Downstep 35M: calibrated pitch accent for dictionary-unlisted Japanese compounds

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+figures/accuracy_by_condition.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,201 @@
+---
+language:
+  - ja
+license: apache-2.0
+library_name: downstep
+pipeline_tag: token-classification
+tags:
+  - text-to-speech
+  - japanese
+  - pitch-accent
+  - prosody
+  - accent-phrase
+  - openjtalk
+  - voicevox
+  - style-bert-vits2
+  - calibration
+  - selective-prediction
+  - onnx
+metrics:
+  - exact_match
+  - f1
+---
+
+# Downstep — Japanese pitch accent for compounds your dictionary has never seen
+
+**Getting 東京大学 right is a lookup. The hard part is 新型感染症対策本部.**
+
+Downstep is a 35M-parameter encoder that predicts the **accent nucleus** and **accent-phrase
+boundaries** of a Japanese noun compound from its morphemes. It reports a calibrated
+confidence and, below a threshold, **abstains** — returning the existing front end's answer
+unchanged — so it can be added to an OpenJTalk-based TTS pipeline without replacing it.
+
+Code, benchmark builder and evaluation: **https://github.com/NagaYu/downstep**
+
+---
+
+## What this model is for, and what it is not for
+
+**Use it for** the fallback path: compounds that are *not* entries in your TTS dictionary,
+where a rule-based front end has to compose the accent from parts.
+
+**Do not use it to replace dictionary lookup.** On compounds that *are* in the dictionary this
+model is clearly worse than OpenJTalk (81.3% vs 96.4%), and worse than the dictionary itself
+(100% by definition). That is not a defect — the dictionary is right and there is nothing to
+add. The shipped adapter tries the dictionary first for exactly this reason.
+
+This is a **text-only** model. It reads a surface string and a kana reading; it has never seen
+speech. Accent models trained on thousands of hours of audio are solving a different problem.
+
+---
+
+## Results
+
+Evaluated on 5,414 held-out compounds, scored identically in both conditions. Gold accents
+come from NAIST-jdic and UniDic 2.1.2. Metric: accent-nucleus exact match.
+
+| system | compound **in** dictionary | compound **not** in dictionary |
+|---|---|---|
+| OpenJTalk (rule-based) | **96.4%** | 40.6% |
+| dictionary lookup only | 100.0% | 40.6% (falls back) |
+| **Downstep** | 81.3% | **81.3%** |
+| **Downstep + abstention** | **96.3%** | **81.3%** |
+
+**+40.7 points** on dictionary-unlisted compounds. On the *hard* subset — compounds whose
+final morpheme never appeared in training — **73.3% vs 40.9%**, so the gain is not memorisation.
+OpenJTalk also wrongly splits an unlisted compound into two accent phrases 23.3% of the time;
+Downstep does so 2.2% of the time.
+
+The "not in dictionary" condition is produced by **ablating the compound's own entry** at query
+time so the gold stays real. It is a held-out-entry simulation, not an observation of genuinely
+novel compounds — see Limitations.
+
+**Calibration.** Temperature scaling fitted on dev: ECE **0.048 → 0.011** (T = 1.498).
+
+**n-best.** top-1 81.3%, top-3 **98.0%**, top-5 99.7%. On words whose dictionaries attest
+*more than one* accent, top-1 is 61.2% but top-3 is 98.0%.
+
+**Speed.** 2.0 ms/word (PyTorch, CPU). ONNX int8: 1.3 ms at batch 1, ~3× faster than fp32.
+
+---
+
+## Files
+
+| file | what it is | size |
+|---|---|---|
+| `model.pt` | PyTorch checkpoint (config + weights) | 134 MB |
+| `vocab.json` | **required** — the learned feature id maps | 12 KB |
+| `downstep.onnx` | ONNX fp32, opset 17, dynamic batch and time | 134 MB |
+| `downstep.int8.onnx` | ONNX int8 dynamic quantisation | 34 MB |
+| `calibrator.json`, `calibration.json` | temperature scaler + dev-chosen thresholds | small |
+
+`vocab.json` is not optional. Feature id spaces are learned from the training corpus, so a
+mismatched vocab does not raise — it silently predicts nonsense.
+
+**The int8 build's decisions are identical** to fp32 (nucleus argmax and boundary sign agree
+100%), but its logits deviate enough to fail a 5e-02 tolerance. Since the abstention gate
+thresholds a *confidence*, re-calibrate before using int8 with a threshold rather than
+inheriting the fp32 one.
+
+There is **no GGUF**. llama.cpp dispatches on `general.architecture` and has no `downstep`
+entry, and its token-ids-in / vocabulary-logits-out interface does not describe this model,
+which takes 13 parallel feature streams and returns three per-mora logit vectors. ONNX is the
+export that actually runs.
+
+---
+
+## Usage
+
+```bash
+pip install "downstep[all] @ git+https://github.com/NagaYu/downstep"
+```
+
+```python
+from huggingface_hub import snapshot_download
+from downstep.adapters.openjtalk import DropInAdapter
+
+path = snapshot_download("NagaYu/downstep")
+adapter = DropInAdapter.load(path)          # dictionary first, model for the rest
+
+# VOICEVOX-shaped accent phrases
+adapter.accent_phrases("新型感染症対策本部を設置した")
+
+# NJD features — same keys as pyopenjtalk.run_frontend(), swap it in directly
+adapter.njd_features("新型感染症対策本部を設置した")
+
+# HTS full-context labels
+adapter.full_context_labels("新型感染症対策本部を設置した")
+```
+
+For **VOICEVOX**, substitute `adapter.accent_phrases(text)` where the engine builds its
+`AudioQuery` accent phrases. For **Style-Bert-VITS2**, point its `pyopenjtalk.run_frontend`
+call at `adapter.njd_features`.
+
+**A trap worth knowing.** In HTS full-context labels the accent field `F2` writes 平板 (flat,
+no fall) as `F2 == F1`, *not* as 0 — so `F2 == F1` is ambiguous between 平板 and 尾高 (nucleus
+on the final mora). Writing a literal 0 there silently corrupts every flat-accent word, and
+reading `F2` naively scores OpenJTalk at 70% where the truth is 94%. The adapter converts in
+both directions and raises `AmbiguousF2Error` rather than guessing.
+
+---
+
+## Training
+
+- **Data:** 71,544 compounds from NAIST-jdic, each decomposing into ≥2 morphemes whose
+  readings compose to the compound's reading under 連濁 (rendaku) and 促音化. Decomposition
+  uses fugashi + UniDic — deliberately a *different* dictionary from the gold, so the parts
+  handed to the model are not derived from the answer.
+- **Architecture:** 10-layer pre-LN transformer encoder, d_model 512, 8 heads, d_ff 2048.
+  13 per-mora integer feature streams (mora identity, position, morpheme boundary, word
+  origin, each component's own dictionary accent and sandhi class, …) summed as embeddings.
+  Three per-mora heads: boundary, nucleus, heiban.
+- **Objective:** boundary BCE + a per-phrase softmax over (each mora position, plus a
+  no-nucleus option). That single softmax per phrase is the training-time counterpart of the
+  decoder's hard constraint that an accent phrase carries **at most one nucleus**.
+- **Run:** 12 epochs, batch 256, AdamW 3e-4 with cosine decay, 82 minutes on Apple MPS.
+- 25% of training items are **constructed** two-phrase examples (two compounds joined by a
+  case particle), because every real benchmark compound is a single accent phrase and the
+  boundary head would otherwise see only negative examples.
+
+---
+
+## Limitations
+
+- **No human annotation exists anywhere in this project.** Every label is a dictionary entry.
+  The compounds this model is built for are precisely the ones no dictionary contains, and
+  their accent can only be settled by asking native speakers. The repository ships the full
+  annotation protocol, multi-annotator format, agreement statistics and an extractor that
+  emits novel compounds **unlabeled** — but nothing here has been checked by a person.
+- **The unlisted condition is simulated.** Held-out-entry ablation on lexicalised compounds
+  is what makes real gold available; it is not an observation of genuinely novel compounds.
+- **Per-word non-degradation is expensive.** At full coverage the model makes 256 of 5,414
+  words worse while making 2,462 better (~10:1). Driving individual regressions to zero
+  requires abstaining on ~93% of items, which scores 42.2% instead of 81.3%. In aggregate the
+  combined system is at or above the baseline at every coverage level; individual words are
+  not. Choose the threshold deliberately.
+- **Proper nouns are 66% of the benchmark.** Place and person names are a different, more
+  arbitrary problem. They are always reported as their own stratum; the common-noun result
+  rests on 2,033 test items.
+- **Boundary F1 is not a hard task here** (dev reaches 1.000) because the only positive
+  boundary examples available offline are constructed. The informative boundary number is the
+  wrong-split rate above.
+- **`hard` is 818 items** and its strictest variant is 98. The generalisation result is real
+  but rests on a small sample.
+
+## On variation
+
+Standard (Tokyo) accent is the default **because that is what the two gold lexicons encode** —
+a statement about available data, not about how anyone should speak. Regional and generational
+accent patterns are ordinary variation, not error. Where the dictionaries attest several
+accents for a word, all are counted correct, and n-best exists for the same reason. Non-Tokyo
+systems belong in separate adapters; that interface ships with no invented accent data behind
+it, because no licensed non-Tokyo accent resource was available.
+
+## License and attribution
+
+Code and weights: **Apache-2.0**.
+
+Trained on data derived from **UniDic 2.1.2**, © The UniDic Consortium (NINJAL), used under
+the BSD 3-clause option of its GPL/LGPL/BSD tri-licence, and from the **NAIST Japanese
+Dictionary** (naist-jdic), © Nara Institute of Science and Technology, BSD 3-clause. Neither
+dictionary is redistributed here.

