In the discussion on embodied representation of language, Cangelosi argues that language learning needs to be grounded in sensori-motor representations -thus, that learning proceeds from actions to language, although not all words need to be directly grounded in sensori-motor representations. In more recent work, new concepts have been learned through explicit verbal definition (e.g. "grab" [is] "close-left-arm" [and] "close-right-arm"). Thus, language can be used to shape action learning. In our comment we want to second and emphasize the importance of language for action learning. It is not the sole purpose of language to function in the role as a symbol; it also plays a role as a social signal influencing attention and the memory performance (e.g. [7,10]) that can be observed in developmental processes. We propose that the relationship between action and language should be viewed as a symbiosis rather than a link.
The reverse impact of language onto action can be seen in the phenomenon that has been termed Acoustic Packaging and which relates to the function of segmenting and structuring an observed action by speech [6]: Imagine that you see a dancer performing a dance. You watch it and it seems to you like a whole. If you would like to learn it, you need to break the whole down into its elements. In dancing classes, the single elements are provided with synchronous speech when a tutor performs them. This way, learners' attention is drawn to the single elements. Similar to this experience, for a child learning actions, it is difficult to break down the whole to its elements. In this process, it is helpful when a tutor provides verbal input that takes the role of a social signal and marks the single elements. Information that is provided in such a redundant manner will be picked up more easily by child's senses [1]. Once the single elements are picked up, they can be memorized. The way, language 'packages' events can be characterized as education of attention [2,4,11] that drives our understanding of actions. Once such a package is discovered and stored as meaningful (in form of a top-down knowledge), the perception becomes more flexible and independent of a direct link between action and language as social signal.
In our first model towards Acoustic Packaging, we are testing which forms of correlation and synchronization between action and language can be verified for action structuring. For this purpose in [8], multimodal units are detected, which consist of overlapping speech and action intervals. Based on this segmentation approach found more Acoustic Packages in child-directed as opposed to adult-directed interaction, which is consistent with the current