next up previous
Next: Example: nlp_speech method. Up: Weights Previous: Weights

Example: speech method.

As a trivial example of such a method, consider the problem of finding the best path through the word-graph ignoring all aspects but the acoustic scores present in the word-graph. In order to implement a method $W^{\mbox{speech}}$ to solve this problem, we define weights using a unary tuple $\langle c\rangle$. The initial weight is $\mbox{\it ini\/}=\langle 0\rangle$ and $\mbox{\it uw\/}$ is defined as follows:
\mbox{\it uw\/}(\langle c\rangle,(v_i,v_j,w,a,q)) =
\left\{ ...
...e c+a\rangle & \mbox{for stopping edges}\\
\end{displaymath} (31)

Note that we specify an infinite weight for category edges because we want to ignore such edges for this simple method (i.e. we are simply ignoring the results of the parser). We define an ordering $\prec$ on such tuples, simply by stating that $\langle w\rangle \prec \langle w'\rangle$ iff w < w'.