2012), and negation detection (Miller et al, in preparation). This also includes contribution of code to open source projects Apache cTAKES (clinical Text Analysis and Knowledge Extraction System) and ClearTK. In cTAKES he developed a constituency parser module, and contributed syntactic features to all the relation extraction modules. In ClearTK he contributed java tree kernel code (part of their version 2.0 release) that dramatically improves tree kernel learning, and enables new kernel development. This code was the backbone for a new kernel (Descending Path Kernel) described
in Lin et al. (2014).
Despite these advances, he is struck by the diversity in clinical sub-domains and how this affects performance. He has been involved with several clinical language annotation projects, and has been lucky enough to be able to use these syntactic and semantic annotations. However, the difficulty of distributing clinical data and the differences between domains will limit the applicability of methods developed on only one corpus. Timothy saw first hand evidence of this by working on different coreference corpora (ODIE and i2b2 Challenge), where performance suffered greatly between corpora. As a result, he has come to be interested in approaches that make use of unsupervised structure learning and world knowledge extraction.


Publications powered by Harvard Catalyst Profiles

  1. Clinical Natural Language Processing for Radiation Oncology: A Review and Practical Primer. Int J Radiat Oncol Biol Phys. 2021 07 01; 110(3):641-655. View abstract
  2. Rethinking domain adaptation for machine learning over clinical language. JAMIA Open. 2020 Jul; 3(2):146-150. View abstract
  3. Does BERT need domain adaptation for clinical negation detection? J Am Med Inform Assoc. 2020 04 01; 27(4):584-591. View abstract
  4. Supervised methods to extract clinical events from cardiology reports in Italian. J Biomed Inform. 2019 07; 95:103219. View abstract
  5. Towards generalizable entity-centric clinical coreference resolution. J Biomed Inform. 2017 05; 69:251-258. View abstract
  6. Multilayered temporal modeling for the clinical domain. J Am Med Inform Assoc. 2016 Mar; 23(2):387-95. View abstract
  7. Automatic identification of methotrexate-induced liver toxicity in patients with rheumatoid arthritis from the electronic medical record. J Am Med Inform Assoc. 2015 Apr; 22(e1):e151-61. View abstract
  8. ClinicalTrials.gov as a data source for semi-automated point-of-care trial eligibility screening. PLoS One. 2014; 9(10):e111055. View abstract
  9. Automatic prediction of rheumatoid arthritis disease activity from the electronic medical records. PLoS One. 2013; 8(8):e69932. View abstract
  10. A system for coreference resolution for the clinical narrative. J Am Med Inform Assoc. 2012 Jul-Aug; 19(4):660-7. View abstract