The state configuration structure sets parameters used to make situational predictions.
- The number of categories can be expanded by specifying the 32-category compile option (
NJ_OPT_STATE_TYPE2
).
Type |
Categories |
State Control |
Description of the Element Number of the Standard State Setting Array |
Application |
iWnn |
Input Field |
Person's name |
○ |
- |
NJ_CATEGORY_FIELD_PERSON sets the priority of people's names. This is used with name input fields, for example. |
Noun concatenation |
NJ_CATEGORY_FIELD_NOUN prioritizes nouns. This is used in situations such as filename input. |
Start of sentence |
○ |
NJ_CATEGORY_FIELD_HEAD prioritizes words that come at the start of sentences. |
Kinds of Expressions |
Colloquial Expressions |
○ |
○ |
NJ_CATEGORY_EXPRESSION_COLLOQUIAL |
Written Expressions |
NJ_CATEGORY_EXPRESSION_WRITTEN |
Time Expressions |
Time period: Morning |
○ |
○ |
NJ_CATEGORY_TIME_MORNING |
Time period: Noon |
NJ_CATEGORY_TIME_NOON |
Time period: Night |
NJ_CATEGORY_TIME_NIGHT |
Relative time: Past |
- |
○ |
NJ_CATEGORY_TIME_PAST |
Relative time: Future |
NJ_CATEGORY_TIME_FUTURE |
Feelings |
Positive Feelings |
○ |
○ |
NJ_CATEGORY_FEEL_PLUS |
Negative Feelings |
NJ_CATEGORY_FEEL_MINUS |
Months |
Month: January |
○ |
○ |
NJ_CATEGORY_MONTH_JANUARY |
Month: February |
NJ_CATEGORY_MONTH_FEBRUARY |
Month: March |
NJ_CATEGORY_MONTH_MARCH |
Month: April |
NJ_CATEGORY_MONTH_APRIL |
Month: May |
NJ_CATEGORY_MONTH_MAY |
Month: June |
NJ_CATEGORY_MONTH_JUNE |
Month: July |
NJ_CATEGORY_MONTH_JULY |
Month: August |
NJ_CATEGORY_MONTH_AUGUST |
Month: September |
NJ_CATEGORY_MONTH_SEPTEMBER |
Month: October |
NJ_CATEGORY_MONTH_OCTOBER |
Month: November |
NJ_CATEGORY_MONTH_NOVEMBER |
Month: December |
NJ_CATEGORY_MONTH_DECEMBER |
By default, when executing learning, the state bias value of categories that include words to learn are all incremented by one by iWnn, except in the case of input field type states. In addition, the bias value is always initialized to
0
for the start of sentence category when learning words.
You can retrieve candidates prioritizing those suited to the state by having the application check for changes in the bias value during learning and updating the state setting accordingly. For example, if the past attribute has been incremented by one, words from the past category can be prioritized during the next prediction conversion by setting +30 for the past category and -800 for the future category. This moves words in the future category to the bottom of the candidate list.
The state setting bias value appropriate for using standard dictionary frequency values can be updated automatically inside iWnn by setting the compile option
NJ_ADD_STATE_TYPE2
.
The bias value is added to the candidate frequency value when determining the priority of candidates. For example, 100 is added to the usual frequency value for a candidate in a category with a bias value of +100.
The frequency value of the candidate varies in the range from 0 to 1000. If the value goes over the upper limit or under the lower limit, it is clamped at either 1000 or 0. In other words, a candidate in a category with a bias value of +1000 always has a maximum frequency value (1000) and has priority over learning dictionary candidates. Also, a candidate with a bias value of -1000, always has the minimum frequency value (0) and the lowest priority.