The amazing durability of infant memory: Three-year-olds show recognition of a person they met once at age one

We all know that infants have incredible capabilities. For instance, we know that a big percentage of babies emit their first words or meaningful combinations of words around the age of 18 months. However, there is more to babies' abilities than just words, we all know that. 

A study addressed at the BPS Research Digest brings up an interesting but puzzling research study, that I haven't quite been able to put together. 

This article focus on the topic of memory formed by babies on moment A and their retrieval on moment B. 

To maximise the chance of uncovering long-term memory through infancy and into early childhood, the researchers devised a scenario involving many many prompts - what they described as "massive cueing". Kingo and his team first renewed contact with parents and their children who'd taken part in an earlier study when the children were age one. That earlier research involved the infant children interacting with one of two researchers for 45 minutes - either a Scandinavian-Caucasian man or a Scandinavian-African man.

Now two years on, 50 of these parents and children - the latter now aged three - were invited back to the exact same lab (hopefully cueing their earlier memories). Here the children were shown two simultaneous 45-second videos side by side. One video was a recording of the researcher - either the Scandinavian-Caucasian or Scandinavian-African man - interacting with them two years earlier; the other video showed the other researcher (the one they hadn't met) interacting with a different child in the exact same way. The children themselves were not visible in these videos.

The key test was whether the three-year-olds would show a preference for looking at one video rather than the other. Amazingly, the children spent significantly more time looking at the video that featured the researcher they'd never met. This is not due to the children having a bias for either the white or black man, because for some of these children the previously unseen researcher was Scandinavian-African  and for others he was Scandinavian-Caucasian. All background features and behaviours in the videos were identical, so this result provides strong evidence that the children had some recognition of the researcher they'd met, and were drawn more strongly to look at the unfamiliar researcher. 

This conclusion puzzles me. Shouldn't children look at the researcher that they had met two years before, as if they were identifying familiarity in the process of memory retrieval? Why is that looking at the unfamiliar researcher makes the case for the conclusion that the babies had memory of the researcher two years before and therefore were drawn to the unfamiliar researcher?

I believe there is something missing in this puzzle.