This talk investigates a set of phonological patterns in the SFU Speech Error Database (SFUSED), with the goal of understanding if and how phonological grammar is involved in phonological encoding. It assesses 2,076 sound errors in English of various types (phonological substitutions, deletions, additions, etc.) for their phonological regularity (do they violated English phonotactics?) and their sensitivity to a variety of markedness constraints commonly used in speech error analysis. The results show that sound errors are much less regular (violate phonotactics more often) than reported in prior work, and they reveal no significant effects from eight distinct measures of phonological markedness. The higher degree of phonological irregularity is attributed to methodology, because the methods for collecting speech errors in SFUSED are demonstrably less prone to perceptual bias. While phonological encoding may require linguistic representations of planning units (e.g., segments), these findings suggest that some of the tools of phonological grammars, like syllable-structure algorithms and phonological markedness constraints, may not be required in the retrieval of phonological segments in words. Pilot results from SFUSED Cantonese (with over 2,500 speech errors in Cantonese speech) will also be explored with a similar research focus.