*** Guide-to-Links ***
SI
SI is used in subject-verb inversion:
  
         +----Pg---+               +---I---+
         +-SI-+    |               +SI-+   |
         |    |    |               |   |   |
	 Is Jane coming      Who did Jane see

All words carrying S+ - nouns, nominative pronouns, and many
other words - carry SI- as well.  (See "S".)  SIs is used for
singular noun and verb forms, SIp for plural forms; these are
exactly analogous to Ss and Sp.

In this entry we provide a general discussion of questions and
subject-verb inversion.

     Contents
     1. Words taking SI+
     2. Questions requiring subject-verb inversion
     3. Cases where s-v inversion may not occur
     4. Simple s-v inversion ("yes-no") questions
     5. Questions without s-v inversion
     6. Indirect questions
     7. Other uses of SI besides questions: SI#j

Words taking SI+
Only verbs which may be inverted (i.e. modals and auxiliaries)
have SI+ connectors. On such verbs, the SI+ is disjoined with
the S-, and is conjoined with whatever complement connectors
the verb may be used with when it is inverted. For example,
forms of "have" may be inverted when they are taking a past
participle, but not when they are taking a direct object or
infinitive. Thus they have

have: (SI+ & PP-) or (S- & (O+ or TO+ or PP+));

This yields:

	They have finished it
	Have they finished it
	They have dogs
	*Have they dogs
	They have to go
	*Have they to go
	
The use of SI and S connectors is highly constrained. In many
situations, subject-verb inversion may not occur; in some
situations, it must occur. In many cases, the enforcement of
this involves post-processing.

Questions requiring subject-verb inversion 
In some cases, s-v inversion must occur: in object-type
questions ("*Which dog you chase", "Which dog did you chase"),
or with question words like "where" ("*Where you will go",
"Where will you go"). This is enforced by post-processing.
When a question word begins a sentence, it must make a Wq
connection to the wall (or a Ws connection: see below). (There
is no other way for the wall to connect to the sentence.)  The
Wq connection starts a 'm' domain, and is included in the
domain.  We then require that a group with a "Wq" contain some
kind of SI link.  This prevents "*Which dog I chase".

This solution works equally well when the question contains an
embedded clause: "What do you think he did?", "*What you think
he did?" In both cases, "What" begins a group; in both cases,
the link between "you" and its finite verb is in the outer
group of this domain. (Since the Bsw is a restricted link, the
subordinate domain does not spread back to the Wq.) Thus this
group is required to contain an "SI" link of some kind.

          +----------Bsw(m(e))--------------+
          |    +----I(m)--+                 |
   +-Wq(m)+    +SI(m)+    +Ce(m(e))+S(m(e))-+
   |      |    |     |    |        |        |
 /////  What  do    you think      he      did

A similar situation arises with adjectival questions:

             +-----AF----+
             |   +---I---+
     +-Wq+EAh+   +SI-+   |
     |   |   |   |   |   |            
  /////	How big will it be

Here again, s-v inversion must be enforced ("*How big it
is"). We therefore give "how" a Wq- connector; this is then in
the same group as whatever S or SI link is in the outer
group, and post-processing insists that it must be an SI.

S-V inversion must also be enforced with the question words
"where", "when" and "how" (when used in this way):
"Where(/when/how) will you go", "*Where you will go".  This is
done with simple connector logic. Unlike with object-type
constructions, the question word here is not making a B link
to the rest of the sentence; it must find some other way to
connect. For this purpose, the question words "where", "when",
"why" and "who" have Q+ conjoined with Wq-; and Q- on verbs is
conjoined with SI+, disjoined with S-.

	where: (Wq- & Q+)...
	have: ({Q-} & SI+ & ...) or ({C-} & S+ & ...)

Thus we allow "Where have you gone"; we prevent "*Where you
have gone".

Cases where s-v inversion may not occur
In many cases, s-v inversion is prohibited: in relative
clauses ("*The dog who did you buy was black") and subordinate
clauses ("*I left the party after did you see Fred"). No
linkages are found for these; again, there is no way for the
illegally-inverted segment to connect to the rest of the
sentence. (The one case where s-v inversion may occur unwanted
is in indirect questions; this problem is discussed below.)

Simple s-v-inversion ("yes-no") questions 
Questions may also be formed by simply inverting the verb and
the auxiliary ("Are you coming", "Did you go"). In this case
the question must make some kind of connection to the wall; we
use the Q link for this, giving the wall Q+. (This connector
is subscripted Qd+, for reasons explained below.)

