Hi. The intermediate phrase boundary model is typically trained to be a three way classifier, capable of outputting INTONATIONAL_BOUNDARY, NONBOUNDARY and INTERMEDIATE_BOUNDARY. The prediction of INTERMEDIATE_BOUNDARY tokens is relatively inaccurate (in comparison to INTONATIONAL_BOUNDARY performance). This is, of course, consistent with human consistency in labeling of full intonational phrase boundaries, in comparison to the agreement with respect to the location of intermediate phrase boundaries. It seems likely that the model simply didn't predict any intermediate phrase boundaries.
i tried to run programmes and found intermediate_phrase_boundary_detection just return INTONATIONAL_BOUNDARY and NONBOUNDARY, does that mean intermediate_phrase_boundary_detection return INTONATIONAL_BOUNDARY which is same as INTERMEDIATE_BOUNDARY