calibration.json

@@ -0,0 +1,19 @@
+{
+  "fitted_on": "benchmarks/_raw_dev_unlisted.json",
+  "n_dev_items": 5441,
+  "temperature": 1.4978117508891504,
+  "ece_before": 0.048161102888937414,
+  "ece_after": 0.011165273365421908,
+  "thresholds": {
+    "max_accuracy": {
+      "threshold": 0.318732,
+      "dev_combined_accuracy": 0.8213563683146481
+    },
+    "zero_regression": {
+      "threshold": 0.984069,
+      "dev_coverage": 0.07755927219261165
+    }
+  },
+  "default_threshold": 0.318732,
+  "note": "Fitted on dev only. scripts/evaluate.py reports the real test-set regression count at these thresholds; a dev-set zero-regression point is an empirical guarantee, not a proof."
+}

calibrator.json

@@ -0,0 +1,10 @@
+{
+  "kind": "temperature",
+  "fit_on": "dev",
+  "n_fit": 5441,
+  "nll_before": 0.4128013098034377,
+  "nll_after": 0.38817239423824257,
+  "temperature": 1.4978117508891504,
+  "mode": "multiclass",
+  "bounds_hit": false
+}

downstep.int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76dc3085a79052726dad6bb2cd1e1efe03ecd71973b8beac3adbd090b2a31b52
+size 36023179

downstep.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:535cb239b7e5ee9b78f254d8d1f4cfa35d773c98dcb78c68494ff324ebdb0c95
+size 140862821

eval_results.json

@@ -0,0 +1,1834 @@
+{
+  "generated_by": "scripts/evaluate.py",
+  "checkpoint": "checkpoints/base",
+  "conditions": {
+    "listed": {
+      "n_items": 9036,
+      "n_reading_matched": 5414,
+      "n_reading_ok_this_condition": 8815,
+      "n_excluded_for_reading": 3622,
+      "openjtalk_seconds": 0.61,
+      "downstep_seconds": 18.13,
+      "strata": {
+        "all": {
+          "n": 5414,
+          "A_openjtalk": 0.9637975618766161,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.022349464351680826,
+          "C_downstep": 0.8134466198743997,
+          "C_wrongly_split": 0.0
+        },
+        "common_nouns": {
+          "n": 2033,
+          "A_openjtalk": 0.9567142154451549,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.017707820954254796,
+          "C_downstep": 0.7968519429414658,
+          "C_wrongly_split": 0.0
+        },
+        "proper_nouns": {
+          "n": 3381,
+          "A_openjtalk": 0.9680567879325643,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.025140490979000295,
+          "C_downstep": 0.8234250221827861,
+          "C_wrongly_split": 0.0
+        },
+        "hard": {
+          "n": 574,
+          "A_openjtalk": 0.9686411149825784,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.0156794425087108,
+          "C_downstep": 0.7334494773519163,
+          "C_wrongly_split": 0.0
+        },
+        "hard_all": {
+          "n": 98,
+          "A_openjtalk": 0.9897959183673469,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.0,
+          "C_downstep": 0.6224489795918368,
+          "C_wrongly_split": 0.0
+        },
+        "variation": {
+          "n": 49,
+          "A_openjtalk": 0.9387755102040817,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.061224489795918366,
+          "C_downstep": 0.6122448979591837,
+          "C_wrongly_split": 0.0
+        },
+        "goshu_和": {
+          "n": 1988,
+          "A_openjtalk": 0.954225352112676,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.02665995975855131,
+          "C_downstep": 0.7907444668008048,
+          "C_wrongly_split": 0.0
+        },
+        "goshu_漢": {
+          "n": 1349,
+          "A_openjtalk": 0.9747961452928094,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.015567086730911787,
+          "C_downstep": 0.8413639733135656,
+          "C_wrongly_split": 0.0
+        },
+        "goshu_外": {
+          "n": 405,
+          "A_openjtalk": 0.9481481481481482,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.024691358024691357,
+          "C_downstep": 0.8222222222222222,
+          "C_wrongly_split": 0.0
+        },
+        "goshu_混": {
+          "n": 41,
+          "A_openjtalk": 1.0,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.0,
+          "C_downstep": 0.8780487804878049,
+          "C_wrongly_split": 0.0
+        },
+        "goshu_固": {
+          "n": 1562,
+          "A_openjtalk": 0.969270166453265,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.023687580025608196,
+          "C_downstep": 0.8149807938540333,
+          "C_wrongly_split": 0.0
+        },
+        "script_kanji": {
+          "n": 2884,
+          "A_openjtalk": 0.9680998613037448,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.025312066574202496,
+          "C_downstep": 0.8311373092926491,
+          "C_wrongly_split": 0.0
+        },
+        "script_katakana": {
+          "n": 683,
+          "A_openjtalk": 0.95900439238653,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.020497803806734993,
+          "C_downstep": 0.7862371888726208,
+          "C_wrongly_split": 0.0
+        },
+        "script_mixed": {
+          "n": 1847,
+          "A_openjtalk": 0.9588521927449919,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.018408229561451002,
+          "C_downstep": 0.7958852192744992,
+          "C_wrongly_split": 0.0
+        },
+        "components_2": {
+          "n": 4595,
+          "A_openjtalk": 0.9608269858541894,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.02372143634385201,
+          "C_downstep": 0.8187159956474429,
+          "C_wrongly_split": 0.0
+        },
+        "components_3": {
+          "n": 735,
+          "A_openjtalk": 0.9782312925170068,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.0163265306122449,
+          "C_downstep": 0.7850340136054422,
+          "C_wrongly_split": 0.0
+        },
+        "components_4": {
+          "n": 80,
+          "A_openjtalk": 1.0,
+          "B_dictionary": 1.0,
+          "A_wrongly_split": 0.0,
+          "C_downstep": 0.775,
+          "C_wrongly_split": 0.0
+        }
+      },
+      "abstention": {
+        "curve": [
+          {
+            "threshold": 0.287819,
+            "coverage": 0.999815293683044,
+            "combined_accuracy": 0.8136313261913557,
+            "covered_accuracy": 0.8135968963606134,
+            "regressions": 955
+          },
+          {
+            "threshold": 0.404532,
+            "coverage": 0.9898411525674178,
+            "combined_accuracy": 0.819541928333949,
+            "covered_accuracy": 0.8186228774025005,
+            "regressions": 921
+          },
+          {
+            "threshold": 0.453785,
+            "coverage": 0.9798670114517917,
+            "combined_accuracy": 0.8254525304765423,
+            "covered_accuracy": 0.8233741753063148,
+            "regressions": 888
+          },
+          {
+            "threshold": 0.48362,
+            "coverage": 0.9698928703361654,
+            "combined_accuracy": 0.8313631326191355,
+            "covered_accuracy": 0.8276518758331747,
+            "regressions": 856
+          },
+          {
+            "threshold": 0.50118,
+            "coverage": 0.9599187292205393,
+            "combined_accuracy": 0.8370890284447728,
+            "covered_accuracy": 0.8322108908985953,
+            "regressions": 824
+          },
+          {
+            "threshold": 0.518026,
+            "coverage": 0.9499445881049132,
+            "combined_accuracy": 0.842260805319542,
+            "covered_accuracy": 0.8362823254909586,
+            "regressions": 795
+          },
+          {
+            "threshold": 0.53355,
+            "coverage": 0.939970446989287,
+            "combined_accuracy": 0.8465090506095309,
+            "covered_accuracy": 0.8400471605423462,
+            "regressions": 771
+          },
+          {
+            "threshold": 0.550858,
+            "coverage": 0.9299963058736609,
+            "combined_accuracy": 0.851496121167344,
+            "covered_accuracy": 0.8438927507447865,
+            "regressions": 743
+          },
+          {
+            "threshold": 0.568295,
+            "coverage": 0.9198374584410787,
+            "combined_accuracy": 0.856483191725157,
+            "covered_accuracy": 0.8481927710843373,
+            "regressions": 714
+          },
+          {
+            "threshold": 0.5827,
+            "coverage": 0.9098633173254526,
+            "combined_accuracy": 0.8598079054303657,
+            "covered_accuracy": 0.8509947218838815,
+            "regressions": 694
+          },
+          {
+            "threshold": 0.595637,
+            "coverage": 0.8998891762098263,
+            "combined_accuracy": 0.8642408570373107,
+            "covered_accuracy": 0.8548850574712644,
+            "regressions": 670
+          },
+          {
+            "threshold": 0.611318,
+            "coverage": 0.8899150350942002,
+            "combined_accuracy": 0.8690432212781677,
+            "covered_accuracy": 0.8590701535907015,
+            "regressions": 643
+          },
+          {
+            "threshold": 0.626212,
+            "coverage": 0.8799408939785741,
+            "combined_accuracy": 0.8723679349833764,
+            "covered_accuracy": 0.8618807724601175,
+            "regressions": 624
+          },
+          {
+            "threshold": 0.640905,
+            "coverage": 0.8699667528629479,
+            "combined_accuracy": 0.8762467676394533,
+            "covered_accuracy": 0.8651804670912951,
+            "regressions": 601
+          },
+          {
+            "threshold": 0.655575,
+            "coverage": 0.8599926117473218,
+            "combined_accuracy": 0.879571481344662,
+            "covered_accuracy": 0.8679123711340206,
+            "regressions": 581
+          },
+          {
+            "threshold": 0.672725,
+            "coverage": 0.8500184706316956,
+            "combined_accuracy": 0.8836350203176949,
+            "covered_accuracy": 0.8720121686223381,
+            "regressions": 556
+          },
+          {
+            "threshold": 0.686029,
+            "coverage": 0.8398596231991134,
+            "combined_accuracy": 0.8869597340229036,
+            "covered_accuracy": 0.8744226962832637,
+            "regressions": 538
+          },
+          {
+            "threshold": 0.700345,
+            "coverage": 0.8298854820834872,
+            "combined_accuracy": 0.8900997414111562,
+            "covered_accuracy": 0.8773647896728244,
+            "regressions": 519
+          },
+          {
+            "threshold": 0.712894,
+            "coverage": 0.8199113409678611,
+            "combined_accuracy": 0.8932397487994089,
+            "covered_accuracy": 0.8806037395809867,
+            "regressions": 500
+          },
+          {
+            "threshold": 0.72452,
+            "coverage": 0.809937199852235,
