Datasets:

aieng-lab
/

gradiend-function-composition

masked large_stringlengths 26 41	label large_stringclasses 60 values	label_class large_stringclasses 1 value	alternative large_stringclasses 60 values	alternative_class large_stringclasses 1 value	chain_depth int64 1 1	query_var large_stringclasses 26 values	difficulty large_stringclasses 1 value	split large_stringclasses 1 value	feature_class_id large_stringclasses 1 value
m = fork; p = seal; h = m; h = [MASK]	fork	RESULT	seal	DISTRACTOR	1	h	easy	train	RESULT
u = copper; q = u; q = [MASK]	copper	RESULT	shark	DISTRACTOR	1	q	easy	train	RESULT
r = lion; p = ring; o = r; o = [MASK]	lion	RESULT	ring	DISTRACTOR	1	o	easy	train	RESULT
w = desk; j = lime; a = w; a = [MASK]	desk	RESULT	lime	DISTRACTOR	1	a	easy	train	RESULT
w = fork; e = white; c = w; c = [MASK]	fork	RESULT	white	DISTRACTOR	1	c	easy	train	RESULT
z = seal; k = z; k = [MASK]	seal	RESULT	fish	DISTRACTOR	1	k	easy	train	RESULT
q = green; n = q; n = [MASK]	green	RESULT	map	DISTRACTOR	1	n	easy	train	RESULT
j = brown; z = deer; l = j; l = [MASK]	brown	RESULT	deer	DISTRACTOR	1	l	easy	train	RESULT
v = gray; r = coral; p = v; p = [MASK]	gray	RESULT	coral	DISTRACTOR	1	p	easy	train	RESULT
n = mirror; v = whale; r = n; r = [MASK]	mirror	RESULT	whale	DISTRACTOR	1	r	easy	train	RESULT
y = box; b = rose; v = y; v = [MASK]	box	RESULT	rose	DISTRACTOR	1	v	easy	train	RESULT
i = copper; s = i; s = [MASK]	copper	RESULT	lamp	DISTRACTOR	1	s	easy	train	RESULT
i = plum; j = shark; x = i; x = [MASK]	plum	RESULT	shark	DISTRACTOR	1	x	easy	train	RESULT
c = slate; j = c; j = [MASK]	slate	RESULT	black	DISTRACTOR	1	j	easy	train	RESULT
f = fork; u = f; u = [MASK]	fork	RESULT	pen	DISTRACTOR	1	u	easy	train	RESULT
o = mouse; d = eagle; i = o; i = [MASK]	mouse	RESULT	eagle	DISTRACTOR	1	i	easy	train	RESULT
b = cat; t = rabbit; f = b; f = [MASK]	cat	RESULT	rabbit	DISTRACTOR	1	f	easy	train	RESULT
p = mint; z = coin; c = p; c = [MASK]	mint	RESULT	coin	DISTRACTOR	1	c	easy	train	RESULT
s = pink; e = orange; y = s; y = [MASK]	pink	RESULT	orange	DISTRACTOR	1	y	easy	train	RESULT
y = door; t = plate; p = y; p = [MASK]	door	RESULT	plate	DISTRACTOR	1	p	easy	train	RESULT
b = clock; m = b; m = [MASK]	clock	RESULT	shark	DISTRACTOR	1	m	easy	train	RESULT
y = seal; p = y; p = [MASK]	seal	RESULT	door	DISTRACTOR	1	p	easy	train	RESULT
s = silver; k = black; m = s; m = [MASK]	silver	RESULT	black	DISTRACTOR	1	m	easy	train	RESULT
k = spoon; h = k; h = [MASK]	spoon	RESULT	deer	DISTRACTOR	1	h	easy	train	RESULT
c = bag; w = mint; e = c; e = [MASK]	bag	RESULT	mint	DISTRACTOR	1	e	easy	train	RESULT
a = fork; x = bag; v = a; v = [MASK]	fork	RESULT	bag	DISTRACTOR	1	v	easy	train	RESULT
w = brown; j = mouse; d = w; d = [MASK]	brown	RESULT	mouse	DISTRACTOR	1	d	easy	train	RESULT
f = lamp; u = f; u = [MASK]	lamp	RESULT	silver	DISTRACTOR	1	u	easy	train	RESULT
d = mirror; y = frog; t = d; t = [MASK]	mirror	RESULT	frog	DISTRACTOR	1	t	easy	train	RESULT
d = bowl; y = lime; k = d; k = [MASK]	bowl	RESULT	lime	DISTRACTOR	1	k	easy	train	RESULT
