Nt1310 Unit 4 Factor Analysis

The first factor template in Equation (3.1) models the co-occurrence between input variables and output variables. Moreover, the factor in the unigram factor template are defined as: φ_u (y_t,x_t;_t)= exp ∑_k^(K(T))▒〖_TK f_TK (y_t,x_t)〗 (3.2)
On the other hand, the second factor template in Equation (3.1) models the co-occurrence between contiguous output variables. The factors in the bigram factor template are defined as: φ_b (y_t,y_(t-1);_)=exp ∑_k^(K())▒〖_K f_K (y_t,y_(t-1))〗 (3.3)
As we can notice from equation (3.2) and equation (3.3), the two factor templates used different sets of feature functions and affected two different parameters. This is one of the advantages of using CRF because it allows the designer of the CRFs to model different “aspects” of the distribution in

In a simple manner, it is liked saying “If the label of the current token is NN given the current token is an occurrence of the word guitar, then return 1, otherwise return 0.”. Another example feature function of the type demonstrated in equation (3.4) is the following: f_(NN,JJ) (y_t,y_(t-1) )= {█(1 if y_t=NN and y_(t-1)=JJ@0 otherwise)┤ (3.7) where JJ is POS tag denoting an adjective; the meaning of this feature function is straightforward. In summary, the feature function demonstrated in equation (3.3) model the relation that occurs between the input variable of type d at position t in the sequence and an output variable, whereas the feature function demonstrated in equation 3.4 model instead the relation that appears between the output variable at position t in the sequence and the output variable at position t –