People stuff their resumes with lots of irrelevant or less pertinent information. Most software developer resumes are stuffed with details on the company or the projects they have worked on which have all kinds of skills. The actual work that the person may have done may have been a small part of the project so they only used a few actual skills. However, the parser will not be able to differentiate and will pick all the skills whether the skill belongs to that of the company or the project or the person. This is misleading.
Contextual reference can be obtained with semantic rules but such rules can run into millions. For example we could write a rule to normalise the following:
I am into Project Management (skill = Project Management)
I have been managing projects (skill = Project Management)
My activities included managing people, finance and projects (skill = Project Management)
I have managed projects (skill = Project Management)
I manage finance and projects (skill = Project Management)
Pronouns themselves are easy to capture, but pronouns are actually not skills. Pronouns with verbs become a skill. Oracle itself is not a skill. Oracle administration, oracle development, oracle support, selling oracle, providing training in oracle are skills. This is the problem of issues with verbs (stated above) coming into play.
In areas where nouns and pronouns are less relevant as in IT, for example, and where verbs are more common like in management, parsing faces even higher difficulty.
While these may be problems related to semantics we have other problems that also need to be addressed particularly when we want to improve efficiency in talent development and talent management.