How to solve the memory error in Python

Assuming your example text is representative of all the text, one line would consume about 75 bytes on my machine:

In [3]: sys.getsizeof('usedfor zipper fasten_coat')
Out[3]: 75

Doing some rough math:

75 bytes * 8,000,000 lines / 1024 / 1024 = ~572 MB

So roughly 572 meg to store the strings alone for one of these files. Once you start adding in additional, similarly structured and sized files, you'll quickly approach your virtual address space limits, as mentioned in @ShadowRanger's answer.

If upgrading your python isn't feasible for you, or if it only kicks the can down the road (you have finite physical memory after all), you really have two options: write your results to temporary files in-between loading in and reading the input files, or write your results to a database. Since you need to further post-process the strings after aggregating them, writing to a database would be the superior approach.

Simplest solution: You're probably running out of virtual address space (any other form of error usually means running really slowly for a long time before you finally get a MemoryError). This is because a 32 bit application on Windows (and most OSes) is limited to 2 GB of user mode address space (Windows can be tweaked to make it 3 GB, but that's still a low cap). You've got 8 GB of RAM, but your program can't use (at least) 3/4 of it. Python has a fair amount of per-object overhead (object header, allocation alignment, etc.), odds are the strings alone are using close to a GB of RAM, and that's before you deal with the overhead of the dictionary, the rest of your program, the rest of Python, etc. If memory space fragments enough, and the dictionary needs to grow, it may not have enough contiguous space to reallocate, and you'll get a MemoryError.

Install a 64 bit version of Python (if you can, I'd recommend upgrading to Python 3 for other reasons); it will use more memory, but then, it will have access to a lot more memory space (and more physical RAM as well).

If that's not enough, consider converting to a sqlite3 database (or some other DB), so it naturally spills to disk when the data gets too large for main memory, while still having fairly efficient lookup.