The analysis of sense frequency distributions presented in this article provides an explanation for the results of Schutze and Pedersen  whose use of a disambiguator on large queries and documents resulted in a 7-14% improvement in retrieval effectiveness, the first published results showing an automatic disambiguator working successfully with an IR system.
To understand the reasons for their results, which apparently contradict those presented here, it is necessary to first explain how Schutze and Pedersen's disambiguator worked.
2) Ng and Lee stated that the error rate of their disambiguator was 30%.
Gale, Church, and Yarowksy introduced and tested a disambiguator using pseudowords in a 1992 paper [Gale et al.
Early disambiguators were based on hand-built rule sets and only worked over a small number of words and senses [Small and Rieger 1982; Weiss 1973].
This means, despite the fact that the majority of the instances show at least one formal disambiguator (86.
It is commonly assumed in such studies that morphological and syntactic phenomena are used as disambiguators that are expected to trigger revision of the analysis pursued so far (see, e.
In Swedish, this holds for all the word order phenomena presented as disambiguators so far.
When there are no formal disambiguators in the sentence, the thematic roles can only be assigned by utilizing word meanings and real world knowledge.
Summing up, then, we contend that the morphological and syntactic factors listed in this section have an impact on the process of decoding and that calling them disambiguators is therefore justified.
More specifically, the following topics will be discussed: (a) do formal disambiguators affect the occurrence of OVS word order (Section 2)?
If formal markers have an impact on the production of utterances, one can expect them to be differently distributed in the two clause categories, that is, formal disambiguators should be more frequent in declaratives than in relatives as the need of structural disambiguation can be assumed to be greater where the word order does not enable the syntactic functions to be assigned with absolute certainty.
In order to explore the dependence between formal disambiguators and clause type, an attempt was made to perform a statistical analysis involving the following factors: clause type, subject morphology, verbal frame, negator, verbal particle, and preposition stranding.
Since NP is the only variable of the three that survived statistical testing, the interdependence between reversibility and formal disambiguators is weak as a whole.
What is meant by this notion is that an utterance can be interpreted on purely semantic grounds, without resorting to formal disambiguators.