The number reported in your quote is 12%.
The other numbers are uncertainties. In general, when two uncertainties are reported separately like this, one is the "statistical" uncertainty and the other is the "systematic." The statistical uncertainty can be reduced by running your experiment for longer. The systematic uncertainty generally can't be improved without redesigning the experiment (but can sometimes be reduced after the fact, by understanding the apparatus better somehow). The abstract for your paper reports a statistical uncertainty first, so in your quote the $\pm9\%$ is statistical and the $\pm14\%$ is systematic.
As a commenter points out, a lot has changed in the proton-spin problem in the thirty years since the paper you're reading was published. But you can see why the puzzle was interesting: $12\pm14$ was consistent with zero of the proton's spin coming from its valence quarks.