Most compression algorithms do consider the context of previously-received data -- but only from the same file/unit/transmission/whetever, aside from predefined

Most compression algorithms do consider the context of previously-received data -- but only from the same file/unit/transmission/whetever, aside from predefined contexts and dictionaries for some algorithms (like the DCT in jpeg or the fixed huffman tree in H.263 decoders). but I think it might be worth considering that compression need not always be bound by the type of data, but also by the receiver of that data and the intention and method of that data's transmission. It is. This is why we have various entropy-coding schemes for important data (arthmetic, golomb, huffman, lzw, burrows-wheeler/bzip2, rle), and lossy compression for data that doesn't need to be perfect, such as audio, video, and imagery (mp3, h.264, jpg) which use irreversible transforms to pack a lot more information in each bit.