The idea that ground de-/anti-icing had something to do with the incident can be safely disregarded. If one has really bad case of iced wing one either: 1) fails to get airborne 2) gets airborne but then mush in ground effect until hitting the obstacle 3) stalls after leaving ground effect 4) gets wing ice ingested by tail mounted engines with consequential trust loss. Since this birdie has flown for hours, obviously it left ground effect at the good side of the trust curve. Even if the aeroplane was not de-iced properly, the aerodynamic effect of contamination was not enough to induce the upset.
Now, there is a picture making circles and it's been claimed it represents the QAR readout of the incident flight. Provided it's real, the triggering event was control law degradation to ALT1. I cannot figure out what caused it; G trace doesn't show any significant turbulence while left and right IAS and baro alt are consistent all throughout the upset. There are intermittent faults of both yaw dampers, but they got recorded later, before and after the wildest of gyrations. It looks like F/O was PF but the captain took over the controls and pushed the nose down, reaching minimum of -8.2°, before pulling up to about 30. Some interesting pitch, roll and speed excursion followed, probably helped by large and alternating stick displacement. Picture resolution is too low to figure out whether occasional dual input exacerbated or damped the ride. Control law degraded to direct but restored to ALT1 after about five minutes. There were some stall, some overspeed and some unspecified red warnings. Eventually everything settled down.