I conducted an experiment in which students rated sentences and I would like to fit a mixed effects model using lme4. My response variable is RATING (i.e. the judgements given by the students), the two predictors (SENTENCE_TYPE (3 levels) and WORD_ORDER (2 levels)) are categorical variables. In addition, I would like to fit random intercepts for SUBJECT and VERB to account for idiosyncratic preferences of individual students and lexical effects of individual verbs. However, the variable VERB only applies to two levels of SENTENCE_TYPE and cannot be coded for tokens that belong to the third level. Is it nevertheless possible to include VERB as a random effect and to get reliable estimates for the intercept adjustments?