Error Detection and Prevention With Crf

¶ … devising a strategy for detecting errors on a CRF, what are the benefits of strict detection parameters (high sensitivity) versus more relaxed parameters (high specificity)? Even though small ASR errors might not impact downstream processes that make use of the transcript, larger error segments like those generated by OOVs can have a considerable impact on applications such as speech-to-speech translation and can eventually lead to communication failure between users of the system.

The most significant error segment is passed to a dialog system for interactive recovery in which rephrased words are reinserted in the original. However benefits can be derived from both a high sensitivity strategy versus a high specificity strategy. A high sensitivity strategy is one that is able to detect more errors in a broader manner than that of a high specificity approach. With high sensitivity, errors that are detected are more likely to be false positives.

They system may elect to detect the error even though the error was detected incorrectly. High sensitivity allows for more potential errors to be detected at the expense of accuracy or false positives. With high specificity, errors are detected at more of a granular level. These errors are specific to the CRF and often have a higher probability of being correct. With specificity however, errors that should be detected, may go unnoticed as the system focuses solely a narrow range of errors.

In this instance, the errors detected are of higher quality and are more accurate. However, they are not as broad based as those detected by a system that emphasized high sensitivity. How would the type of variable influence your decision? How critical the errors being detected are to the overall operation of the system is a critical variable that can influence the decisions of others. Errors that can potentially cause massive amounts of damage are different than many routine errors that cause minor inconveniences.

Therefore, the severity of not detecting the error is paramount in the decision making process. Costs are also a large component that can influence an individual's decision. Cost includes not only the physical systems, processes, and personally to properly monitor them, but also the opportunity cost associated with them. Depending on the decision made, a team will forego the opportunity to use their resources in other endeavors. For example, there is an opportunity cost of electing a high sensitivity as oppose to electing a high specificity.

The firm must be prepared to bare these hidden costs in the decisions they make. Innovation and nimbleness are also variables that need consideration as well. Innovation, particularly in the technology field is rampant. How quickly a strategy and processes can be altered is also critical. Within the decision making process, unexpected events undoubtedly occur. The variable of nimbleness is important so that decisions can be changed if the end result is not what is expected. How would the error detection strategies differ between paper and electronic.