n = camel; p = n; p = [MASK]	camel	RESULT	plum	DISTRACTOR	1	p	easy	train	RESULT
a = fish; c = a; c = [MASK]	fish	RESULT	rust	DISTRACTOR	1	c	easy	train	RESULT
w = map; g = pen; b = w; b = [MASK]	map	RESULT	pen	DISTRACTOR	1	b	easy	train	RESULT
i = fish; l = knife; f = i; f = [MASK]	fish	RESULT	knife	DISTRACTOR	1	f	easy	train	RESULT
z = silver; g = gold; y = z; y = [MASK]	silver	RESULT	gold	DISTRACTOR	1	y	easy	train	RESULT
u = bowl; d = u; d = [MASK]	bowl	RESULT	knife	DISTRACTOR	1	d	easy	train	RESULT
v = stone; n = rope; t = v; t = [MASK]	stone	RESULT	rope	DISTRACTOR	1	t	easy	train	RESULT
k = ring; m = k; m = [MASK]	ring	RESULT	shark	DISTRACTOR	1	m	easy	train	RESULT
j = silver; x = shark; p = j; p = [MASK]	silver	RESULT	shark	DISTRACTOR	1	p	easy	train	RESULT
y = ring; v = y; v = [MASK]	ring	RESULT	turtle	DISTRACTOR	1	v	easy	train	RESULT
y = crow; w = y; w = [MASK]	crow	RESULT	ring	DISTRACTOR	1	w	easy	train	RESULT
v = orange; j = brown; z = v; z = [MASK]	orange	RESULT	brown	DISTRACTOR	1	z	easy	train	RESULT
b = rust; g = b; g = [MASK]	rust	RESULT	lion	DISTRACTOR	1	g	easy	train	RESULT
p = wolf; y = gray; q = p; q = [MASK]	wolf	RESULT	gray	DISTRACTOR	1	q	easy	train	RESULT
m = mirror; b = m; b = [MASK]	mirror	RESULT	yellow	DISTRACTOR	1	b	easy	train	RESULT
m = deer; z = m; z = [MASK]	deer	RESULT	eagle	DISTRACTOR	1	z	easy	train	RESULT
z = copper; h = z; h = [MASK]	copper	RESULT	bear	DISTRACTOR	1	h	easy	train	RESULT
n = knife; e = cat; v = n; v = [MASK]	knife	RESULT	cat	DISTRACTOR	1	v	easy	train	RESULT
e = lion; h = bowl; b = e; b = [MASK]	lion	RESULT	bowl	DISTRACTOR	1	b	easy	train	RESULT
q = rose; n = q; n = [MASK]	rose	RESULT	camel	DISTRACTOR	1	n	easy	train	RESULT
p = rose; w = fish; r = p; r = [MASK]	rose	RESULT	fish	DISTRACTOR	1	r	easy	train	RESULT
q = ring; b = q; b = [MASK]	ring	RESULT	sand	DISTRACTOR	1	b	easy	train	RESULT
j = plum; w = knife; x = j; x = [MASK]	plum	RESULT	knife	DISTRACTOR	1	x	easy	train	RESULT
e = white; n = dog; t = e; t = [MASK]	white	RESULT	dog	DISTRACTOR	1	t	easy	train	RESULT
o = frog; a = eagle; x = o; x = [MASK]	frog	RESULT	eagle	DISTRACTOR	1	x	easy	train	RESULT
n = black; p = rabbit; x = n; x = [MASK]	black	RESULT	rabbit	DISTRACTOR	1	x	easy	train	RESULT
p = coin; q = p; q = [MASK]	coin	RESULT	knife	DISTRACTOR	1	q	easy	train	RESULT
w = copper; o = whale; z = w; z = [MASK]	copper	RESULT	whale	DISTRACTOR	1	z	easy	train	RESULT
h = camel; e = pen; r = h; r = [MASK]	camel	RESULT	pen	DISTRACTOR	1	r	easy	train	RESULT
g = bag; r = bear; c = g; c = [MASK]	bag	RESULT	bear	DISTRACTOR	1	c	easy	train	RESULT
l = fish; p = l; p = [MASK]	fish	RESULT	eagle	DISTRACTOR	1	p	easy	train	RESULT
d = coral; g = fork; v = d; v = [MASK]	coral	RESULT	fork	DISTRACTOR	1	v	easy	train	RESULT
v = cat; r = v; r = [MASK]	cat	RESULT	whale	DISTRACTOR	1	r	easy	train	RESULT
e = purple; g = e; g = [MASK]	purple	RESULT	white	DISTRACTOR	1	g	easy	train	RESULT
a = spoon; r = a; r = [MASK]	spoon	RESULT	whale	DISTRACTOR	1	r	easy	train	RESULT
