Three new Dynamic Programming (DP) algorithms have been proposed for connected-word recognition: constant time delay, O(n) DP, and augmented continuous DP matching algorithms.

The constant time delay DP algorithm is the same as the two level DP algorithm [1], except that local distance computations are reduced by a factor by executing computations after a constant time delay (every W frames) and by increased storage for local distances.

The O(n) DP algorithm is the same as the Bridle algorithm [2] for variable length connected-word recognition. The algorithm provides one pass recognition of connected words and is computationally the least expensive of all three algorithms proposed in the paper. However, the O(n) DP algorithm cannot perform syntax driven recognition while the other two can. The O(n) DP algorithm can be modified to provide unconstrained endpoint recognition by skipping reference frames, or test frames, or both. Another possible modification for the O(n) DP algorithm is the use of frame weights to emphasize certain dissimilar segments of the reference patterns.

The augmented continuous DP matching algorithm is derived from word spotting [3]. This algorithm produces only a suboptimal solution, but is considerably faster than the constant time delay DP algorithm. The algorithm is also well suited for decoding word sequences from a given syntax.

A few comments would be appropriate at this time. The author has made the paper as mathematically complete as possible. Careful reading gives a good insight into the algorithms involved. However, to a more casual reader the algorithms would be difficult to understand without going into the mathematical details. Some insight into the algorithms without having to understand the equations would have been useful.

The author implies that syntactical constraints cannot be applied to O(n) DP programming algorithm. However, Bridle [2] has shown that syntactic constraints can be applied easily. This is evident if one separate copy of the reference is maintained for each place in the syntax where the reference is associated.