post hoc pairwise comparison I have a model testing whether the modifications to my texts (fixed factor: Modified) affect several measures (for example IA_FIXATION_COUNT). The Texts used in my experiments are 8, and when i want to see the difference in general between original and modified texts i imput them as random factor "Text", together with the random factor "Participant". There are few additional fixed factors that also are supposed to affect measures, such as word length and the position of the word on the line.
glmer(IA_FIXATION_COUNT ~ Modified +
                         Word_length +
                         Position_line +
                         Page +
                         Trial +
                         (1 | Text) +
                         (1 | Participant),
                       data = paired_dataset,
                       family = poisson())

When I run my glmer(), I get that the modifications have a significant effect on my measure.
Now, if I wanted to have a look at the different texts, and specifically whether each of my 8 texts features (or not) a significant effect of Modified, I thought I could add Text as fixed factor. I have been advised, however, to run a post hoc comparison instead. 
I had a look at multcomp and emmeans, but not only I have to get rid of my random variables, also the result i get is a multiple comparison among all my texts, whereas I just want to see whether Text 1 original differs from Text 1 modified and so on.
Does anyone have any recommendation?
 A: I think it is OK to have a model where Text is a random factor, used for purposes of general inferences about other factors; and another model where Text is a fixed factor, used for making specific comparisons involving texts. Perhaps some others disagree; I'd be interested in what they say.
To get the comparisons you want, you need to have Text interact with Modified in the fixed part of the model:
model2 <- glmer(IA_FIXATION_COUNT ~ Text * Modified +
                         Word_length +
                         Position_line +
                         Page +
                         Trial +
                         (1 | Participant),
                       data = paired_dataset,
                       family = poisson())

Then, if I understand the comparisons you want, they may be done like this:
library(emmeans)
emm <- emmeans(model2, ~ Modified | Text)
pairs(emm)

The above compares the levels of Modified for each Text. But perhaps you want a multiplicity correction for the set of all 8 of these comparisons:
summary(pairs(emm), by = NULL, adjust = "mvt")

This ignores the by variable (after the comparisons are made), and summarizes them using the multivariate t correction.
Comment
The reason you need the interaction is because with only additive effects, you are fitting a model on the assumption that the effect of Modified is the same for each Test, and vice-versa. 
Which leads me to wonder if you also want them to interact in the first model as well: ...fixed part... + (1|Participant) + (1|Text:Modified)
And in fact, carefully consider whether some other fixed factors should interact with one another...