+            "combined_accuracy": 0.8960103435537495,
+            "covered_accuracy": 0.883466362599772,
+            "regressions": 482
+          },
+          {
+            "threshold": 0.73972,
+            "coverage": 0.7999630587366088,
+            "combined_accuracy": 0.8978574067233099,
+            "covered_accuracy": 0.8857076887554837,
+            "regressions": 468
+          },
+          {
+            "threshold": 0.754053,
+            "coverage": 0.7899889176209827,
+            "combined_accuracy": 0.9017362393793867,
+            "covered_accuracy": 0.8891746551321019,
+            "regressions": 447
+          },
+          {
+            "threshold": 0.765463,
+            "coverage": 0.7800147765053564,
+            "combined_accuracy": 0.9052456594015516,
+            "covered_accuracy": 0.8924934880416765,
+            "regressions": 428
+          },
+          {
+            "threshold": 0.77653,
+            "coverage": 0.7700406353897303,
+            "combined_accuracy": 0.9065386036202439,
+            "covered_accuracy": 0.893739505876709,
+            "regressions": 418
+          },
+          {
+            "threshold": 0.785172,
+            "coverage": 0.7598817879571481,
+            "combined_accuracy": 0.9089397857406724,
+            "covered_accuracy": 0.8964511424404472,
+            "regressions": 403
+          },
+          {
+            "threshold": 0.794848,
+            "coverage": 0.749907646841522,
+            "combined_accuracy": 0.9102327299593647,
+            "covered_accuracy": 0.8972906403940887,
+            "regressions": 396
+          },
+          {
+            "threshold": 0.804995,
+            "coverage": 0.7399335057258958,
+            "combined_accuracy": 0.9124492057628371,
+            "covered_accuracy": 0.8996505242136795,
+            "regressions": 381
+          },
+          {
+            "threshold": 0.817838,
+            "coverage": 0.7299593646102697,
+            "combined_accuracy": 0.9142962689323975,
+            "covered_accuracy": 0.9015688259109311,
+            "regressions": 369
+          },
+          {
+            "threshold": 0.826393,
+            "coverage": 0.7199852234946436,
+            "combined_accuracy": 0.9157739194680458,
+            "covered_accuracy": 0.9030271934325295,
+            "regressions": 360
+          },
+          {
+            "threshold": 0.835651,
+            "coverage": 0.7100110823790173,
+            "combined_accuracy": 0.916697451052826,
+            "covered_accuracy": 0.9037460978147763,
+            "regressions": 352
+          },
+          {
+            "threshold": 0.8443,
+            "coverage": 0.7000369412633912,
+            "combined_accuracy": 0.9183598079054304,
+            "covered_accuracy": 0.9055408970976253,
+            "regressions": 341
+          },
+          {
+            "threshold": 0.853646,
+            "coverage": 0.689878093830809,
+            "combined_accuracy": 0.9203915773919468,
+            "covered_accuracy": 0.9081659973226238,
+            "regressions": 327
+          },
+          {
+            "threshold": 0.861325,
+            "coverage": 0.6799039527151829,
+            "combined_accuracy": 0.9216845216106391,
+            "covered_accuracy": 0.9092637870143983,
+            "regressions": 318
+          },
+          {
+            "threshold": 0.868863,
+            "coverage": 0.6699298115995567,
+            "combined_accuracy": 0.9229774658293314,
+            "covered_accuracy": 0.9098428453267163,
+            "regressions": 311
+          },
+          {
+            "threshold": 0.877202,
+            "coverage": 0.6599556704839306,
+            "combined_accuracy": 0.925932766900628,
+            "covered_accuracy": 0.9129582983487265,
+            "regressions": 295
+          },
+          {
+            "threshold": 0.883933,
+            "coverage": 0.6499815293683044,
+            "combined_accuracy": 0.9274104174362763,
+            "covered_accuracy": 0.9147485080988917,
+            "regressions": 284
+          },
+          {
+            "threshold": 0.890918,
+            "coverage": 0.6400073882526782,
+            "combined_accuracy": 0.9287033616549686,
+            "covered_accuracy": 0.9163059163059163,
+            "regressions": 274
+          },
+          {
+            "threshold": 0.895454,
+            "coverage": 0.6300332471370521,
+            "combined_accuracy": 0.930735131141485,
+            "covered_accuracy": 0.9187921430665494,
+            "regressions": 261
+          },
+          {
+            "threshold": 0.900932,
+            "coverage": 0.6200591060214259,
+            "combined_accuracy": 0.9314739564093092,
+            "covered_accuracy": 0.9192731605600238,
+            "regressions": 256
+          },
+          {
+            "threshold": 0.905733,
+            "coverage": 0.6099002585888438,
+            "combined_accuracy": 0.9335057258958256,
+            "covered_accuracy": 0.922168382798304,
+            "regressions": 242
+          },
+          {
+            "threshold": 0.911608,
+            "coverage": 0.5999261174732176,
+            "combined_accuracy": 0.935352789065386,
+            "covered_accuracy": 0.9245689655172413,
+            "regressions": 232
+          },
+          {
+            "threshold": 0.916809,
+            "coverage": 0.5899519763575914,
+            "combined_accuracy": 0.9368304396010344,
+            "covered_accuracy": 0.926111458985598,
+            "regressions": 224
+          },
+          {
+            "threshold": 0.920427,
+            "coverage": 0.5799778352419652,
+            "combined_accuracy": 0.9379386775027706,
+            "covered_accuracy": 0.9273885350318471,
+            "regressions": 216
+          },
+          {
+            "threshold": 0.925197,
+            "coverage": 0.5700036941263391,
+            "combined_accuracy": 0.9405245659401551,
+            "covered_accuracy": 0.9309786130913804,
+            "regressions": 201
+          },
+          {
+            "threshold": 0.929965,
+            "coverage": 0.560029553010713,
+            "combined_accuracy": 0.9421869227927595,
+            "covered_accuracy": 0.933377308707124,
+            "regressions": 191
+          },
+          {
+            "threshold": 0.933174,
+            "coverage": 0.5500554118950868,
+            "combined_accuracy": 0.9438492796453639,
+            "covered_accuracy": 0.9358629952988583,
+            "regressions": 180
+          },
+          {
+            "threshold": 0.93709,
+            "coverage": 0.5400812707794607,
+            "combined_accuracy": 0.9447728112301441,
+            "covered_accuracy": 0.9374145006839946,
+            "regressions": 174
+          },
+          {
+            "threshold": 0.940724,
+            "coverage": 0.5299224233468784,
+            "combined_accuracy": 0.9458810491318803,
+            "covered_accuracy": 0.9393516904844894,
+            "regressions": 165
+          },
+          {
+            "threshold": 0.944268,
+            "coverage": 0.5199482822312523,
+            "combined_accuracy": 0.9466198743997045,
+            "covered_accuracy": 0.9399644760213144,
+            "regressions": 161
+          },
+          {
+            "threshold": 0.948099,
+            "coverage": 0.5099741411156261,
+            "combined_accuracy": 0.9471739933505726,
+            "covered_accuracy": 0.9406012314378849,
+            "regressions": 156
+          },
+          {
+            "threshold": 0.951436,
+            "coverage": 0.5,
+            "combined_accuracy": 0.9475434059844847,
+            "covered_accuracy": 0.9408939785740672,
+            "regressions": 152
+          },
+          {
+            "threshold": 0.954023,
+            "coverage": 0.49002585888437383,
+            "combined_accuracy": 0.9495751754710011,
+            "covered_accuracy": 0.944214097248398,
+            "regressions": 140
+          },
+          {
+            "threshold": 0.956498,
+            "coverage": 0.48005171776874767,
+            "combined_accuracy": 0.9504987070557813,
+            "covered_accuracy": 0.9453636013851482,
+            "regressions": 134
+          },
+          {
+            "threshold": 0.958537,
+            "coverage": 0.47007757665312155,
+            "combined_accuracy": 0.9514222386405615,
+            "covered_accuracy": 0.9469548133595285,
+            "regressions": 127
+          },
+          {
+            "threshold": 0.960376,
+            "coverage": 0.45991872922053934,
+            "combined_accuracy": 0.9521610639083856,
+            "covered_accuracy": 0.948995983935743,
+            "regressions": 120
+          },
+          {
+            "threshold": 0.963265,
+            "coverage": 0.44994458810491317,
+            "combined_accuracy": 0.9530845954931658,
+            "covered_accuracy": 0.9499178981937603,
+            "regressions": 115
+          },
+          {
+            "threshold": 0.965379,
+            "coverage": 0.43997044698928706,
+            "combined_accuracy": 0.9532693018101219,
+            "covered_accuracy": 0.9500419815281276,
+            "regressions": 113
+          },
+          {
+            "threshold": 0.96692,
+            "coverage": 0.4299963058736609,
+            "combined_accuracy": 0.9541928333949021,
+            "covered_accuracy": 0.9510309278350515,
+            "regressions": 108
+          },
+          {
+            "threshold": 0.968873,
+            "coverage": 0.4200221647580347,
+            "combined_accuracy": 0.9549316586627262,
+            "covered_accuracy": 0.951627088830255,
+            "regressions": 104
+          },
+          {
+            "threshold": 0.970763,
+            "coverage": 0.41004802364240855,
+            "combined_accuracy": 0.9554857776135944,
+            "covered_accuracy": 0.9518018018018019,
+            "regressions": 101
+          },
+          {
+            "threshold": 0.972469,
+            "coverage": 0.40007388252678244,
+            "combined_accuracy": 0.9558551902475064,
+            "covered_accuracy": 0.9519852262234534,
+            "regressions": 98
+          },
+          {
+            "threshold": 0.974047,
+            "coverage": 0.3900997414111563,
+            "combined_accuracy": 0.9562246028814185,
+            "covered_accuracy": 0.9535984848484849,
+            "regressions": 94
+          },
+          {
+            "threshold": 0.975516,
+            "coverage": 0.37994089397857406,
+            "combined_accuracy": 0.9564093091983746,
+            "covered_accuracy": 0.9533300923675255,
+            "regressions": 92
+          },
