🧑‍🔬 Burr Settles

Burr Settles works in machine learning for natural language processing. The work centres on training statistical models when labelled examples are scarce or costly to obtain. The papers here deal with supervised learning from text. One asks how a learner might pick which instances deserve human annotation. The other builds a sequence labeller for biomedical literature, within the wider area of named entity recognition. The methods include conditional random fields, engineered features and lexicons, and the systematic organisation of prior research.

The most cited work is the Active Learning Literature Survey (2009). This technical report reviews approaches in which the model chooses its own queries to an annotator, aiming for good accuracy with far fewer labels. It arranges the existing literature rather than reporting new experiments. The earlier Biomedical Named Entity Recognition using Conditional Random Fields and Rich Feature Sets (2004) was prepared for the BioNLP/NLPBA 2004 shared task. It tags proteins, DNA, RNA, cell lines and cell types in MEDLINE-style abstracts using a linear-chain CRF. The paper compares a model that relies only on spelling and context with one that adds several further resources: semantic lexicons, class keyword lists selected by a χ² test, and a gazetteer grown from a handful of seed terms.

Across their papers here, the direction runs from a single applied tagging task toward the general problem of reducing annotation effort, with the 2009 survey as the newest work. Bylines connect Settles with the University of Wisconsin–Madison, matching the years of both papers, and later with Carnegie Mellon University and Duolingo in Pittsburgh.

Burr Settles has 2 papers in our library, in 2004–2009, most of them on Named entity recognition.