                +---I--+
          +--Q--+-SI+  |
          |     |   |  |
	/////  Did you go

Questions without s-v inversion
Post-processing cannot insist on subject-verb inversion in all
question-word questions. Subject-type questions do not contain
s-v inversion:
         
         +-S--+
         |    |
	Who  hit John

To allow this, question words like "who" and "which" may also
make a Ws connection to the wall. Ws is exempted from the
post-processing constraints applied to Wq; a group containing
a Ws need not contain an SI link (indeed, they may not; see
below).

	who: (S+ or B+) or (Ws- or Wq-);
	which: (D**w+ or S+ or B+) or (Ws- or Wq-);

(The special subscript on the D+ of "which" will be explained
below.)  But in that case what will prevent illegal sentences
being formed using the Ws: "*Who John hit", and "*Where John
goes"? As described above, the second case is prevented by
link logic; the sentence cannot form. For the first case, we
require post-processing: we dictate that a group with a Ws may
not contain an B#m. (We must also prevent the redundant parses
resulting from Ws being used in "where/when/why" questions
with s-v inversion. We do this by simply not giving these
words Ws.)  In short: if there is a question-word present in
the main clause, there must either be a Wq, in which case s-v
inversion is enforced, or there must be a Ws, in which case
B#m connections are prohibited in post-processing and
"where-when-why" questions do not form. The result is that
with Ws all s-v inversion questions are prevented either at
the linkage stage or in post-processing; Ws is therefore used
with all and only non-s-v questions, Wq with all and only s-v
questions.

There is one final problem: enforcing s-v inversion in
object-type questions with embedded clauses.

                     +----------Bsm(m(e))----+
   +-Ws(m)+D**w(m(e))+   +S(m)+Ce(m)-+S(m(e))+
   |      |          |   |    |      |       |
*/////  Which      dog  you think   you     hit

Here, "which" can use its Ws connection.  The parser then
thinks it is a subject-type question; it finds no s-v
inversion, and no B link in the outer group, therefore it
accepts the sentence. To prevent this, we assign question-word
determiners ("which" and "what" a D**w+ subscript; we then
stipulate that a group with a Ws must contain a "D##w"
link. We further make the B#m link not a restricted link;
the D##w is restricted, however. (See "B: Object-type
Questions".)  In this case, then, the 'e' group spreads back
to include the "D##w" link; the group containing the Ws
therefore no longer contains a D**w; and the sentence is
rejected. (In effect, we want post-processing to know when
something is a well-formed subject-type question. It only
knows this is the case if a) there is an S in the outer group
of the sentence, and b) the D link of that subject is
also in that group.)

(One more little annoyance: We've said that groups with Ws
correspond to subject-type questions. We've insisted that they
may not contain SI links; and, to avoid the false positive
just discussed, we've insisted that they must contain a
D**w. Now the only problem is, they sometimes don't contain a
D**w; sometimes the question word is itself the subject: "Who
is coming?".  So we give question-words specially subscripted
"S**w+" connectors; and then we say, a group with a Ws must
contain either a D##w or an S##w.)

Indirect Questions
It was mentioned that in most cases where s-v inversion is
prohibited (relative clauses and the like), it is prevented at
the linkage level. There is one case where it is not, however,
namely indirect questions.

                +----B(s)-----+
                |   +----I(s)-+
            +-QI+   +SI(s)+   |
            |   |   |     |   |
      * I know who did   you hit

We prevent this in post-processing by saying that an SI
link may only occur when Wq or Qd is present.  Wq
connections can only form between a question-word and the
wall; Qd can only form between an auxiliary verb and the wall.
Thus the above construction will be rejected.  (With
"where/when/how" questions, this is accomplished at the
linkage level. QI- on these words is disjoined with Q+; thus
there is no way for the question word to connect to a
s-v-inverted clause.)  The same applies to "how [adjective]"
questions: "*I wonder how big is it" is therefore rejected.

Other Uses of SI Besides Questions: SI#j
There are a few other situations besides questions where SI is
used.  In subjunctive clauses, a nominative noun phrase is
used (e.g., "he" rather than "him"). In such cases, however,
is most convenient to connect it to the preceding "that",
rather than the following verb. Therefore we use SI here.

                     +----I--+
              +--TS--+-SI-+  |
              |      |    |  |
	I suggested that  he go

We also use SI in quotation constructions with certain verbs like
"say":

	 +-------------CP---------------+----SI----+
	 |                              |          |
       /////   The President is busy, said the spokesman

As described above, the use of SI is usually tightly
constrained by post-processing. Rather than try to adjust the
post-processing constraints to handle these situations, we
avoid the problem. We simply give the SI+ a "*j" subscript
here, and make the post-processing constraints apply only
to unsubscripted "SI#*" links. Thus the post-processing
constraints do not apply, and SI#j can be used anywhere that a
linkage can be found. Verbs (and the word "that") that take
subjunctive clauses thus carry "SI*j+".

SI#j is also used in conditional constructions like "I would
have seen you, HAD YOU been there". 

Grammar Documentation Page.