Do infants detect A→V articulator congruency for non-native click consonants?

In a prior study infants habituated to an audio-only labial or alveolar, native English voiceless or non-native ejective stop, then saw silent videos of stops at each place [1]. 4-month-olds gazed more at congruent videos for native and non-native stops. 11-month-olds preferred congruence for native stops but incongruence for non-native ejectives, suggesting language experience biases but does not block detection of non-native AâžV speech relations. But as English adults perceive ejectives as deviant stops [2], we asked whether infants detect AâžV congruence in non-native phones adults hear as nonspeech, i.e., click consonants [3-6]. 4-month-olds preferred incongruency; 11-month-olds showed no preference. We posit that infants prefer AâžV congruency for phones heard as native-like speech; prefer incongruency for phones heard as speech that deviates from native segments; notice extreme deviance earlier (clicks: 4 mo; ejectives: 11 mo); and later treat very deviant phones as discriminable nonspeech sounds [3, 4] that are unrelated to visual speech. Results are at odds with existing AV models, but may be handled by a hybrid of Amodal Articulatory and Intersensory Narrowing views.