TrainContext 

1.0     2.0     

Provides the recognizer a previous recognition result that may contain errors, plus the correct interpretation of the raw data.

Note  This function is provided only for compatibility with version 1.0 of the Pen API and will not be supported in future versions. Use TrainHREC1AOCU_ instead.

BOOL TrainContext( LPRCRESULT lprcresult, LPSYE lpsye, int csye, LPSYC lpsyc, int csyc )

Parameters

lprcresult

Address of the RCRESULT232HBID structure containing the handle to the pen data that contains the raw data and the recognizer s original interpretation of that data. This parameter cannot be NULL.

lpsye

Address of an array of SYE3BE5AM5 structures that specify the correct interpretation of the raw data. The values of the  iSyc members of these structures index the SYC1AOCY0 structures pointed to be the lpsyc parameter.

csye

The number of SYE structures in the lpsye array.

lpsyc

An array of SYC structures that establish the mapping between the raw data and the characters in the hpendata member of the structure pointed to by the lprcresult parameter.

csyc

The number of SYC structures in the lpsyc array.

Return Value

Returns TRUE if the ink is accepted for training; otherwise, returns FALSE.

Comments

TrainContext is called by an application with a recognition result that may contain mistakes, along with a correct interpretation, so that the recognizer can learn from the mistake and improve subsequent recognition. A second, simpler training function for 1.0 recognizers is provided by TrainInk232HBEC.

TrainContext internally calls the function TrainContextInternal exported by the recognizer identified by the hrec member of the RC232HBHC structure pointed to by the lprc member of the RCRESULT232HBID structure. A version 1.0 recognizer should export both TrainContextInternal and TrainInkInternal, but can simply return FALSE from both functions if the recognizer does not support this type of training.

When a training application is able to provide contextual information (such as segmentation suggestions) to the version 1.0 recognizer, it calls the TrainContext function. The lprcresult parameter points to an RCRESULT structure that contains the results of a previous recognition. The raw data is also contained in the hpendata member of the structure pointed to be lprcresult.

In addition to providing the incorrect interpretation of the data (by means of the symbol graph, the lpsyv member in the RCRESULT structure), a more detailed, correct interpretation is also provided by the SYE3BE5AM5 structures and SYC1AOCY0 structures. Because the correct interpretation is passed by SYE structures, it is possible to suggest segmentation boundaries to the recognizer.

Suppose, for example, that a user writes  lc,  and the recognizer interprets it as  k . A trainer calls TrainContext using, first, an array of SYC structures that point to the ink of the  lc  and, second, the two SYE structures with the SYV values  l  and  c . These two SYE structures share the same index into the lpsyc array, indicating that both use the ink that was interpreted as  k .

Segmentation errors can be corrected in the other direction as well. Suppose, for example, the user writes  k  and the recognizer interprets it as  lc . A trainer could call TrainContext using a single SYE with SYV values equal to  k  and an array of SYC structures that incorporate the ink the recognizer had previously assigned to the  l  and the  c .

To train several SYV symbol values to a single piece of ink (for example, a long stroke that is an  he  ligature), there will be two consecutive SYE3BE5AM5 structures one for the  h  and one for the  e . Both SYE structures have the same iSyc member; this means that the SYE structures both point to the same ink. A recognizer must take this into consideration to avoid training the two characters separately using the same ink for both; that would result in having  he  trained as  he he .

A recognizer can supply its own custom training dialog boxes. An application should check whether the recognizer supports custom training by calling ConfigRecognizer232HADD with the WCR_TRAIN subfunction.

The trainer does not display an error message if TrainInk232HBEC or TrainContext returns FALSE. Error messages that occur when training fails must be handled by the recognizer.

See Also