+          {
+            "threshold": 0.976913,
+            "coverage": 0.3699667528629479,
+            "combined_accuracy": 0.9571481344661987,
+            "covered_accuracy": 0.9550673989016475,
+            "regressions": 86
+          },
+          {
+            "threshold": 0.978492,
+            "coverage": 0.3599926117473218,
+            "combined_accuracy": 0.9577022534170668,
+            "covered_accuracy": 0.9563878912262699,
+            "regressions": 82
+          },
+          {
+            "threshold": 0.979541,
+            "coverage": 0.3500184706316956,
+            "combined_accuracy": 0.9573328407831548,
+            "covered_accuracy": 0.9551451187335093,
+            "regressions": 82
+          },
+          {
+            "threshold": 0.980824,
+            "coverage": 0.34004432951606944,
+            "combined_accuracy": 0.9578869597340229,
+            "covered_accuracy": 0.955458989679522,
+            "regressions": 79
+          },
+          {
+            "threshold": 0.981792,
+            "coverage": 0.3300701884004433,
+            "combined_accuracy": 0.9580716660509789,
+            "covered_accuracy": 0.9557918298824846,
+            "regressions": 77
+          },
+          {
+            "threshold": 0.98304,
+            "coverage": 0.32009604728481716,
+            "combined_accuracy": 0.958256372367935,
+            "covered_accuracy": 0.9561454125793422,
+            "regressions": 74
+          },
+          {
+            "threshold": 0.984012,
+            "coverage": 0.30993719985223495,
+            "combined_accuracy": 0.9591799039527151,
+            "covered_accuracy": 0.9576877234803337,
+            "regressions": 69
+          },
+          {
+            "threshold": 0.984861,
+            "coverage": 0.2999630587366088,
+            "combined_accuracy": 0.9593646102696712,
+            "covered_accuracy": 0.9568965517241379,
+            "regressions": 68
+          },
+          {
+            "threshold": 0.985914,
+            "coverage": 0.2899889176209826,
+            "combined_accuracy": 0.9593646102696712,
+            "covered_accuracy": 0.9573248407643312,
+            "regressions": 66
+          },
+          {
+            "threshold": 0.986784,
+            "coverage": 0.2800147765053565,
+            "combined_accuracy": 0.9602881418544514,
+            "covered_accuracy": 0.9597625329815304,
+            "regressions": 60
+          },
+          {
+            "threshold": 0.98768,
+            "coverage": 0.27004063538973033,
+            "combined_accuracy": 0.9601034355374953,
+            "covered_accuracy": 0.9596443228454172,
+            "regressions": 58
+          },
+          {
+            "threshold": 0.988424,
+            "coverage": 0.26006649427410417,
+            "combined_accuracy": 0.9612116734392316,
+            "covered_accuracy": 0.9630681818181818,
+            "regressions": 51
+          },
+          {
+            "threshold": 0.989225,
+            "coverage": 0.250092353158478,
+            "combined_accuracy": 0.9623199113409678,
+            "covered_accuracy": 0.96602658788774,
+            "regressions": 45
+          },
+          {
+            "threshold": 0.989847,
+            "coverage": 0.24011821204285186,
+            "combined_accuracy": 0.9626893239748799,
+            "covered_accuracy": 0.9661538461538461,
+            "regressions": 43
+          },
+          {
+            "threshold": 0.990648,
+            "coverage": 0.22995936461026967,
+            "combined_accuracy": 0.9625046176579239,
+            "covered_accuracy": 0.9654618473895582,
+            "regressions": 42
+          },
+          {
+            "threshold": 0.991326,
+            "coverage": 0.21998522349464353,
+            "combined_accuracy": 0.9625046176579239,
+            "covered_accuracy": 0.964735516372796,
+            "regressions": 41
+          },
+          {
+            "threshold": 0.991868,
+            "coverage": 0.21001108237901736,
+            "combined_accuracy": 0.9626893239748799,
+            "covered_accuracy": 0.9665787159190853,
+            "regressions": 37
+          },
+          {
+            "threshold": 0.992456,
+            "coverage": 0.20003694126339122,
+            "combined_accuracy": 0.963243442925748,
+            "covered_accuracy": 0.9676823638042474,
+            "regressions": 34
+          },
+          {
+            "threshold": 0.993025,
+            "coverage": 0.19006280014776505,
+            "combined_accuracy": 0.963058736608792,
+            "covered_accuracy": 0.966958211856171,
+            "regressions": 33
+          },
+          {
+            "threshold": 0.993511,
+            "coverage": 0.1800886590321389,
+            "combined_accuracy": 0.9626893239748799,
+            "covered_accuracy": 0.9661538461538461,
+            "regressions": 32
+          },
+          {
+            "threshold": 0.993984,
+            "coverage": 0.17011451791651275,
+            "combined_accuracy": 0.963058736608792,
+            "covered_accuracy": 0.9674267100977199,
+            "regressions": 29
+          },
+          {
+            "threshold": 0.99437,
+            "coverage": 0.15995567048393056,
+            "combined_accuracy": 0.963243442925748,
+            "covered_accuracy": 0.9665127020785219,
+            "regressions": 28
+          },
+          {
+            "threshold": 0.994859,
+            "coverage": 0.15016623568526044,
+            "combined_accuracy": 0.9636128555596601,
+            "covered_accuracy": 0.9692496924969249,
+            "regressions": 24
+          },
+          {
+            "threshold": 0.995294,
+            "coverage": 0.14000738825267825,
+            "combined_accuracy": 0.9639822681935722,
+            "covered_accuracy": 0.9709762532981531,
+            "regressions": 22
+          },
+          {
+            "threshold": 0.995697,
+            "coverage": 0.13021795345400813,
+            "combined_accuracy": 0.9645363871444403,
+            "covered_accuracy": 0.9730496453900709,
+            "regressions": 19
+          },
+          {
+            "threshold": 0.996291,
+            "coverage": 0.12005910602142593,
+            "combined_accuracy": 0.9649057997783524,
+            "covered_accuracy": 0.9738461538461538,
+            "regressions": 17
+          },
+          {
+            "threshold": 0.996686,
+            "coverage": 0.11008496490579978,
+            "combined_accuracy": 0.9643516808274843,
+            "covered_accuracy": 0.9714765100671141,
+            "regressions": 17
+          },
+          {
+            "threshold": 0.99705,
+            "coverage": 0.10011082379017362,
+            "combined_accuracy": 0.9645363871444403,
+            "covered_accuracy": 0.974169741697417,
+            "regressions": 14
+          },
+          {
+            "threshold": 0.997386,
+            "coverage": 0.09013668267454747,
+            "combined_accuracy": 0.9643516808274843,
+            "covered_accuracy": 0.9774590163934426,
+            "regressions": 11
+          },
+          {
+            "threshold": 0.997703,
+            "coverage": 0.08016254155892132,
+            "combined_accuracy": 0.9641669745105282,
+            "covered_accuracy": 0.9746543778801844,
+            "regressions": 11
+          },
+          {
+            "threshold": 0.997964,
+            "coverage": 0.07000369412633912,
+            "combined_accuracy": 0.9645363871444403,
+            "covered_accuracy": 0.978891820580475,
+            "regressions": 8
+          },
+          {
+            "threshold": 0.998248,
+            "coverage": 0.060029553010712965,
+            "combined_accuracy": 0.9652752124122645,
+            "covered_accuracy": 0.9907692307692307,
+            "regressions": 3
+          },
+          {
+            "threshold": 0.998595,
+            "coverage": 0.05005541189508681,
+            "combined_accuracy": 0.9649057997783524,
+            "covered_accuracy": 0.988929889298893,
+            "regressions": 3
+          },
+          {
+            "threshold": 0.998848,
+            "coverage": 0.04008127077946066,
+            "combined_accuracy": 0.9647210934613963,
+            "covered_accuracy": 0.9953917050691244,
+            "regressions": 1
+          },
+          {
+            "threshold": 0.99905,
+            "coverage": 0.029922423346878464,
+            "combined_accuracy": 0.9643516808274843,
+            "covered_accuracy": 0.9938271604938271,
+            "regressions": 1
+          },
+          {
+            "threshold": 0.99924,
+            "coverage": 0.020132988548208348,
+            "combined_accuracy": 0.9641669745105282,
+            "covered_accuracy": 0.9908256880733946,
+            "regressions": 1
+          },
+          {
+            "threshold": 0.999454,
+            "coverage": 0.009974141115626155,
+            "combined_accuracy": 0.9643516808274843,
+            "covered_accuracy": 1.0,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.99978,
+            "coverage": 0.0,
+            "combined_accuracy": 0.9637975618766161,
+            "covered_accuracy": null,
+            "regressions": 0
+          }
+        ],
+        "baseline_accuracy": 0.9637975618766161,
+        "no_abstention_accuracy": 0.8134466198743997,
+        "best_combined": {
+          "threshold": 0.998248,
+          "coverage": 0.060029553010712965,
+          "combined_accuracy": 0.9652752124122645,
+          "covered_accuracy": 0.9907692307692307,
+          "regressions": 3
+        },
+        "conservative_zero_regression": {
+          "threshold": 0.999454,
+          "coverage": 0.009974141115626155,
+          "combined_accuracy": 0.9643516808274843,
+          "covered_accuracy": 1.0,
+          "regressions": 0
+        },
+        "ece": 0.0576062929816676
+      },
+      "nbest_coverage": {
+        "all": {
+          "1": 0.8134466198743997,
+          "2": 0.9436645733284078,
+          "3": 0.9804211304026598,
+          "4": 0.9918729220539343,
+          "5": 0.9970446989287034
+        },
+        "variation_only": {
+          "1": 0.6122448979591837,
+          "2": 0.9387755102040817,
+          "3": 0.9795918367346939,
+          "4": 1.0,
+          "5": 1.0
+        },
+        "n_variation": 49
+      },
+      "operating_point_dev_chosen": {
+        "chosen_on": "dev split, same condition",
+        "max_accuracy": {
+          "threshold": 0.994487,
+          "coverage": 0.15755448836350203,
+          "combined_accuracy": 0.963243442925748,
+          "regressions": 27,
+          "improvements": 24,
+          "n": 5414
