you can use both ANOVA or Paired Sample t test. what i understood, better is to use paired sample t test to measure the difference. I can help you more if you share data and hypothesis(s). 

The paired sample t-test might tell you whether readers in general percieved the modified texts as being less difficult than the original texts, but it won´t tell you anything about the influence of the other variables, and importantly, won´t control for a potential confounding effect of the other variables. If, for example, the overall difference in ratings is driven by 2 out of the 4 questionnaires, percieved difficulty is related to differences between the texts or participant differences between the four groups of participants. A t-test will not be informative regarding this. Also, although ordinal data such as rating scores often is treated as continuous, there are other tests availible for ordinal data.

I would recommend either using linear regression (treating your dependent variable as continuous) or, even better, ordinal regression. In that way, you will be able to include all other variables as well as the groups as independent variables, and thereby determine to what extent these factors influence your results.

Ideally, you would want to use ordinal mixed effects modeling, including participants and narrative texts as random effects, but I am not sure if this is implemented in SPSS.

Good luck!

You can do 'ordinal mixed effects modeling', as suggested by Thomas, in R using the ordinal package. 

As an additional suggestion, I strongly encourage you to do some descriptive analysis of your data! Overall your sample is a bit small for use of inferential modeling alone, and it generally helps to see descriptive analysis even in cases where a sample is somewhat larger. Presenting some contingency tables to complement your ANOVA tables should do the trick. Also, if your outcome measures weren't continuous you will likely fare better with the McNemar test of significance for between-group comparison tests.