Two research strategies aimed at understanding how maternal speech input enables pre-productive infants to segment words from fluent speech are summarized. The first strategy consists of gathering audio-recordings of maternal speech directed to 12-month-olds in two different tasks: teaching the infant a new word and reading from a story book. The second strategy consists of constructing a simulation of the word-segmentation process using neural networks. Transcripts and detailed acoustic analyses of maternal speech from native speakers of English and Turkish revealed the following: Mothers (1) do not consistently use the target (or topical) word in isolation, (2) do not avoid or enhance difficult-to-segment word boundaries, (3) typically highlight the target word in an utterance by using exaggerated pitch contours, and (4) typically place the target word in utterance-final position, even when such placement violates strict grammatically. Taken together, these findings suggest that mothers are aware of neither the acoustic difficulty of word segmentation nor the need to enhance the segmental information for word boundaries. Mothers do, however, spontaneously use exaggerated pitch, but this pitch-marking is insufficient for word segmentation because the preceding and following words are blended with the change in pitch. Perhaps most importantly, mothers appear to have tacit knowledge of the potential benefit associated with placing the target word in utterance-final (or utterance-initial) position. Not only does this placement reduce the segmentation problem (by eliminating one of the word boundaries), but it capitalizes on either recency (or primacy) in the encoding of the word into memory.