Current research on information extraction from spoken documents is mainly focused on the recognition of named entities, such as names of organizations, locations and persons, within transcripts automatically generated by a speech recognizer. In this work we present research carried out at ITC-irst on named entity recognition in Italian broadcast news. In particular, an original statistical named entity tagger is described which can be trained with relatively little language resources: a seed list of named entities and a large untagged text corpus. Moreover, the paper discusses and presents named entity recognition experiments with case sensitive automatic transcripts, generated by the ITC-irst speech recognizer, and by training the named entity model with seed lists of different size.