kylelovesllms/grammar-v2-frames-ce-rb

Grammar V2 (determiners): CE-RB Populated Frames

The 1336 depth-1 syntactic frames of grammar-v2-frames (Grammar V2, experiments/02_grammar_v2/src/cfg_v2/grammar.py in the translation-syntax project), each populated with 2 randomly sampled lexicalizations per task (seeded RNG, seed=0).

Grammar V2 is the number-agreement grammar with optional determiners. Every argument (matrix subject, object, embedded subject) is a bare noun or a determiner plus noun; this and a are singular, these is plural and the goes with both numbers. The determiner keeps its position in Head Final while the relative clause moves in front of the noun: the dog that these birds like chases cat translates to the these birds like that dog cat chases.

Partition

A frame is ambiguous if the linear rule gives the hierarchical translation. The linear rule splits the sentence at the first verb and emits each half with its determiners first, in source order, followed by its other words reversed.

• ambiguous frames (360): 324 are in train with copy and translate rows, 36 are held out and are in validation with copy and translate rows
• unambiguous frames (976): copy rows in train, translate rows in test

Validation holds out frames, not sentences: 10% of the ambiguous frames of every category, spread over the determiner conditions, and only frames that keep another number variant in train. Copy and translate rows of a frame never share a sentence, so no test translate input appears as a train copy input.

Copy-task variant ce-rb: copy every structure: center embedded and right branching

Train copy rows are kept only for frames whose relative clause attaches at one of these sites: None, Subject, Object (None = no relative clause). The validation and test splits are identical across the ce-rb, ce and rb datasets. Every frame has exactly 2 sentences per task in all three, so the datasets differ in size: the dropped copy rows are not replaced.

• train: 3,248 rows = 648 translate + 2,600 copy, 64,920 tokens (<bos> src <task> <sep> tgt <eos> = 2 * n_tokens + 4 per row)
• validation: 144 rows (36 held-out frames, copy and translate)
• test: 1,952 rows (translate, unambiguous)

Train rows per structure

Task	Attachment Site	Gap Type	Frames	Rows per frame	Rows	Tokens
translate	None	None	36	2	72	904
translate	Object	Subject Gap	288	2	576	11,732
copy	None	None	36	2	72	904
copy	Subject	Subject Gap	400	2	800	15,840
copy	Subject	Object Gap	320	2	640	13,056
copy	Object	Subject Gap	288	2	576	11,732
copy	Object	Object Gap	256	2	512	10,752
Total					3,248	64,920

Columns

HF Sentence is the task target: the HI sentence itself for copy rows, the hierarchical (Head Final) translation for translate rows. Frames use indexed bracketed notation where [N_sg_3] means the terminal filled by HI token 3. Matrix Number is the number of the matrix subject, RC Number that of the noun the RC verb agrees with (None without an RC), and n_attractors counts nouns of the other number between a verb and the noun it agrees with.

Determiners lists the arguments that carry a determiner (subject, object, RC = the noun inside the relative clause). Determiner Condition groups them: no-determiner, matrix (only arguments of the matrix clause), RC (only the noun inside the RC) and matrix-RC (both). Tier is ID (in distribution) for ambiguous frames, T1 for Subject / Subject Gap, T2 for Object / Object Gap and T3 for Subject / Object Gap.

Frame counts per category (before population)

Attachment Site	Gap Type	Frames	Ambiguous	Unambiguous
None	None	40	40	0
Subject	Subject Gap	400	0	400
Subject	Object Gap	320	0	320
Object	Subject Gap	320	320	0
Object	Object Gap	256	0	256
Total		1336	360	976

Sentence counts per category (after population)

Over every split and both tasks. Sample factor is the number of sentences per frame, copy and translate together; it is lower where the variant dropped a structure's train copy rows.

Attachment Site	Gap Type	Frames	Ambiguous	Unambiguous	Sentences	Sample factor	RC number = matrix number	RC number differs
None	None	40	40	0	160	4	—	—
Subject	Subject Gap	400	0	400	1,600	4	1,600	0
Subject	Object Gap	320	0	320	1,280	4	640	640
Object	Subject Gap	320	320	0	1,280	4	640	640
Object	Object Gap	256	0	256	1,024	4	512	512
Total		1336	360	976	5,344		3,392	1,792

Translate sentences per determiner condition

Split	Attachment Site	Gap Type	no-determiner	matrix	RC	matrix-RC	Sentences
train	None	None	22	50	0	0	72
train	Object	Subject Gap	86	260	58	172	576
validation	None	None	2	6	0	0	8
validation	Object	Subject Gap	10	28	6	20	64
test	Subject	Subject Gap	144	336	96	224	800
test	Subject	Object Gap	96	224	96	224	640
test	Object	Object Gap	64	192	64	192	512