Frequentist likelihood-based inference includes (but is not limited to) maximum likelihood, likelihood ratio tests, and confidence intervals. Because generalized linear mixed models have unobservable random effects, the likelihood function is often a high-dimensional integral that cannot be expressed in closed form. Therefore, inference based on the likelihood is very difficult.
The R package glmm enables likelihood-based inference by producing a Monte Carlo approximation the likelihood. Any likelihood-based inference can be performed using this approximation. The package also produces Monte Carlo maximum likelihood estimates and Fisher information (which can be used for creating confidence intervals or for performing hypothesis tests).