Multidimensional IRT in brms

Interfaces brms
5

I have some experimental data where respondents saw a random political slogan and then reported how it made them feel on a 0-10 scale across 10 emotions which theory suggests load onto three latent factors.

They look as follows:

# A tibble: 94,530 x 7
      id    w8 gender slogan          emotion      resp  group          
   <dbl> <dbl> <fct>  <fct>           <fct>        <ord> <fct>          
 1 97322  1.07 Female Get Brexit done Enthusiastic 5     Positive Affect
 2 97322  1.07 Female Get Brexit done Hopeful      10    Positive Affect
 3 97322  1.07 Female Get Brexit done Proud        5     Positive Affect
 4 97322  1.07 Female Get Brexit done Scared       5     Anxiety        
 5 97322  1.07 Female Get Brexit done Worried      5     Anxiety        
 6 97322  1.07 Female Get Brexit done Afraid       6     Anxiety        
 7 97322  1.07 Female Get Brexit done Hateful      5     Aversion       
 8 97322  1.07 Female Get Brexit done Angry        5     Aversion       
 9 97322  1.07 Female Get Brexit done Bitter       5     Aversion       
10 97322  1.07 Female Get Brexit done Resentful    5     Aversion

I’d like to model how varying the slogan and respondent gender affects their latent emotions.

I fit the following model to a single slogan, which seemed to work well:

brm(
  formula =
    bf(resp | weights(w8) ~ 1 + (1 |i| emotion) + (0 + group | id),
       disc ~ 1 + (1 |i| emotion)),
  family = 
    cumulative(
      link = "probit",
      link_disc = "log"),
  prior = 
    prior(normal(0, 1), class = "Intercept") +
    prior(constant(1), class = "sd", group = "id") +
    prior(normal(0, 3), class = "sd", group = "emotion") +
    prior(normal(0, 1), class = "sd", dpar = "disc") +
    prior(lkj(2), class = "cor"),
  data = 
    dta %>%
    filter(slogan == "Get Brexit done"),
  iter = 2e3,
  chains = 2,
  cores = 2
 )

My question: How can I extend the model to handle covariates like gender that might themselves vary over the slogan that each respondent saw?