Test 10 · How a language model learns: guessing the next word

Which of these did a language model NOT need to learn?

Mova's choice: option 4 · The right answer: option 1 · wrong ✗ · What happens: not need, learn

This is test 10 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	Which	which	PRON	PronType=Rel	4 did	obj
2	of	of	ADP	_	3 these	case
3	these	this	PRON	Number=Plur PronType=Dem	4 did	obl
4	did	do	AUX	Mood=Ind Number=Sing Person=1 Tense=Past VerbForm=Fin	9 need	aux
5	a	a	DET	Definite=Ind PronType=Art	7 model	det
6	language	language	NOUN	Number=Sing	7 model	compound
7	model	model	NOUN	Number=Sing	4 did	obj
8	NOT	not	PART	Polarity=Neg	9 need	advmod
9	need	need	VERB	Mood=Imp VerbForm=Fin	root	root
10	to	to	PART	_	11 learn	mark
11	learn	learn	VERB	VerbForm=Inf	9 need	xcomp
12	?	?	PUNCT	_	9 need	punct

#	verb	tense, mood	who	what, whom	to whom	where	when	other	roles by
1	not need	past · do-support							vn want-32.1-1-1 «NP V NP S_INF»: agent=Pivot clause=Theme
2	learn not said to happen	to-infinitive							vn learn-14-1 «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 · ✓ the right answer

A body to feel things with

The words, one row each (CoNLL-U)

#	word	lemma	UPOS	features	head	relation
1	A	a	DET	Definite=Ind PronType=Art	2 body	det
2	body	body	NOUN	Number=Sing	4 feel	nsubj
3	to	to	PART	_	4 feel	mark
4	feel	feel	VERB	VerbForm=Inf	root	root
5	things	thing	NOUN	Number=Plur	4 feel	obj
6	with	with	ADP	_	4 feel	obl

#	verb	tense, mood	who	what, whom	to whom	where	when	other	roles by
1	feel	to-infinitive	a body	things					vn hunt-35.1 «NP V NP»: agent=Agent thing=Theme

Option 2

People to build and train it

The words, one row each (CoNLL-U)

#	word	lemma	UPOS	features	head	relation
1	People	people	NOUN	Number=Plur	3 build	nsubj
2	to	to	PART	_	3 build	mark
3	build	build	VERB	Mood=Imp VerbForm=Fin	root	root
4	and	and	CCONJ	_	5 train	cc
5	train	train	VERB	Mood=Imp VerbForm=Fin	3 build	conj
6	it	it	PRON	Case=Acc Gender=Neut Number=Sing Person=3 PronType=Prs	5 train	obj

#	verb	tense, mood	who	what, whom	to whom	where	when	other	roles by
1	build	to-infinitive	People						vn calibratable_cos-45.6.1 «NP.attribute V»: agent=Attribute
2	train	infinitive	People not said here	it					vn hire-13.5.3 «NP V NP»: agent=Agent it=Theme

Option 3

Powerful computers to run on

The words, one row each (CoNLL-U)

#	word	lemma	UPOS	features	head	relation
1	Powerful	powerful	ADJ	Degree=Pos	2 computers	amod
2	computers	computer	NOUN	Number=Plur	root	root
3	to	to	PART	_	4 run	mark
4	run	run	VERB	VerbForm=Inf	2 computers	acl
5	on	on	ADP	_	4 run	compound:prt

#	verb	tense, mood	who	what, whom	to whom	where	when	other	roles by
1	run on	to-infinitive							vn run-51.3.2-2-1 «NP V»: agent=Theme

Option 4 · ← Mova's choice

Huge amounts of text to learn from

The words, one row each (CoNLL-U)

#	word	lemma	UPOS	features	head	relation
1	Huge	huge	ADJ	Degree=Pos	2 amounts	amod
2	amounts	amount	NOUN	Number=Plur	root	root
3	of	of	ADP	_	4 text	case
4	text	text	NOUN	Number=Sing	2 amounts	nmod
5	to	to	PART	_	6 learn	mark
6	learn	learn	VERB	VerbForm=Inf	2 amounts	acl
7	from	from	ADP	_	6 learn	obl

#	verb	tense, mood	who	what, whom	to whom	where	when	other	roles by
1	learn	to-infinitive							vn learn-14-1 «NP V»: agent=Agent
2	amount			text					nom amount «possession transitive»: of=patient

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/t10/