Test 9 · How a language model learns: guessing the next word

Why do words appear one after another when a bot answers?

Mova's choice: option 1 · The right answer: option 3 · wrong ✗ · Who acts: words, bot · What happens: appear, answer · To what, where: one

This is test 9 of lesson 2 «How a language model learns: guessing the next word» of the AI course of Mova's primary school. The page gives the question's syntax tree — a dependency tree in the style of Universal Dependencies — and its pragmatics clause by clause, then the 4 options as the student saw them, each with its tree and pragmatics. Every text is parsed by Mova itself. How the student answered it, with the score of every option and why: the test at school.

The question

The words, one row each (CoNLL-U)

#	word	lemma	UPOS	features	head	relation
1	Why	why	ADV	PronType=Int	4 appear	advmod
2	do	do	AUX	Mood=Ind Number=Plur Person=3 Tense=Pres VerbForm=Fin	4 appear	aux
3	words	word	NOUN	Number=Plur	4 appear	nsubj
4	appear	appear	VERB	VerbForm=Inf	root	root
5	one	one	NUM	NumForm=Word NumType=Card	4 appear	obj
6	after	after	ADP	_	7 another	case
7	another	another	DET	PronType=Ind	4 appear	obl
8	when	when	ADV	PronType=Int	11 answers	advmod
9	a	a	DET	Definite=Ind PronType=Art	10 bot	det
10	bot	bot	NOUN	Number=Sing	11 answers	nsubj
11	answers	answer	VERB	Mood=Ind Number=Sing Person=3 Tense=Pres VerbForm=Fin	4 appear	parataxis
12	?	?	PUNCT	_	4 appear	punct

#	verb	tense, mood	who	what, whom	to whom	where	when	other	roles by
1	appear	present · do-support	words	one			after another		vn appear-48.1.1 «PP.location V NP»: agent=Location one=Theme
2	answer speech: assert	present	a bot						vn reciprocate-112-1 «NP V»: agent=Agent

One row per clause the reader finds. «not said here»: a role filled from the clause above. «roles by»: a VerbNet class and frame when one fits the verb, else the reader's rules on the tree.

The options

Option 1 · ← Mova's choice

It is waiting for you to read each word

The words, one row each (CoNLL-U)

#	word	lemma	UPOS	features	head	relation
1	It	it	PRON	Case=Nom Gender=Neut Number=Sing Person=3 PronType=Prs	3 waiting	nsubj
2	is	be	AUX	Mood=Ind Number=Sing Person=3 Tense=Pres VerbForm=Fin	3 waiting	aux
3	waiting	wait	VERB	Tense=Pres VerbForm=Part	root	root
4	for	for	SCONJ	_	7 read	mark
5	you	you	PRON	Case=Nom Person=2 PronType=Prs	7 read	nsubj
6	to	to	PART	_	7 read	mark
7	read	read	VERB	VerbForm=Inf	3 waiting	advcl
8	each	each	DET	PronType=Tot	9 word	det
9	word	word	NOUN	Number=Sing	7 read	obj

#	verb	tense, mood	who	what, whom	to whom	where	when	other	roles by
1	wait	present · progressive	It						vn exist-47.1 «NP V»: agent=Theme
2	read	to-infinitive	you	each word					vn transfer_mesg-37.1.1-1 «NP V NP»: agent=Agent word=Topic

Option 2

The internet is too slow to send a whole sentence at once

The words, one row each (CoNLL-U)

#	word	lemma	UPOS	features	head	relation
1	The	the	DET	Definite=Def PronType=Art	2 internet	det
2	internet	internet	NOUN	Number=Sing	5 slow	nsubj
3	is	be	AUX	Mood=Ind Number=Sing Person=3 Tense=Pres VerbForm=Fin	5 slow	cop
4	too	too	ADV	_	5 slow	advmod
5	slow	slow	ADJ	Degree=Pos	root	root
6	to	to	PART	_	7 send	mark
7	send	send	VERB	VerbForm=Inf	5 slow	xcomp
8	a	a	DET	Definite=Ind PronType=Art	10 sentence	det
9	whole	whole	ADJ	Degree=Pos	10 sentence	amod
10	sentence	sentence	NOUN	Number=Sing	7 send	obj
11	at	at	ADP	_	12 once	case
12	once	once	ADV	NumForm=Word NumType=Mult	7 send	obl

