We present a generic natural language processing (NLP) architecture, acronym QTIL, based on a system of cooperating multiple agents (Q/A, T, I, and L agents) which can be used in any information system incorporating Internet Information Retrieval. We then introduce a hybrid multi-agent system (MAS) architecture, acronym QTIP, for the privacy domain through integrating the PeCAN (Personal Context Agent Networking) and QTIL MAS architectures. There are two areas where NLP is used: in the user-MAS interaction and in the process of resource indexing and matching. These two areas map to the Q/Aagent and to the I-agents. We propose using a lightweight Head-driven Phrase Structure Grammar (HPSG) natural language method for the Q architectural layers and qualitatively justify its applicability. We provide an example of employing the HPSG formalism for Information Retrieval using natural language capability via Privacy Web Services in one instantiation of the QTIP architecture. Independent preliminary results for HPSG on the Q level show that our approaches for enhancing the usability of PET tools are promising.