Datasets:

bbunzeck
/

lexical-frames

id int64 1 50	n_words int64 2 6	freq int64 15 432	beginning stringlengths 4 33
1	6	22	I like to play with my
2	6	22	and then you put it in
3	6	21	but I don't know how to
4	6	19	when I grow up I want
5	6	17	I don't know how to get
6	6	17	but I don't know where the
7	6	17	and then put it in the
8	6	16	and then they went to the
9	6	16	and how I would do it
10	6	15	can I have a bit of
11	5	224	and then there was a
12	5	76	I don't know what it
13	5	73	I don't want you to
14	5	72	I wanna play with the
15	5	54	I don't know where it
16	5	53	I don't know what to
17	5	52	I wanna go to the
18	5	49	what do you do with
19	5	45	I want to play with
20	5	45	the boy and the dog
21	4	194	what is that funny
22	4	172	and this is a
23	4	161	how do you spell
24	4	151	do you want a
25	4	128	I think it's a
26	4	122	it looks like a
27	4	115	because I don't like
28	4	108	no I don't want
29	4	100	put it on the
30	4	93	what do you wanna
31	3	369	look at the
32	3	302	and that's the
33	3	292	I like the
34	3	251	is that the
35	3	233	that is a
36	3	225	it is a
37	3	185	a lot of
38	3	181	and I was
39	3	175	that's a big
40	3	169	what is it
41	2	432	to the
42	2	342	a baby
43	2	316	not that
44	2	236	another one
45	2	235	that's your
46	2	233	it a
47	2	224	where's your
48	2	217	take a
49	2	196	a red
50	2	192	do that
1	6	122	It is one of the most
2	6	106	It is interesting to note that
3	6	92	All you have to do is
4	6	91	The reason for this is that
5	6	91	This is due to the fact
6	6	90	Note that depending on the number
7	6	84	It can also be used to
8	6	77	This is one of the most
9	6	75	It should also be noted that
10	6	72	It is a good idea to
11	5	141	To learn more about the
12	5	104	The bottom line is that
13	5	102	It can be used to
14	5	98	At the same time the
15	5	96	There are three types of
16	5	92	For example, if you are
17	5	91	At the start of the
18	5	89	In the middle of the
19	5	88	This can be done by
20	5	84	What are the benefits of
21	4	143	However, there is a
22	4	133	This allows you to
23	4	118	While there is no
24	4	114	In recent years, the
25	4	108	That is why the
26	4	104	There are also some
27	4	102	In any case, the
28	4	102	In many cases, the
29	4	102	Thank you for your
30	4	98	In other words, a
31	3	180	I am a
32	3	176	In short, the
33	3	159	Learn about the
34	3	141	The system is
35	3	140	He said the
36	3	136	Many people have
37	3	135	Several of the
38	3	133	In conclusion, the
39	3	132	He became a
40	3	129	Because of these
41	2	166	The police
42	2	160	Using these
43	2	156	Get your
44	2	154	Eventually the
45	2	142	So to
46	2	140	Accordingly, the
47	2	139	Otherwise, the
48	2	135	This simple
49	2	135	Consider a
50	2	133	Since a

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

language:

en ---+

This repository contains the lexical frames used to prompt the models in the frame completion task we introduced in Child-directed speech facilitates production, not comprehension, in BabyLMs (Bunzeck & Zarrieß @ CoNLL 2026).

If you use this data, please cite:

@inproceedings{bunzeck-zarriess-2026-child,
    title = "Child-directed speech facilitates production, not comprehension, in {B}aby{LM}s",
    author = "Bunzeck, Bastian  and
      Zarrie{\ss}, Sina",
    editor = "Bonial, Claire  and
      Berzak, Yevgeni",
    booktitle = "Proceedings of the 30th Conference on Computational Natural Language Learning",
    month = jul,
    year = "2026",
    address = "San Diego, California, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.conll-main.14/",
    doi = "10.18653/v1/2026.conll-main.14",
    pages = "227--249",
    ISBN = "979-8-89176-410-1",
    abstract = "Recent studies suggest that child-directed speech is not conducive to language learning in BabyLMs. However, current evaluations focus predominantly on comprehension and not production, which is central to usage-based theories of language acquisition which argue how CDS facilitates early language use through constructional ``frames'' (frequent lexical patterns with open slots). We introduce a novel generation-based evaluation inspired by such theories in form of a \textbf{frame-completion task}, and compare Llama models trained with CDS, the BabyLM corpus, and web-crawl data (FineWeb-edu) on comprehension benchmarks and our novel framework. Our results reveal a clear dissociation between models' comprehension and production capabilities: while FineWeb-trained models excel at minimal pairs, CDS-trained models produce grammatical completions substantially earlier in training and concentrate probability mass on appropriate slot-fillers. These findings show that comprehension benchmarks underestimate what CDS affords to BabyLMs."
}

Downloads last month: 18

Collection including bbunzeck/lexical-frames

Child-directed speech facilitates production (CoNLL 2026)

Collection

Models and datasets for *Child-directed speech facilitates production, not comprehension, in BabyLMs* (Bunzeck & Zarrieß 2026) • 3 items • Updated 2 days ago