ABSTRACT
Recent spectrogram-reading experiments (Cole, Rudnicky, Reddy, & Zue, Chap. 1) have shown that the acoustic signal is rich in phonetic information. Without knowing anything about the words that are present, an expert spectrogram reader can produce a broad phonetic transcription that agrees with a panel of phoneticians from 80 to 90% of the time, depending on the scoring method used. Furthermore, perceptual experi ments by Liberman and Nakatani (personal communication) indicate that listeners can transcribe nonsense names embedded in sentences (and obeying the phonological constraints of English) with better than 90% phonemic accuracy.
