Your understanding of the situation is correct: The angle of attack was beyond stall and the aircraft was in a descent which reduces the pitch attitude, so the resulting acceleration felt to the pilots close to that in normal flight.
Now it is important to understand that the aircraft produced lift equivalent to its weight. It was in a steady state - if lift had been too small, it would had accelerated downwards. Let's calculate how big the descent angle could have been: The minimum speed of an A330 in clean configuration is 166 KIAS at the maximum landing weight, which is 396,800 lbs or 182 t. Since the aircraft had just flown for less than four hours, I would estimate that it weighed 205 t at the time of the accident, which makes its stall speed 176 KIAS or 90.6 m/s. In about 3½ minutes it went from 11,600 m to sea level, which gives us an average sink speed of 55 m/s. For a very rough estimate we use the mean between the stall speed at 11,600 m, which is 175 m/s, and the stall speed at sea level: 132 m/s. The resulting angle is 22.5° (note that the flight speed is along the flight path, so you need the arctan of 55 / 132.6). The real value was probably a little lower since the aircraft flew a little faster than its stall speed to compensate for the lower lift coefficient in the stalled state. Modern supercritical airfoils have a very benign stall behavior, and I would estimate that the speed was not much above stall speed. Also, it did not fly constantly at the same attitude, but the copilot released his pitch-up command twice, only to resume it seconds later.
When I use the cited values at the time of impact (107 kts ground speed and 108 kts sink speed), the airspeed is 78.2 m/s. Assuming standard atmospheric conditions, this gives a lift coefficient of 1.48 - way more than any stalled wing can manage. This can only be explained by a strong headwind, of which there is no word in the Wikipedia article. Since the plane was flying through a storm, strong winds should be expected, and then the figures become credible again. Tom's value of 60° sounds very unlikely to me - at that angle the wing produces mostly drag and the aerodynamic forces act at mid chord, producing a strong pitch-down moment which cannot be compensated by a tail surface at those 60°, regardless of elevator position. Around 20° to 30° angle of attack the whole situation becomes much more plausible.
EDIT: Now I spent some time reading the BEA report (thank you @mins for the link!) and are deeply troubled by some details. In FL360 (figure 65) and with 1g (figure 66), the airplane was supposed to have flown at only Mach 0.4. This translates into a post-stall lift coefficient of 2.09, which is physically impossible. Some lift is contributed by the engines due to the positive pitch angle, granted, but by far not enough to make this low speed possible. At least BEA agrees with my mass estimate.
The dynamic pressure at this point is only 2660 N/mm², and in clean configuration this is too little to prevent the plane from dropping like a stone at any angle of attack. The same goes for the condition right before it impacted the ocean surface: 78.2 m/s is too low; at least 95 m/s would be needed for a barely credible post-stall lift coefficient of 1.0. If I add 17 m/s wind speed, things get back into normal territory. Unfortunately, the only wind information in the plots is in figure 64, when the stall begins. Headwind is around 0, but crosswind is between 20 and 30 m/s. If the airplane had sideslipped so badly during the stall, it would had entered a spin. It just does not make sense.
In the text we get a headwind information at the time of the autopilot disconnect (page 91):
Prior to the disconnection of the autopilot, a constant headwind
component of 15 kt had to be added in order to make the simulation’s
ground speed match the recorded parameter. This value was consistent
with the wind parameters recorded.
But the most important line is at the start of the analysis, hidden away at the bottom of page 90. It should be bold and underlined, but isn't:
The validity of the model is limited to the known flight envelope
based on flight tests.
Airbus did some more flights at the same configuration and loading as AF447 had at the time of the accident, but it appears that buffet limited their flown angles of attack to below 10°.
It must be concluded that all angle of attack values exceeding 10° are purely speculative and not backed by flight test data.