SRILM sometimes omits backoff weights (presumably for N-grams that don't participate in any N+1-grams), which breaks the text language model reader.
Log in to post a comment.
