Thank you Peter and Tim, your responses were very helpful. On Tue, Aug 20, 2019, 5:01 PM Peter Abramowitsch <pabramowit...@gmail.com> wrote:
> Hi Jeff > > I've experimented with three approaches. > > One is with the LabValueFinder which is included in the cTakes release - > that looks specifically for values associated with LabMentions. It also > has an "eager" mode where it converts some MedicationMentions into > LabMentions, when the context seems right. O2, Sodium etc. I can't say > it works all that well and it is not capable of many different semantic > forms of the Name/Value association. It is also too eager.. sometimes > creating LabMentions out of Medications when it shouldn't. > > Another approach was to use something like Stanford's TokensRegex, that > allows you to construct regex-like rules where the segments are not strings > but Tokens, where you can query the attributes like POS, and NER . For > Ctakes I had to adapt a UIMA package that must have been someone's thesis > project from the university of Nantes. > > Copyright 2015 - CNRS (Centre National de Recherche Scientifique) > package fr.univnantes.lina.uima.tkregex > > What I have is not ready for prime time and is still very rough. It works > well but only for a limited set of rules > > I used it to create a vitals detector. Here's a snippet of the rules that > this package loads in at runtime, that creates an annotation called WGT > given these matchers > matcher NUM: [ postag == "CD" ]; > matcher BE: [ lemma == "be" | lemma == "at"]; > matcher WT: /(?i)^wt|^weight/; > matcher WUOM: /(?i)^kg|^lb|^pounds/; > term "WGT": WT BE? SYM? NUM WUOM; > > The last approach was a home-built mechanism using the ConllDependencyNode > collection and the RelationArguments to detect the same connection between > certain typed pairs of Identified annotations. > > Problem is. I've always been in prototyping mode and never had time to push > these methods to production ready status > > Peter > > On Tue, Aug 20, 2019 at 1:15 PM Jeffrey Miller <jeff...@gmail.com> wrote: > > > Hi, > > > > Is there any configuration or component in cTAKES that can be used to > > attribute a measurement annotation to another annotation that it applies > > to? For example, for "2 mm incision" where we relate "2 mm" to > "incision"? > > It looks like there might be a roundabout way to find the head of the > span > > of the MeasurementAnnotation in the output of the dependency parser, but > I > > was wondering if this has been explored before? Perhaps the > > RelationExtractor component? > > > > I also have another more general question if anyone can help- how does > the > > structure of the cTAKES type system effect how cTAKES works? I am looking > > for a general intuition of how the structure of the typesystem drives the > > larger cTAKES architecture? > > > > Thanks! > > Jeff > > >
