Context free parsing appears to have polynomial time complexity at best. To posit something closer to linear time, we must therefore give something up - preferably something we did not want in the first place. I propose that they should be structures per strings that involve more than limited amounts of sensor imbedding. Linear time complexity that this allows in theory is however still hard to achieve in practice.

• Research Fellow, Natural Language Theory and Technology Area, Information Sciences and Technologies Laboratory, Xerox PARC.
• Professor of Linguistics, Stanford University.
• Research Interests: Design of algorithms for computational linguistics, machine translation.