Perhaps a somewhat different take on the situation?
A) To translate the FE model from ABAQUS to Nastran, do as suggested by the other authors.
B) If you wish to use ABAQUS modal data inside a Nastran FE model.
As you can see here. Two FE software provide different results.
Presentation Analysis Technology:; A discussion on some of the fundamentals
I learned in a different RG discussion that this difference can be traced back to how solvers operate.
If recovering the ABAQUS results in Nastran is a theme, you can calculate mass normalized modes, export these and use them in Nastran or any other software using e.g. McNeal's method to substrcturing.
In the McNeal method, a mode is a SDOF system with unity mass, k = w2 and the mode shape is implemented as constraint equations. There are some examples of this in the presentation as well (see the SEREP stuff).
Using the McNeal method, model output data translation becomes the exact ABAQUS output.
It works for real modes, not for complex modes.
C) If you refer to data translation from Abaqus format to Nastran format.
Then export from ABAQUS .into ascii FIL, write a small translator yourself and convert it into Nastran punch format.
These formats are easy and there are several open source translators around at least for each of them, if not for the exact combination. LMS Virtual Lab has such data input/output capabilities.
If you need further operations on Nastran output data, take a look at PyNastran. https://github.com/SteveDoyle2/pyNastran
I happened to come across this online converter. It handles Abaqus, Nastran, Ansys, Code Aster and LS Dyna.
https://www.simright.com/applications/converter/
It is free to use for small models.
You may want to take a look also at GMSH which is free and can make part of the model conversions, not from Abaqus though. Nastran is a better start point.