c = green; s = rabbit; f = c; f = [MASK]	green	RESULT	rabbit	DISTRACTOR	1	f	easy	train	RESULT
t = copper; x = duck; m = t; m = [MASK]	copper	RESULT	duck	DISTRACTOR	1	m	easy	train	RESULT
y = pink; b = y; b = [MASK]	pink	RESULT	door	DISTRACTOR	1	b	easy	train	RESULT
t = whale; p = mint; k = t; k = [MASK]	whale	RESULT	mint	DISTRACTOR	1	k	easy	train	RESULT
n = copper; b = lamp; v = n; v = [MASK]	copper	RESULT	lamp	DISTRACTOR	1	v	easy	train	RESULT
j = seal; x = mirror; u = j; u = [MASK]	seal	RESULT	mirror	DISTRACTOR	1	u	easy	train	RESULT
z = ring; v = z; v = [MASK]	ring	RESULT	deer	DISTRACTOR	1	v	easy	train	RESULT
t = deer; z = t; z = [MASK]	deer	RESULT	lamp	DISTRACTOR	1	z	easy	train	RESULT
f = deer; r = slate; t = f; t = [MASK]	deer	RESULT	slate	DISTRACTOR	1	t	easy	train	RESULT
p = camel; n = p; n = [MASK]	camel	RESULT	slate	DISTRACTOR	1	n	easy	train	RESULT
a = rope; l = a; l = [MASK]	rope	RESULT	bear	DISTRACTOR	1	l	easy	train	RESULT
d = rope; k = d; k = [MASK]	rope	RESULT	ring	DISTRACTOR	1	k	easy	train	RESULT
y = seal; b = y; b = [MASK]	seal	RESULT	yellow	DISTRACTOR	1	b	easy	train	RESULT
y = slate; r = y; r = [MASK]	slate	RESULT	copper	DISTRACTOR	1	r	easy	train	RESULT
m = gold; s = bear; x = m; x = [MASK]	gold	RESULT	bear	DISTRACTOR	1	x	easy	train	RESULT
x = lion; r = x; r = [MASK]	lion	RESULT	eagle	DISTRACTOR	1	r	easy	train	RESULT
w = camel; n = fork; c = w; c = [MASK]	camel	RESULT	fork	DISTRACTOR	1	c	easy	train	RESULT
a = pink; l = a; l = [MASK]	pink	RESULT	mirror	DISTRACTOR	1	l	easy	train	RESULT
z = rope; o = z; o = [MASK]	rope	RESULT	clock	DISTRACTOR	1	o	easy	train	RESULT
r = crow; j = r; j = [MASK]	crow	RESULT	mouse	DISTRACTOR	1	j	easy	train	RESULT
b = pink; n = rose; m = b; m = [MASK]	pink	RESULT	rose	DISTRACTOR	1	m	easy	train	RESULT
u = green; z = u; z = [MASK]	green	RESULT	bowl	DISTRACTOR	1	z	easy	train	RESULT
h = lamp; j = slate; x = h; x = [MASK]	lamp	RESULT	slate	DISTRACTOR	1	x	easy	train	RESULT
s = eagle; t = coral; l = s; l = [MASK]	eagle	RESULT	coral	DISTRACTOR	1	l	easy	train	RESULT
f = bag; s = f; s = [MASK]	bag	RESULT	rabbit	DISTRACTOR	1	s	easy	train	RESULT
d = shark; u = mint; o = d; o = [MASK]	shark	RESULT	mint	DISTRACTOR	1	o	easy	train	RESULT
e = crow; c = duck; s = e; s = [MASK]	crow	RESULT	duck	DISTRACTOR	1	s	easy	train	RESULT
j = map; f = clock; q = j; q = [MASK]	map	RESULT	clock	DISTRACTOR	1	q	easy	train	RESULT
e = rose; h = e; h = [MASK]	rose	RESULT	candle	DISTRACTOR	1	h	easy	train	RESULT
w = wolf; d = rose; p = w; p = [MASK]	wolf	RESULT	rose	DISTRACTOR	1	p	easy	train	RESULT
o = fish; t = lime; f = o; f = [MASK]	fish	RESULT	lime	DISTRACTOR	1	f	easy	train	RESULT
r = spoon; y = r; y = [MASK]	spoon	RESULT	shark	DISTRACTOR	1	y	easy	train	RESULT
p = coral; b = p; b = [MASK]	coral	RESULT	knife	DISTRACTOR	1	b	easy	train	RESULT
j = fork; z = sand; p = j; p = [MASK]	fork	RESULT	sand	DISTRACTOR	1	p	easy	train	RESULT
o = mint; s = o; s = [MASK]	mint	RESULT	fish	DISTRACTOR	1	s	easy	train	RESULT

