Halliday & Matthiessen (1999: 150, 157):
As Figure 4-7 above shows, the generalised participant roles of Medium and Agent correspond to different sets of roles in the particularistic model — one or more for each type of figure. For example, in a figure of saying the Medium corresponds to the Sayer, whereas in a figure of sensing it corresponds to the Senser. We summarise the correspondences in Table 4(8), where the generalised participant roles are represented in columns.
Blogger Comments:
Note that, for encoding identifying figures, Agent corresponds to Token, and Medium to Value. This is because, in all identifying figures, Medium corresponds to Identified, and Identified conflates with Token in decoding figures (as in the table), but conflates with Value in encoding figures (not shown in the table). See Figure 4-7.