Analysis of an 8-state Ergodic HMM

Figure  1  illustrates the analysis of an 8-state Ergodic HMM.

The circles represent the states of an Ergodic HMM. The line widths of the circles show the initial probabilities. The line width of the transition arcs show transition probabilities. The POS codes labeling the transition s in this figure are shown in Table  4 .

The results of this analysis are shown in the following for each POS.

1. Interjection

   Interjections tend to move in the state 0.

2. Common noun

   Common noun mainly move the HMM to state ○ 1○ 2 from ○ 0○ 3○ 7.

   In transition of ○ 3 , the probability of a formal noun is greater than a common noun. Note that the transition of ○ 3 is output for complementizer /no/. The transition of state 0 is mainly output for complementizer /hou/.

3. Conjugate word

   Verbs, auxiliary verbs or adjectives that can be conjugated POS are not grouping by the POS but grouping by the conjunction. For example, an attributive base is output from states ○ 4,○ 1 to ○ 3. A continuative base is output from states ○ 4,○ 5,○ 6 to ○ 4. A conclusive base of auxiliary verb is output state ○ 4,○ 5,○ 6 to ○ 4.

4. Case particle

   Case particles are found in a lot of transitions; each transition is output another word. Examples are given in the following.

   
   

   • state ○ 2$\Rightarrow$○ 0 /no/ (94%)
     
   • state ○ 2$\Rightarrow$○ 7 /kara/ (28%) /de/ (25%)
     
   • state ○ 2$\Rightarrow$○ 4 /to/ (94%)
     
   • state ○ 2$\Rightarrow$○ 5 /ni/ (73%) /de/ (18%)
     
   • state ○ 1$\Rightarrow$○ 5 /wo/ (39%) /ga/ (27%)