Methods and systems are provided for natural-language surveillance of packet-based communications. Speech data is received, representing at least part of a speech session. The speech data is compared with at least one speech-characteristics profile in an automated manner. The comparing step comprises using natural-language-processing (NLP) to compare the speech data with the profiles on at least one of a lexical level, a syntactic level, a semantic level, a discourse level, and a pragmatic level. It is determined in an automated manner whether the speech data matches any of the profiles. If the speech data matches at least one profile, an indication that the speech data matches the at least one matching profile is responsively stored.