The article presents named entity recognition system, which participated in the second task of PolEval 2018 competition. It utilizes recurrent and convolutional neural networks with conditional random fields. The only external resources are provided by the morphological tagger KRNNT and word embeddings. The distinctive aspect of the solution is the lack of use of gazetteers or lexicons. Two approaches are presented to address nested annotation of named entities, each with its own advantages. The solution obtains 81.4% F1 measure.