Limitations of Linguistic Analysis

There are some limitations to linguistic analysis to arrive at conclusions that may make assumptions that are probabilistic rather than deterministic. This is touched upon briefly in the textbook under the section "Data Analysis" under the discussion on Centering Resonance Analysis (CRA, page 461), but the caveats seem to be missing.

One example I can think of where linguistic analysis gave wrong results that did not agree with the rest of the evidence is in the case of "Forensic Linguistics" that was used in the aftermath of the Anthrax attacks. Donald Foster was a forensic linguist who used the written messages in the Anthrax letters to identify the perpetrator of the attacks. Foster claimed that a virologist, Dr. Steven Hatfill, was a possible suspect after noting some similarities in language used by Hatfill to the words in the Anthrax letters. After an exhaustive investigation, Hatfill was declared to be innocent, and even successfully sued the Government for millions of dollars. This was a case where a large amount of linguistic evidence suggested a conclusion (Hatfill's guilt) that was later proven to be false.