+        },
+        "zero_regression_on_dev": {
+          "threshold": 0.999395,
+          "coverage": 0.013668267454746953,
+          "combined_accuracy": 0.9641669745105282,
+          "regressions": 1,
+          "improvements": 3,
+          "n": 5414
+        },
+        "dev_zero_regression_coverage": 0.015083798882681564
+      }
+    },
+    "unlisted": {
+      "n_items": 9036,
+      "n_reading_matched": 5414,
+      "n_reading_ok_this_condition": 5462,
+      "n_excluded_for_reading": 3622,
+      "openjtalk_seconds": 4.07,
+      "downstep_seconds": 18.16,
+      "strata": {
+        "all": {
+          "n": 5414,
+          "A_openjtalk": 0.4059844846693757,
+          "B_dictionary": 0.4059844846693757,
+          "A_wrongly_split": 0.2332840783154784,
+          "C_downstep": 0.8134466198743997,
+          "C_wrongly_split": 0.0
+        },
+        "common_nouns": {
+          "n": 2033,
+          "A_openjtalk": 0.31431382193802265,
+          "B_dictionary": 0.31431382193802265,
+          "A_wrongly_split": 0.2867683226758485,
+          "C_downstep": 0.7968519429414658,
+          "C_wrongly_split": 0.0
+        },
+        "proper_nouns": {
+          "n": 3381,
+          "A_openjtalk": 0.46110618160307604,
+          "B_dictionary": 0.46110618160307604,
+          "A_wrongly_split": 0.20112392783200236,
+          "C_downstep": 0.8234250221827861,
+          "C_wrongly_split": 0.0
+        },
+        "hard": {
+          "n": 574,
+          "A_openjtalk": 0.4094076655052265,
+          "B_dictionary": 0.4094076655052265,
+          "A_wrongly_split": 0.21602787456445993,
+          "C_downstep": 0.7334494773519163,
+          "C_wrongly_split": 0.0
+        },
+        "hard_all": {
+          "n": 98,
+          "A_openjtalk": 0.3163265306122449,
+          "B_dictionary": 0.3163265306122449,
+          "A_wrongly_split": 0.19387755102040816,
+          "C_downstep": 0.6224489795918368,
+          "C_wrongly_split": 0.0
+        },
+        "variation": {
+          "n": 49,
+          "A_openjtalk": 0.40816326530612246,
+          "B_dictionary": 0.40816326530612246,
+          "A_wrongly_split": 0.22448979591836735,
+          "C_downstep": 0.6122448979591837,
+          "C_wrongly_split": 0.0
+        },
+        "goshu_和": {
+          "n": 1988,
+          "A_openjtalk": 0.3631790744466801,
+          "B_dictionary": 0.3631790744466801,
+          "A_wrongly_split": 0.23943661971830985,
+          "C_downstep": 0.7907444668008048,
+          "C_wrongly_split": 0.0
+        },
+        "goshu_漢": {
+          "n": 1349,
+          "A_openjtalk": 0.42031134173461826,
+          "B_dictionary": 0.42031134173461826,
+          "A_wrongly_split": 0.21793921423276502,
+          "C_downstep": 0.8413639733135656,
+          "C_wrongly_split": 0.0
+        },
+        "goshu_外": {
+          "n": 405,
+          "A_openjtalk": 0.5506172839506173,
+          "B_dictionary": 0.5506172839506173,
+          "A_wrongly_split": 0.14814814814814814,
+          "C_downstep": 0.8222222222222222,
+          "C_wrongly_split": 0.0
+        },
+        "goshu_混": {
+          "n": 41,
+          "A_openjtalk": 0.24390243902439024,
+          "B_dictionary": 0.24390243902439024,
+          "A_wrongly_split": 0.24390243902439024,
+          "C_downstep": 0.8780487804878049,
+          "C_wrongly_split": 0.0
+        },
+        "goshu_固": {
+          "n": 1562,
+          "A_openjtalk": 0.41037131882202305,
+          "B_dictionary": 0.41037131882202305,
+          "A_wrongly_split": 0.26568501920614596,
+          "C_downstep": 0.8149807938540333,
+          "C_wrongly_split": 0.0
+        },
+        "script_kanji": {
+          "n": 2884,
+          "A_openjtalk": 0.4791955617198336,
+          "B_dictionary": 0.4791955617198336,
+          "A_wrongly_split": 0.17094313453536755,
+          "C_downstep": 0.8311373092926491,
+          "C_wrongly_split": 0.0
+        },
+        "script_katakana": {
+          "n": 683,
+          "A_openjtalk": 0.424597364568082,
+          "B_dictionary": 0.424597364568082,
+          "A_wrongly_split": 0.22547584187408493,
+          "C_downstep": 0.7862371888726208,
+          "C_wrongly_split": 0.0
+        },
+        "script_mixed": {
+          "n": 1847,
+          "A_openjtalk": 0.28478613968597727,
+          "B_dictionary": 0.28478613968597727,
+          "A_wrongly_split": 0.3335138061721711,
+          "C_downstep": 0.7958852192744992,
+          "C_wrongly_split": 0.0
+        },
+        "components_2": {
+          "n": 4595,
+          "A_openjtalk": 0.41653971708378673,
+          "B_dictionary": 0.41653971708378673,
+          "A_wrongly_split": 0.2132752992383025,
+          "C_downstep": 0.8187159956474429,
+          "C_wrongly_split": 0.0
+        },
+        "components_3": {
+          "n": 735,
+          "A_openjtalk": 0.363265306122449,
+          "B_dictionary": 0.363265306122449,
+          "A_wrongly_split": 0.3197278911564626,
+          "C_downstep": 0.7850340136054422,
+          "C_wrongly_split": 0.0
+        },
+        "components_4": {
+          "n": 80,
+          "A_openjtalk": 0.2125,
+          "B_dictionary": 0.2125,
+          "A_wrongly_split": 0.55,
+          "C_downstep": 0.775,
+          "C_wrongly_split": 0.0
+        }
+      },
+      "abstention": {
+        "curve": [
+          {
+            "threshold": 0.287819,
+            "coverage": 0.999815293683044,
+            "combined_accuracy": 0.8134466198743997,
+            "covered_accuracy": 0.8135968963606134,
+            "regressions": 256
+          },
+          {
+            "threshold": 0.404532,
+            "coverage": 0.9898411525674178,
+            "combined_accuracy": 0.8127077946065755,
+            "covered_accuracy": 0.8186228774025005,
+            "regressions": 248
+          },
+          {
+            "threshold": 0.453785,
+            "coverage": 0.9798670114517917,
+            "combined_accuracy": 0.8104913188031031,
+            "covered_accuracy": 0.8233741753063148,
+            "regressions": 244
+          },
+          {
+            "threshold": 0.48362,
+            "coverage": 0.9698928703361654,
+            "combined_accuracy": 0.8090136682674548,
+            "covered_accuracy": 0.8276518758331747,
+            "regressions": 235
+          },
+          {
+            "threshold": 0.50118,
+            "coverage": 0.9599187292205393,
+            "combined_accuracy": 0.8084595493165866,
+            "covered_accuracy": 0.8322108908985953,
+            "regressions": 221
+          },
+          {
+            "threshold": 0.518026,
+            "coverage": 0.9499445881049132,
+            "combined_accuracy": 0.8064277798300702,
+            "covered_accuracy": 0.8362823254909586,
+            "regressions": 215
+          },
+          {
+            "threshold": 0.53355,
+            "coverage": 0.939970446989287,
+            "combined_accuracy": 0.8051348356113779,
+            "covered_accuracy": 0.8400471605423462,
+            "regressions": 209
+          },
+          {
+            "threshold": 0.550858,
+            "coverage": 0.9299963058736609,
+            "combined_accuracy": 0.8040265977096417,
+            "covered_accuracy": 0.8438927507447865,
+            "regressions": 197
+          },
+          {
+            "threshold": 0.568295,
+            "coverage": 0.9198374584410787,
+            "combined_accuracy": 0.8018101219061692,
+            "covered_accuracy": 0.8481927710843373,
+            "regressions": 191
+          },
+          {
+            "threshold": 0.5827,
+            "coverage": 0.9098633173254526,
+            "combined_accuracy": 0.7984854082009605,
+            "covered_accuracy": 0.8509947218838815,
+            "regressions": 184
+          },
+          {
+            "threshold": 0.595637,
+            "coverage": 0.8998891762098263,
+            "combined_accuracy": 0.7964536387144441,
+            "covered_accuracy": 0.8548850574712644,
+            "regressions": 178
+          },
+          {
+            "threshold": 0.611318,
+            "coverage": 0.8899150350942002,
+            "combined_accuracy": 0.7953454008127078,
+            "covered_accuracy": 0.8590701535907015,
+            "regressions": 168
+          },
+          {
+            "threshold": 0.626212,
+            "coverage": 0.8799408939785741,
+            "combined_accuracy": 0.791466568156631,
+            "covered_accuracy": 0.8618807724601175,
+            "regressions": 162
+          },
+          {
+            "threshold": 0.640905,
+            "coverage": 0.8699667528629479,
+            "combined_accuracy": 0.7885112670853344,
+            "covered_accuracy": 0.8651804670912951,
+            "regressions": 154
+          },
+          {
+            "threshold": 0.655575,
+            "coverage": 0.8599926117473218,
+            "combined_accuracy": 0.7859253786479498,
+            "covered_accuracy": 0.8679123711340206,
+            "regressions": 144
+          },
+          {
+            "threshold": 0.672725,
+            "coverage": 0.8500184706316956,
+            "combined_accuracy": 0.7837089028444773,
+            "covered_accuracy": 0.8720121686223381,
+            "regressions": 137
+          },
+          {
+            "threshold": 0.686029,
+            "coverage": 0.8398596231991134,
+            "combined_accuracy": 0.7805688954562247,
+            "covered_accuracy": 0.8744226962832637,
+            "regressions": 129
+          },
+          {
+            "threshold": 0.700345,
+            "coverage": 0.8298854820834872,
+            "combined_accuracy": 0.7768747691171038,
+            "covered_accuracy": 0.8773647896728244,
+            "regressions": 124
+          },
+          {
+            "threshold": 0.712894,
+            "coverage": 0.8199113409678611,
+            "combined_accuracy": 0.773180642777983,
+            "covered_accuracy": 0.8806037395809867,
+            "regressions": 117
+          },
+          {
+            "threshold": 0.72452,
+            "coverage": 0.809937199852235,
+            "combined_accuracy": 0.7698559290727743,
+            "covered_accuracy": 0.883466362599772,
+            "regressions": 110
+          },
+          {
+            "threshold": 0.73972,
+            "coverage": 0.7999630587366088,
+            "combined_accuracy": 0.7656076837827853,
+            "covered_accuracy": 0.8857076887554837,
+            "regressions": 106
+          },
+          {
+            "threshold": 0.754053,
+            "coverage": 0.7899889176209827,
+            "combined_accuracy": 0.7615441448097525,
+            "covered_accuracy": 0.8891746551321019,
+            "regressions": 102
+          },
+          {
+            "threshold": 0.765463,
+            "coverage": 0.7800147765053564,
+            "combined_accuracy": 0.7578500184706317,
+            "covered_accuracy": 0.8924934880416765,
+            "regressions": 97
+          },
+          {
+            "threshold": 0.77653,
+            "coverage": 0.7700406353897303,
+            "combined_accuracy": 0.7524935352789065,
+            "covered_accuracy": 0.893739505876709,
+            "regressions": 93
+          },
+          {
+            "threshold": 0.785172,
+            "coverage": 0.7598817879571481,
+            "combined_accuracy": 0.7475064647210935,
+            "covered_accuracy": 0.8964511424404472,
+            "regressions": 90
+          },
+          {
+            "threshold": 0.794848,
+            "coverage": 0.749907646841522,
+            "combined_accuracy": 0.7421499815293683,
+            "covered_accuracy": 0.8972906403940887,
+            "regressions": 86
+          },
+          {
+            "threshold": 0.804995,
+            "coverage": 0.7399335057258958,
+            "combined_accuracy": 0.7380864425563354,
+            "covered_accuracy": 0.8996505242136795,
+            "regressions": 83
+          },
+          {
+            "threshold": 0.817838,
+            "coverage": 0.7299593646102697,
+            "combined_accuracy": 0.7340229035833026,
+            "covered_accuracy": 0.9015688259109311,
+            "regressions": 81
+          },
+          {
+            "threshold": 0.826393,
+            "coverage": 0.7199852234946436,
+            "combined_accuracy": 0.7286664203915774,
+            "covered_accuracy": 0.9030271934325295,
+            "regressions": 80
+          },
+          {
+            "threshold": 0.835651,
+            "coverage": 0.7100110823790173,
+            "combined_accuracy": 0.7227558182489842,
+            "covered_accuracy": 0.9037460978147763,
+            "regressions": 79
+          },
+          {
+            "threshold": 0.8443,
+            "coverage": 0.7000369412633912,
+            "combined_accuracy": 0.7183228666420391,
+            "covered_accuracy": 0.9055408970976253,
+            "regressions": 75
+          },
+          {
+            "threshold": 0.853646,
+            "coverage": 0.689878093830809,
+            "combined_accuracy": 0.7137052087181381,
+            "covered_accuracy": 0.9081659973226238,
+            "regressions": 73
+          },
+          {
+            "threshold": 0.861325,
+            "coverage": 0.6799039527151829,
+            "combined_accuracy": 0.708533431843369,
+            "covered_accuracy": 0.9092637870143983,
+            "regressions": 67
+          },
+          {
+            "threshold": 0.868863,
+            "coverage": 0.6699298115995567,
+            "combined_accuracy": 0.7026228297007757,
+            "covered_accuracy": 0.9098428453267163,
+            "regressions": 65
+          },
+          {
+            "threshold": 0.877202,
+            "coverage": 0.6599556704839306,
+            "combined_accuracy": 0.6981898780938308,
+            "covered_accuracy": 0.9129582983487265,
+            "regressions": 60
+          },
+          {
+            "threshold": 0.883933,
+            "coverage": 0.6499815293683044,
+            "combined_accuracy": 0.6928333949021056,
+            "covered_accuracy": 0.9147485080988917,
+            "regressions": 57
+          },
+          {
+            "threshold": 0.890918,
+            "coverage": 0.6400073882526782,
+            "combined_accuracy": 0.6878463243442926,
+            "covered_accuracy": 0.9163059163059163,
+            "regressions": 56
+          },
+          {
+            "threshold": 0.895454,
+            "coverage": 0.6300332471370521,
+            "combined_accuracy": 0.6839674916882157,
+            "covered_accuracy": 0.9187921430665494,
+            "regressions": 52
+          },
+          {
+            "threshold": 0.900932,
+            "coverage": 0.6200591060214259,
+            "combined_accuracy": 0.6782415958625785,
+            "covered_accuracy": 0.9192731605600238,
+            "regressions": 51
+          },
+          {
+            "threshold": 0.905733,
+            "coverage": 0.6099002585888438,
+            "combined_accuracy": 0.6738086442556336,
+            "covered_accuracy": 0.922168382798304,
+            "regressions": 50
+          },
+          {
+            "threshold": 0.911608,
+            "coverage": 0.5999261174732176,
+            "combined_accuracy": 0.6695603989656447,
+            "covered_accuracy": 0.9245689655172413,
+            "regressions": 48
+          },
+          {
+            "threshold": 0.916809,
+            "coverage": 0.5899519763575914,
+            "combined_accuracy": 0.6638345031400074,
+            "covered_accuracy": 0.926111458985598,
+            "regressions": 45
+          },
+          {
+            "threshold": 0.920427,
+            "coverage": 0.5799778352419652,
+            "combined_accuracy": 0.657739194680458,
+            "covered_accuracy": 0.9273885350318471,
+            "regressions": 42
+          },
+          {
+            "threshold": 0.925197,
+            "coverage": 0.5700036941263391,
+            "combined_accuracy": 0.6536756557074251,
+            "covered_accuracy": 0.9309786130913804,
+            "regressions": 36
+          },
+          {
+            "threshold": 0.929965,
+            "coverage": 0.560029553010713,
+            "combined_accuracy": 0.648503878832656,
+            "covered_accuracy": 0.933377308707124,
+            "regressions": 34
+          },
+          {
+            "threshold": 0.933174,
+            "coverage": 0.5500554118950868,
+            "combined_accuracy": 0.6433321019578869,
+            "covered_accuracy": 0.9358629952988583,
+            "regressions": 31
+          },
+          {
+            "threshold": 0.93709,
+            "coverage": 0.5400812707794607,
+            "combined_accuracy": 0.6377909124492057,
+            "covered_accuracy": 0.9374145006839946,
+            "regressions": 30
+          },
+          {
+            "threshold": 0.940724,
+            "coverage": 0.5299224233468784,
+            "combined_accuracy": 0.6328038418913927,
+            "covered_accuracy": 0.9393516904844894,
+            "regressions": 26
+          },
+          {
+            "threshold": 0.944268,
+            "coverage": 0.5199482822312523,
+            "combined_accuracy": 0.6268932397487994,
+            "covered_accuracy": 0.9399644760213144,
+            "regressions": 25
+          },
+          {
+            "threshold": 0.948099,
+            "coverage": 0.5099741411156261,
+            "combined_accuracy": 0.6215367565570743,
+            "covered_accuracy": 0.9406012314378849,
+            "regressions": 24
+          },
+          {
+            "threshold": 0.951436,
+            "coverage": 0.5,
+            "combined_accuracy": 0.6161802733653491,
+            "covered_accuracy": 0.9408939785740672,
+            "regressions": 24
+          },
+          {
+            "threshold": 0.954023,
+            "coverage": 0.49002585888437383,
+            "combined_accuracy": 0.6126708533431844,
+            "covered_accuracy": 0.944214097248398,
+            "regressions": 20
+          },
+          {
+            "threshold": 0.956498,
+            "coverage": 0.48005171776874767,
+            "combined_accuracy": 0.6078684891023273,
+            "covered_accuracy": 0.9453636013851482,
+            "regressions": 18
+          },
+          {
+            "threshold": 0.958537,
+            "coverage": 0.47007757665312155,
+            "combined_accuracy": 0.6026967122275582,
+            "covered_accuracy": 0.9469548133595285,
+            "regressions": 15
+          },
+          {
+            "threshold": 0.960376,
+            "coverage": 0.45991872922053934,
+            "combined_accuracy": 0.597709641669745,
+            "covered_accuracy": 0.948995983935743,
+            "regressions": 13
+          },
+          {
+            "threshold": 0.963265,
+            "coverage": 0.44994458810491317,
+            "combined_accuracy": 0.5919837458441078,
+            "covered_accuracy": 0.9499178981937603,
+            "regressions": 12
+          },
+          {
+            "threshold": 0.965379,
+            "coverage": 0.43997044698928706,
+            "combined_accuracy": 0.5860731437015146,
+            "covered_accuracy": 0.9500419815281276,
+            "regressions": 12
+          },
+          {
+            "threshold": 0.96692,
+            "coverage": 0.4299963058736609,
+            "combined_accuracy": 0.5799778352419652,
+            "covered_accuracy": 0.9510309278350515,
+            "regressions": 12
+          },
+          {
+            "threshold": 0.968873,
+            "coverage": 0.4200221647580347,
+            "combined_accuracy": 0.5744366457332841,
+            "covered_accuracy": 0.951627088830255,
+            "regressions": 9
+          },
+          {
+            "threshold": 0.970763,
+            "coverage": 0.41004802364240855,
+            "combined_accuracy": 0.5685260435906908,
+            "covered_accuracy": 0.9518018018018019,
+            "regressions": 9
+          },
+          {
+            "threshold": 0.972469,
+            "coverage": 0.40007388252678244,
+            "combined_accuracy": 0.5626154414480975,
+            "covered_accuracy": 0.9519852262234534,
+            "regressions": 9
+          },
+          {
+            "threshold": 0.974047,
+            "coverage": 0.3900997414111563,
+            "combined_accuracy": 0.5567048393055043,
+            "covered_accuracy": 0.9535984848484849,
+            "regressions": 7
+          },
+          {
+            "threshold": 0.975516,
+            "coverage": 0.37994089397857406,
+            "combined_accuracy": 0.5520871813816033,
+            "covered_accuracy": 0.9533300923675255,
+            "regressions": 7
+          },
+          {
+            "threshold": 0.976913,
+            "coverage": 0.3699667528629479,
+            "combined_accuracy": 0.5469154045068342,
+            "covered_accuracy": 0.9550673989016475,
+            "regressions": 6
+          },
+          {
+            "threshold": 0.978492,
+            "coverage": 0.3599926117473218,
+            "combined_accuracy": 0.5421130402659771,
+            "covered_accuracy": 0.9563878912262699,
+            "regressions": 6
+          },
+          {
+            "threshold": 0.979541,
+            "coverage": 0.3500184706316956,
+            "combined_accuracy": 0.5362024381233839,
+            "covered_accuracy": 0.9551451187335093,
+            "regressions": 6
+          },
+          {
+            "threshold": 0.980824,
+            "coverage": 0.34004432951606944,
+            "combined_accuracy": 0.5308459549316586,
+            "covered_accuracy": 0.955458989679522,
+            "regressions": 6
+          },
+          {
+            "threshold": 0.981792,
+            "coverage": 0.3300701884004433,
+            "combined_accuracy": 0.5247506464721093,
+            "covered_accuracy": 0.9557918298824846,
+            "regressions": 6
+          },
+          {
+            "threshold": 0.98304,
+            "coverage": 0.32009604728481716,
+            "combined_accuracy": 0.5205024011821204,
+            "covered_accuracy": 0.9561454125793422,
+            "regressions": 4
+          },
+          {
+            "threshold": 0.984012,
+            "coverage": 0.30993719985223495,
+            "combined_accuracy": 0.5157000369412634,
+            "covered_accuracy": 0.9576877234803337,
+            "regressions": 4
+          },
+          {
+            "threshold": 0.984861,
+            "coverage": 0.2999630587366088,
+            "combined_accuracy": 0.5099741411156261,
+            "covered_accuracy": 0.9568965517241379,
+            "regressions": 4
+          },
+          {
+            "threshold": 0.985914,
+            "coverage": 0.2899889176209826,
+            "combined_accuracy": 0.5060953084595493,
+            "covered_accuracy": 0.9573248407643312,
+            "regressions": 4
+          },
+          {
+            "threshold": 0.986784,
+            "coverage": 0.2800147765053565,
+            "combined_accuracy": 0.5007388252678242,
+            "covered_accuracy": 0.9597625329815304,
+            "regressions": 4
+          },
+          {
+            "threshold": 0.98768,
+            "coverage": 0.27004063538973033,
+            "combined_accuracy": 0.49556704839305504,
+            "covered_accuracy": 0.9596443228454172,
+            "regressions": 4
+          },
+          {
+            "threshold": 0.988424,
+            "coverage": 0.26006649427410417,
+            "combined_accuracy": 0.4915035094200222,
+            "covered_accuracy": 0.9630681818181818,
+            "regressions": 3
+          },
+          {
+            "threshold": 0.989225,
+            "coverage": 0.250092353158478,
+            "combined_accuracy": 0.48762467676394533,
+            "covered_accuracy": 0.96602658788774,
+            "regressions": 2
+          },
+          {
+            "threshold": 0.989847,
+            "coverage": 0.24011821204285186,
+            "combined_accuracy": 0.4826376062061323,
+            "covered_accuracy": 0.9661538461538461,
+            "regressions": 2
+          },
+          {
+            "threshold": 0.990648,
+            "coverage": 0.22995936461026967,
+            "combined_accuracy": 0.4778352419652752,
+            "covered_accuracy": 0.9654618473895582,
+            "regressions": 2
+          },
+          {
+            "threshold": 0.991326,
+            "coverage": 0.21998522349464353,
+            "combined_accuracy": 0.47340229035833026,
+            "covered_accuracy": 0.964735516372796,
+            "regressions": 2
+          },
+          {
+            "threshold": 0.991868,
+            "coverage": 0.21001108237901736,
+            "combined_accuracy": 0.4684152198005172,
+            "covered_accuracy": 0.9665787159190853,
+            "regressions": 2
+          },
+          {
+            "threshold": 0.992456,
+            "coverage": 0.20003694126339122,
+            "combined_accuracy": 0.46490579977835245,
+            "covered_accuracy": 0.9676823638042474,
+            "regressions": 1
+          },
+          {
+            "threshold": 0.993025,
+            "coverage": 0.19006280014776505,
+            "combined_accuracy": 0.46121167343923164,
+            "covered_accuracy": 0.966958211856171,
+            "regressions": 1
+          },
+          {
+            "threshold": 0.993511,
+            "coverage": 0.1800886590321389,
+            "combined_accuracy": 0.4569634281492427,
+            "covered_accuracy": 0.9661538461538461,
+            "regressions": 1
+          },
+          {
+            "threshold": 0.993984,
+            "coverage": 0.17011451791651275,
+            "combined_accuracy": 0.4528998891762098,
+            "covered_accuracy": 0.9674267100977199,
+            "regressions": 1
+          },
+          {
+            "threshold": 0.99437,
+            "coverage": 0.15995567048393056,
+            "combined_accuracy": 0.44883635020317697,
+            "covered_accuracy": 0.9665127020785219,
+            "regressions": 1
+          },
+          {
+            "threshold": 0.994859,
+            "coverage": 0.15016623568526044,
+            "combined_accuracy": 0.4449575175471001,
+            "covered_accuracy": 0.9692496924969249,
+            "regressions": 1
+          },
+          {
+            "threshold": 0.995294,
+            "coverage": 0.14000738825267825,
+            "combined_accuracy": 0.44200221647580346,
+            "covered_accuracy": 0.9709762532981531,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.995697,
+            "coverage": 0.13021795345400813,
+            "combined_accuracy": 0.43867750277059475,
+            "covered_accuracy": 0.9730496453900709,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.996291,
+            "coverage": 0.12005910602142593,
+            "combined_accuracy": 0.43590690801625415,
+            "covered_accuracy": 0.9738461538461538,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.996686,
+            "coverage": 0.11008496490579978,
+            "combined_accuracy": 0.43202807536017734,
+            "covered_accuracy": 0.9714765100671141,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.99705,
+            "coverage": 0.10011082379017362,
+            "combined_accuracy": 0.4287033616549686,
+            "covered_accuracy": 0.974169741697417,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.997386,
+            "coverage": 0.09013668267454747,
+            "combined_accuracy": 0.425748060583672,
+            "covered_accuracy": 0.9774590163934426,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.997703,
+            "coverage": 0.08016254155892132,
+            "combined_accuracy": 0.4239009974141116,
+            "covered_accuracy": 0.9746543778801844,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.997964,
+            "coverage": 0.07000369412633912,
+            "combined_accuracy": 0.4218692279275951,
+            "covered_accuracy": 0.978891820580475,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.998248,
+            "coverage": 0.060029553010712965,
+            "combined_accuracy": 0.4200221647580347,
+            "covered_accuracy": 0.9907692307692307,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.998595,
+            "coverage": 0.05005541189508681,
+            "combined_accuracy": 0.41688215736978207,
+            "covered_accuracy": 0.988929889298893,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.998848,
+            "coverage": 0.04008127077946066,
+            "combined_accuracy": 0.41466568156630956,
+            "covered_accuracy": 0.9953917050691244,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.99905,
+            "coverage": 0.029922423346878464,
+            "combined_accuracy": 0.4130033247137052,
+            "covered_accuracy": 0.9938271604938271,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.99924,
+            "coverage": 0.020132988548208348,
+            "combined_accuracy": 0.4107868489102327,
+            "covered_accuracy": 0.9908256880733946,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.999454,
+            "coverage": 0.009974141115626155,
+            "combined_accuracy": 0.40801625415589216,
+            "covered_accuracy": 1.0,
+            "regressions": 0
+          },
+          {
+            "threshold": 0.99978,
+            "coverage": 0.0,
+            "combined_accuracy": 0.4059844846693757,
+            "covered_accuracy": null,
+            "regressions": 0
+          }
+        ],
+        "baseline_accuracy": 0.4059844846693757,
+        "no_abstention_accuracy": 0.8134466198743997,
+        "best_combined": {
+          "threshold": 0.287819,
+          "coverage": 0.999815293683044,
+          "combined_accuracy": 0.8134466198743997,
+          "covered_accuracy": 0.8135968963606134,
+          "regressions": 256
+        },
+        "conservative_zero_regression": {
+          "threshold": 0.995294,
+          "coverage": 0.14000738825267825,
+          "combined_accuracy": 0.44200221647580346,
+          "covered_accuracy": 0.9709762532981531,
+          "regressions": 0
+        },
+        "ece": 0.0576062929816676
+      },
+      "nbest_coverage": {
+        "all": {
+          "1": 0.8134466198743997,
+          "2": 0.9436645733284078,
+          "3": 0.9804211304026598,
+          "4": 0.9918729220539343,
+          "5": 0.9970446989287034
+        },
+        "variation_only": {
+          "1": 0.6122448979591837,
+          "2": 0.9387755102040817,
+          "3": 0.9795918367346939,
+          "4": 1.0,
+          "5": 1.0
+        },
+        "n_variation": 49
+      },
+      "operating_point_dev_chosen": {
+        "chosen_on": "dev split, same condition",
+        "max_accuracy": {
+          "threshold": 0.242734,
+          "coverage": 1.0,
+          "combined_accuracy": 0.8134466198743997,
+          "regressions": 256,
+          "improvements": 2462,
+          "n": 5414
+        },
+        "zero_regression_on_dev": {
+          "threshold": 0.997893,
+          "coverage": 0.07258958256372368,
+          "combined_accuracy": 0.42223864056150723,
+          "regressions": 0,
+          "improvements": 88,
+          "n": 5414
+        },
+        "dev_zero_regression_coverage": 0.08007448789571694
+      }
+    }
+  },
+  "notes": {
+    "fairness": "Accuracy is computed on items where OpenJTalk's own reading matched the gold reading, so the comparison is about accent rather than G2P.",
+    "downstep_condition_invariance": "Downstep's prediction does not depend on the condition: it reads components, never the compound's own dictionary entry.",
+    "no_human_annotation": "All gold is dictionary-derived. No native-speaker annotation was collected for this release.",
+    "comparable_set": "Strata are scored on the 5,414 test compounds whose reading OpenJTalk produced correctly in BOTH the listed and the unlisted condition, so the two columns describe the same words."
