File size: 3,023 Bytes
634e51e
 
6af6b10
 
634e51e
 
6af6b10
634e51e
 
 
6af6b10
634e51e
6af6b10
634e51e
 
6af6b10
634e51e
 
 
6af6b10
 
 
 
 
 
 
 
 
634e51e
 
 
6af6b10
 
 
634e51e
 
 
6af6b10
 
 
634e51e
 
 
6af6b10
 
 
634e51e
 
 
6af6b10
 
 
 
 
 
 
 
 
634e51e
 
 
 
6af6b10
634e51e
 
 
 
 
5f631f2
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
{
  "schema_version": 1,
  "title": "Reproduction: Context-free Recognition with Transformers",
  "emoji": "🧩",
  "space_id": "snaykey/repro-gp-thompson-sampling-regret-bayesian-optimization",
  "paper": {
    "openreview_id": "JOyxs9ElI7"
  },
  "tags": [
    "icml2026-repro",
    "paper-JOyxs9ElI7"
  ],
  "updated_at": "2026-07-31T00:00:00+00:00",
  "root": {
    "slug": "index",
    "title": "Reproduction: Context-free Recognition with Transformers",
    "file": "pages/index.md",
    "children": [
      {
        "slug": "claim-1-theorem-3-1-general-cfl",
        "title": "Theorem 3.1 proves that all context-free languages can be recognized by looped transformers using O(log(n)) looping layers and O(n^6) padding tokens (Section 3, Theorem 3.1).",
        "file": "pages/claim-1-theorem-3-1-general-cfl/page.md",
        "children": []
      },
      {
        "slug": "claim-2-theorem-4-1-unambiguous",
        "title": "Theorem 4.1 shows unambiguous context-free languages can be recognized with fewer resources than the general case: O(log^2(n)) looping layers and O(n^3) padding tokens (Section 4, Theorem 4.1).",
        "file": "pages/claim-2-theorem-4-1-unambiguous/page.md",
        "children": []
      },
      {
        "slug": "claim-3-theorem-4-2-linear",
        "title": "Theorem 4.2 further reduces requirements for unambiguous linear context-free languages to O(log(n)) looping layers and O(n^2) padding tokens by exploiting the linearity constraint (Section 4.2, Theorem 4.2).",
        "file": "pages/claim-3-theorem-4-2-linear/page.md",
        "children": []
      },
      {
        "slug": "claim-4-lemma-4-1-pebbling",
        "title": "Lemma 4.1 shows that Boolean formula evaluation can be computed in O(log(n)) steps via a parallel pebbling argument, which underlies the padding-token constructions used in the main theorems (Section 4, Lemma 4.1).",
        "file": "pages/claim-4-lemma-4-1-pebbling/page.md",
        "children": []
      },
      {
        "slug": "claim-5-corollary-4-1-bfvp",
        "title": "Corollary 4.1 shows that Boolean Formula Value Problem (BFVP) recognition requires zero padding tokens while retaining logarithmic depth (Section 4, Corollary 4.1).",
        "file": "pages/claim-5-corollary-4-1-bfvp/page.md",
        "children": []
      },
      {
        "slug": "claim-6-table-1-tradeoff",
        "title": "Table 1 summarizes the padding-versus-depth tradeoff across general, unambiguous, and linear-unambiguous context-free language classes, showing that grammatical restrictions systematically reduce required padding (Table 1).",
        "file": "pages/claim-6-table-1-tradeoff/page.md",
        "children": []
      },
      {
        "slug": "executive-summary",
        "title": "executive-summary",
        "file": "pages/executive-summary/page.md",
        "children": []
      },
      {
        "slug": "conclusion",
        "title": "conclusion",
        "file": "pages/conclusion/page.md",
        "children": []
      }
    ]
  }
}