End of preview. Expand in Data Studio

GRADIEND Function Composition Data

Synthetic alias-resolution cloze data used in the GRADIEND/ACTIEND/SAE/CAA comparison (aieng-lab/iend-study).

Usage

from datasets import load_dataset

ds = load_dataset("aieng-lab/gradiend-function-composition", "default", split="train")

Splits: train, validation, test.

Dataset Details

Dataset Description

One variable aliases a variable holding the target value, e.g. m = fork; p = seal; h = m; h = [MASK] (target fork). The factual class is RESULT.

Dataset Structure

Configs default (easy), medium, and hard; each has all splits (8,000 / 1,000 / 1,000 rows).

masked: the input text with a single [MASK] slot to be predicted
label: the factual target token(s) for the slot
label_class: class of the factual target
alternative: the alternative (counterfactual) target for the slot
alternative_class: class of the alternative target
feature_class_id: identifier of the feature class (equal to label_class)
split: train, validation, or test
difficulty: difficulty level of the row

default = easy (one alias step, canonical format). medium (chain depth 2) and hard (chain depth 3) add longer alias chains and format variation.

Dataset Sources

Repository: https://github.com/aieng-lab/iend-study
Generated with study.data.synthetic in the repository above; no external data.

Dataset Creation

Variable names are sampled without replacement from the 26 lowercase Latin letters; values come from the same split-disjoint 75-value vocabulary as gradiend-key-value. Easy examples contain one alias step and zero or one unrelated distractor assignment. If a distractor is present its value is the alternative, otherwise a different value from the split-local vocabulary is used. Alias and distractor variables are distinct, distractor values never equal the target, and duplicate masked contexts are removed.

Bias, Risks, and Limitations

Fully synthetic and templated; it probes one narrow symbolic mechanism, not general reasoning, and results need not transfer to natural text. Value/token vocabularies are small and hand-curated.

Citation

BibTeX: TODO

Downloads last month: -