Clause-embedding verbs: obligatoriness of complementizers and possibility of adjuncts

I am looking at some ambiguity in embedded questions, trying to understand whether the last reading is actually crazy and if so, how to get rid of it. The wh-adverb in the examples can be replaced with any adverb for generality, in examples (4)-(6).

CONTEXT (I hope it is not too distracting).

Here are two readings (2)-(3) that I plan to admit in my Russian regression grammar, for the following sentence:

(1) Я   вижу,    как Иван     читает   книгу
    Ya  vizhu,   kak Ivan     chitaet  knigu
    1SG see.1SG  HOW Ivan.NOM read.3SG book.ACC
"I see how Ivan is reading the book." [rus]

(2) Screen Shot 2020-04-27 at 2.07.25 PM

(2) is licensed at the top by a subject-head rule and the embedded clause is a filler-gap phrase.

Now, as for (3), there was a related thread where I (implicitly) ended up deciding that it might also be OK to have (3), perhaps with some very heavy emphasis on HOW (the top node is the in situ question rule, the embedded clause is adjunct-head phrase):

(3) Screen Shot 2020-04-27 at 2.47.07 PM

ISSUE.

Now here’s one tree which it will be harder to convince myself that it’s OK:

(4) Screen Shot 2020-04-27 at 2.51.31 PM

In (4), my wh-adverb decided to modify the embedding verb. I really don’t like that, however I cannot say this adverb is [ POSTHEAD - ] or anything like that; in a given clause, it can appear anywhere:

(5) Ivan     chitaet kak knigu?
    Ivan.NOM reads   HOW book.ACC 
"How is it that Ivan is reading the book?" [rus]

Is there an elegant way to get rid of this while maintaining the otherwise flexible adjunct position?

Or… Is this crazy tree possible after all? The following is OK:

(6) Я   вижу    КАК, что Иван     читает   книгу?
    Ya  vizhu   KAK, chto Ivan     chitaet  knigu?
    1SG see.1SG HOW  that Ivan.NOM read.3SG book.ACC
"How do I see that Ivan is reading a book?" [rus]

But here the complementizer that helps enormously to understand what is being asked. Without the complementizer… But, complementizers are generally optional.

It sounds like it would be ideal to use the MODIFIED feature on the clause-embedding verb, set it to notmod but only in cases where there is no complementizer. Is that possible to do? Two lexical entries for each verb?.. Not too bad, perhaps? (This would be a change in the clausal complements library.)

I think this an interesting puzzle (reminds me of adverb amelioration of the that-trace effect in English) but one to set aside for future work :slight_smile: