Pošalji zapis e-poštom: How semantic biases in simple adjacencies affect learning a complex structure with non-adjacencies in AGL: a statistical account