The following cmavo are discussed in this section:

The cmavo si (of selma'o SI) is a metalinguistic operator that erases the preceding word, as if it had never been spoken:

means the same thing as ti mlatu. Multiple si cmavo in succession erase the appropriate number of words:

In order to erase the word zo, it is necessary to use three si cmavo in a row:

The first use of si does not erase anything, but completes the zo quotation. Two more si cmavo are then necessary to erase the first si and the zo.

Incorrect names can likewise cause trouble with si:

The Lojbanized spelling .esperanto breaks up, as a consequence of the Lojban morphology rules (see Chapter 4) into two Lojban words, the cmavo e and the undefined lujvo speranto. Therefore, two si cmavo are needed to erase them. Of course, .e speranto is not grammatical after la, but recognition of si is done before grammatical analysis.

Even more messy is the result of an incorrect zoi:

In Example 19.79, the first fy is taken to be the delimiting word. The next word must be different from the delimiting word, and gy., the Lojban name for the letter g, was chosen arbitrarily. Then the delimiting word must be repeated. For purposes of si erasure, the entire quoted text is taken to be a word, so four words have been uttered, and four more si cmavo are needed to erase them altogether. Similarly, a stray lo'u quotation mark must be erased with fy. le'u si si si, by completing the quotation and then erasing it all with three si cmavo.

What if less than the entire zo or zoi construct is erased? The result is something which has a loose zo or zoi in it, without its expected sequels, and which is incurably ungrammatical. Thus, to erase just the word quoted by zo, it turns out to be necessary to erase the zo as well:

The parser will reject zo .djan. si .djordj., because in that context djordj. is a name (of selma'o CMENE) rather than a quoted word.

Note: The current machine parser does not implement si erasure.

As the above examples plainly show, precise erasures with si can be extremely hard to get right. Therefore, the cmavo sa (of selma'o SA) is provided for erasing more than one word. The cmavo following sa should be the starting marker of some grammatical construct. The effect of the sa is to erase back to and including the last starting marker of the same kind. For example:

Since the word following sa is i, the sentence separator, its effect is to erase the preceding sentence. So Example 19.81 is equivalent to:

Another example, erasing a partial description rather than a partial sentence:

In Example 19.83, le blanu .zdan. is ungrammatical, but clearly reflects the speaker's original intention to say le blanu zdani. However, the zdani was cut off before the end and changed into a name. The entire ungrammatical le construct is erased and replaced by le xekri zdani.

Note: The current machine parser does not implement sa erasure. Getting sa right is even more difficult (for a computer) than getting si right, as the behavior of si is defined in terms of words rather than in terms of grammatical constructs (possibly incorrect ones) and words are conceptually simpler entities. On the other hand, sa is generally easier for human beings, because the rules for using it correctly are less finicky.

The cmavo su (of selma'o SU) is yet another metalinguistic operator that erases the entire text. However, if the text involves multiple speakers, then su will only erase the remarks made by the one who said it, unless that speaker has said nothing. Therefore susu is needed to eradicate a whole discussion in conversation.

Note: The current machine parser does not implement either su or susu erasure.