#	verb	tense, mood	who	what, whom	to whom	where	when	other	roles by
1	is	present	the internet					slow property	tree rules
2	send	to-infinitive	the internet not said here	a whole sentence					vn confine-92-1 «NP V NP»: agent=Agent sentence=Theme

Option 3 · ✓ the right answer

It works out the answer in pieces and shows them as they come

The words, one row each (CoNLL-U)

#	word	lemma	UPOS	features	head	relation
1	It	it	PRON	Case=Nom Gender=Neut Number=Sing Person=3 PronType=Prs	2 works	nsubj
2	works	work	VERB	Mood=Ind Number=Sing Person=3 Tense=Pres VerbForm=Fin	root	root
3	out	out	ADP	_	2 works	compound:prt
4	the	the	DET	Definite=Def PronType=Art	5 answer	det
5	answer	answer	NOUN	Number=Sing	2 works	obj
6	in	in	ADP	_	7 pieces	case
7	pieces	piece	NOUN	Number=Plur	5 answer	nmod
8	and	and	CCONJ	_	9 shows	cc
9	shows	show	VERB	Mood=Ind Number=Sing Person=3 Tense=Pres VerbForm=Fin	2 works	conj
10	them	they	PRON	Case=Acc Number=Plur Person=3 PronType=Prs	9 shows	obj
11	as	as	SCONJ	_	13 come	mark
12	they	they	PRON	Case=Nom Number=Plur Person=3 PronType=Prs	13 come	nsubj
13	come	come	VERB	Mood=Ind Number=Plur Person=3 Tense=Pres VerbForm=Fin	9 shows	advcl

#	verb	tense, mood	who	what, whom	where	roles by
1	work out	present	It	the answer	pieces	vn discover-84 «NP V NP»: agent=Agent answer=Topic
2	show	present		pieces = piece		vn transfer_mesg-37.1.1-1-1 «NP V NP.recipient»: agent=Agent they=Recipient
3	come	present	pieces = piece			vn escape-51.1-1 «NP V»: agent=Theme

Option 4

It is a real person somewhere typing as fast as they possibly can for you

The tree, 15 words

The words, one row each (CoNLL-U)

#	word	lemma	UPOS	features	head	relation
1	It	it	PRON	Case=Nom Gender=Neut Number=Sing Person=3 PronType=Prs	5 person	nsubj
2	is	be	AUX	Mood=Ind Number=Sing Person=3 Tense=Pres VerbForm=Fin	5 person	cop
3	a	a	DET	Definite=Ind PronType=Art	5 person	det
4	real	real	ADJ	Degree=Pos	5 person	amod
5	person	person	NOUN	Number=Sing	root	root
6	somewhere	somewhere	ADV	PronType=Ind	5 person	advmod
7	typing	type	VERB	VerbForm=Ger	5 person	acl
8	as	as	ADV	_	9 fast	advmod
9	fast	fast	ADV	Degree=Pos	7 typing	advmod
10	as	as	SCONJ	_	13 can	mark
11	they	they	PRON	Case=Nom Number=Plur Person=3 PronType=Prs	13 can	nsubj
12	possibly	possibly	ADV	_	13 can	advmod
13	can	can	AUX	VerbForm=Fin	9 fast	advcl
14	for	for	ADP	_	15 you	case
15	you	you	PRON	Case=Acc Person=2 PronType=Prs	7 typing	obl

#	verb	tense, mood	who	what, whom	to whom	where	when	other	roles by
1	is a	present	It	a real person				somewhere property	tree rules
2	type	gerund	a real person not said here					for you; fast property	vn scribble-25.2-1 «NP V»: agent=Agent

Text: the tests of Mova's primary school, written for Mova (world/data/school/), as the course runner gave them to the student. Trees: Mova's own parse — the en model on the CPU, when the site is built — not checked by a teacher; they may be wrong. Frames: Mova's frames reader (prag), each text read on its own. Elamur Cosmic License.

This page in Markdown → https://md.mova.elamur.ai/free/en/corpora/questions/ai/2-guessing-the-next-word/t9/