Is Replay Gain correctly named Normalization? How is it different? Is it more like an internal volume control applied to each track in order to hit the target output volume?
The goal is the same -- even volume across multiple tracks that might have been recorded using different notions of "standard" volume. But the method differs, and there is some debate about the pros/cons.
Both Replay Gain and Normalization analyze the track and determine by how much it should be adjusted to playback at "standard" volume. Replay Gain then stores this adjustment value, but does not alter the actual audio. The stored RG value is used by the playback system to adjust volume of that track every time it is played. This requries that the playback system can read and use the Replay Gain value to make the correct adjustment. In contrast, Normalization rewrites the physical track with a new volume so it always plays back at the same new volume on any system.
Of course, both processes are applied to the total track, in-effect adjusting the volume control at the start of the track and not touching it until the next track. There's no mid-track or dynamic adjustment as would be done with compression or limiting.
The two main controversies are: What is the "standard" volume and how is it determined? And, does Normalization degrade the track's "quality"?
The max clean volume of a track is the level just before the track gets distorted, typically by "clipping" where the top of the loudest waveforms get bent or chopped off. The way to avoid this distortion is, don't let any waveforms get clipped anywhere in an audio chain. A "safe" preventative is to keep all audio levels low so there's no chance of clipping, but this can put the desired audio too close to residual and extraneous noise. The "perfect" recording level is way above noise and just below clipping.
Normalization typically looks at the one track being Normalized. It usually determines how far below clipping the loudest part is, then adjusts the entire track so this loudest point is just a smidgen below clipping. So, "standard" volume is essentially "near-max before clipping", with the max value adjustable in many Normalization systems (a couple of peak values are in common use).
Replay Gain, because it is a combination of an analysis of the track by itself, AND an adjustment made at playback time, can take into account the set of tracks being played, and try to adjust them all to be fairly consistent with each other. This assumes the playback system that uses Replay Gain is sophisticated enough. MC 12 seems to do this very well. Without adjustment relative to other tracks, the volume result of Replay Gain might be the same Normalization: play every track at the loudest level without distorting.
The end result of volume adjustments, at playback time with Replay Gain, or previously applied to the track with Normalization, is often virtually identical to the listener.
The other controversy is whether Normalization affects the quality of the track, because when it rewrites the track at an adjusted volume, it potentially introduces digital alterations that affect sound accuracy/quality. It's a legitimate point, but the debate is sometimes over theoretical differences in digitized signal/waveform that might not be noticeable to human listeners.
Normalization, because it's a one time adjustment of a track, is an appealing fix-and-forget method. In that way it's just getting back to the audio level that presumably existed in the original recording, where normal practice is to keep peaks just below clipping. A Normalized track will play at the "correct" volume anywhere, anytime. A Replay Gain coded track must be played on a Replay Gain system or there's no benefit at all.
So, Normalization would be the perfect solution, if it wasn't for the "degradation" debate. The real question is whether the process of Normalization has a negative effect on the track's quality/accuracy that matters to the listener. I'm on the fence, but mostly use Replay Gain. I used to Normalize everything, before Replay Gain became available, and I still sometimes Normalize certain tracks using Sony SoundForge for certain reasons. But the vast majority of my library uses only Replay Gain. However, when I listen to the same songs done both ways, it's hard to convince myself there's a difference, other than RG and Normalization don't always end up with precisely the same playback volume. There might be a much greater difference converting a "pure" waveform to .mp3, itself a debatable practice.
So, if some audio alteration is desired by the listener for practical reasons -- consistent volume, smaller file size, audio equalization -- then the practical question is, "how much is acceptable", and the answer is in the ears of the listeners. Determine what sounds good, and use it. Or, if more alteration is needed (diddle with volume, trade off file size for quality, tweak EQ), apply it and see if the result is still "good". If not, back off until the desired balance is achieved. In this context, Normalization might be perfectly acceptable -- or not.
Of course, another justification for avoiding practices that might permanently alter the audio is the argument that you CAN'T trust your ears and/or playback system. Usually the weakest link in playback accuracy is the particular amp and speakers used to listen, a combination that is quite unlikely to reproduce the sound exactly. What sounds "good enough" on one system might be horriblly deficient on another. This is a key argument for saving music as .flac or .ape rather than .mp3, and not Normalizing -- you might not hear the difference now, but maybe someday...
I've simplified the technical details, but hopefully provided a sense of how Replay Gain and Normalization differ in getting to a similar outcome.