+  }
+}

export_manifest.json

@@ -0,0 +1,422 @@
+{
+  "tool": "scripts/export.py",
+  "generated_at": "2026-09-03T21:19:59+0900",
+  "git_commit": "HEAD",
+  "environment": {
+    "python": "3.13.5",
+    "platform": "macOS-26.6.2-arm64-arm-64bit-Mach-O",
+    "machine": "arm64",
+    "processor": "arm",
+    "torch": "2.12.0",
+    "onnx": "1.21.0",
+    "onnxruntime": "1.26.0",
+    "numpy": "2.3.3"
+  },
+  "weights": {
+    "provenance": "checkpoint",
+    "path": "checkpoints/base/model.pt",
+    "sha256": "fc20409727cbc84b297d1215b124ce10791e04f8663d025420fe618df849055f",
+    "preset": "base",
+    "n_params": 35054595,
+    "checkpoint_extra": {
+      "epoch": 11,
+      "dev_nucleus_exact": 0.6557333333333333
+    }
+  },
+  "model_config": {
+    "d_model": 512,
+    "n_layers": 10,
+    "n_heads": 8,
+    "d_ff": 2048,
+    "dropout": 0.1,
+    "max_len": 256,
+    "vocab_sizes": {
+      "mora": 128,
+      "pos_fwd": 66,
+      "pos_bwd": 66,
+      "comp_idx": 18,
+      "comp_boundary": 4,
+      "goshu": 8,
+      "comp_accent": 12,
+      "acon_type": 28,
+      "comp_len": 33,
+      "is_final_comp": 4,
+      "comp_first_char": 2945,
+      "comp_last_char": 2799,
+      "in_lexicon": 4
+    },
+    "head_hidden": 256,
+    "preset_name": "base"
+  },
+  "vocab": {
+    "path": "artifacts/export/vocab.json",
+    "note": "loaded from the checkpoint's own checkpoints/base/vocab.json",
+    "sizes": {
+      "mora": 128,
+      "pos_fwd": 66,
+      "pos_bwd": 66,
+      "comp_idx": 18,
+      "comp_boundary": 4,
+      "goshu": 8,
+      "comp_accent": 12,
+      "acon_type": 28,
+      "comp_len": 33,
+      "is_final_comp": 4,
+      "comp_first_char": 2945,
+      "comp_last_char": 2799,
+      "in_lexicon": 4
+    },
+    "bytes": 61786,
+    "mib": 0.059,
+    "sha256": "80a6f40f6679a1efb8e31ffad3626659f5ef071f3e20ee1fa55e3f9c20f34af8"
+  },
+  "onnx_fp32": {
+    "exporter": "legacy",
+    "requested_opset": 17,
+    "opset_in_file": {
+      "ai.onnx": 17
+    },
+    "ir_version": 8,
+    "export_seconds": 0.587,
+    "dynamic_axes": {
+      "batch": 0,
+      "time": 1
+    },
+    "input_names": [
+      "mora",
+      "pos_fwd",
+      "pos_bwd",
+      "comp_idx",
+      "comp_boundary",
+      "goshu",
+      "comp_accent",
+      "acon_type",
+      "comp_len",
+      "is_final_comp",
+      "comp_first_char",
+      "comp_last_char",
+      "in_lexicon",
+      "mask"
+    ],
+    "output_names": [
+      "boundary",
+      "nucleus",
+      "heiban"
+    ],
+    "n_initializers": 143,
+    "n_nodes": 815,
+    "files": [
+      {
+        "path": "artifacts/export/downstep.onnx",
+        "bytes": 140862821,
+        "mib": 134.337,
+        "sha256": "535cb239b7e5ee9b78f254d8d1f4cfa35d773c98dcb78c68494ff324ebdb0c95"
+      }
+    ],
+    "total_bytes": 140862821
+  },
+  "verification_fp32_vs_torch": {
+    "backend": "onnxruntime CPUExecutionProvider vs torch eager, float32",
+    "compared_at": "masked (real) mora positions only",
+    "shapes": [
+      {
+        "batch": 1,
+        "moras": 4,
+        "max_abs_diff": {
+          "boundary": 1.239776611328125e-05,
+          "nucleus": 3.5762786865234375e-06,
+          "heiban": 5.252659320831299e-07
+        }
+      },
+      {
+        "batch": 1,
+        "moras": 8,
+        "max_abs_diff": {
+          "boundary": 9.5367431640625e-06,
+          "nucleus": 5.7220458984375e-06,
+          "heiban": 1.2218952178955078e-06
+        }
+      },
+      {
+        "batch": 1,
+        "moras": 16,
+        "max_abs_diff": {
+          "boundary": 1.1444091796875e-05,
+          "nucleus": 8.58306884765625e-06,
+          "heiban": 2.2649765014648438e-06
+        }
+      },
+      {
+        "batch": 32,
+        "moras": 12,
+        "max_abs_diff": {
+          "boundary": 1.52587890625e-05,
+          "nucleus": 2.1219253540039062e-05,
+          "heiban": 2.86102294921875e-06
+        }
+      }
+    ],
+    "max_abs_diff_overall": 2.1219253540039062e-05,
+    "tolerance": 0.0001,
+    "passed": true
+  },
+  "int8": {
+    "method": "onnxruntime.quantization.quantize_dynamic, QInt8, weight-only",
+    "preprocess": "quant_pre_process(skip_symbolic_shape=True) succeeded",
+    "quantise_seconds": 1.03,
+    "n_nodes": 1000,
+    "file": {
+      "path": "artifacts/export/downstep.int8.onnx",
+      "bytes": 36023179,
+      "mib": 34.354,
+      "sha256": "76dc3085a79052726dad6bb2cd1e1efe03ecd71973b8beac3adbd090b2a31b52"
+    },
+    "fp32_bytes": 140862821,
+    "compression_ratio": 3.91
+  },
+  "verification_int8_vs_fp32": {
+    "shapes": [
+      {
+        "batch": 1,
+        "moras": 4,
+        "max_abs_diff": {
+          "boundary": 0.04032421112060547,
+          "nucleus": 0.008849620819091797,
+          "heiban": 0.002766907215118408
+        }
+      },
+      {
+        "batch": 1,
+        "moras": 8,
+        "max_abs_diff": {
+          "boundary": 0.04124927520751953,
+          "nucleus": 0.04478585720062256,
+          "heiban": 0.007645726203918457
+        }
+      },
+      {
+        "batch": 1,
+        "moras": 16,
+        "max_abs_diff": {
+          "boundary": 0.08224201202392578,
+          "nucleus": 0.12763333320617676,
+          "heiban": 0.034450650215148926
+        }
+      },
+      {
+        "batch": 32,
+        "moras": 12,
+        "max_abs_diff": {
+          "boundary": 0.1212320327758789,
+          "nucleus": 0.09279704093933105,
+          "heiban": 0.04297083616256714
+        }
+      }
+    ],
+    "max_abs_diff_overall": 0.12763333320617676,
+    "tolerance": 0.05,
+    "passed": false,
+    "nucleus_argmax_agreement": {
+      "same": 35,
+      "rows": 35,
+      "rate": 1.0
+    },
+    "boundary_sign_agreement": {
+      "same": 252,
+      "positions": 252,
+      "rate": 1.0
+    }
+  },
+  "latency": {
+    "shapes_note": "random ids; shapes exercise the graph, they are not a corpus sample",
+    "onnxruntime_fp32": {
+      "provider": "CPUExecutionProvider",
+      "intra_op_num_threads": 1,
+      "warmup_iters": 10,
+      "note": "random ids; shapes exercise the graph, they are not a corpus sample",
+      "per_shape": [
+        {
+          "iters": 100,
+          "mean_ms": 3.3892,
+          "median_ms": 3.3489,
+          "p90_ms": 3.5683,
+          "min_ms": 3.2684,
+          "max_ms": 3.7994,
+          "batch": 1,
+          "moras": 4,
+          "ms_per_item_median": 3.3489
+        },
+        {
+          "iters": 100,
+          "mean_ms": 6.2854,
+          "median_ms": 6.2355,
+          "p90_ms": 6.4416,
+          "min_ms": 6.0739,
+          "max_ms": 7.5214,
+          "batch": 1,
+          "moras": 8,
+          "ms_per_item_median": 6.2355
+        },
+        {
+          "iters": 100,
+          "mean_ms": 11.6767,
+          "median_ms": 11.6795,
+          "p90_ms": 11.8093,
+          "min_ms": 11.5046,
+          "max_ms": 11.9721,
+          "batch": 1,
+          "moras": 16,
+          "ms_per_item_median": 11.6795
+        },
+        {
+          "iters": 100,
+          "mean_ms": 260.3052,
+          "median_ms": 260.2948,
+          "p90_ms": 260.9332,
+          "min_ms": 258.0178,
+          "max_ms": 267.6222,
+          "batch": 32,
+          "moras": 12,
+          "ms_per_item_median": 8.1342
+        }
+      ]
+    },
+    "onnxruntime_int8": {
+      "provider": "CPUExecutionProvider",
+      "intra_op_num_threads": 1,
+      "warmup_iters": 10,
+      "note": "random ids; shapes exercise the graph, they are not a corpus sample",
+      "per_shape": [
+        {
+          "iters": 100,
+          "mean_ms": 1.2802,
+          "median_ms": 1.2639,
+          "p90_ms": 1.3232,
+          "min_ms": 1.2197,
+          "max_ms": 1.4885,
+          "batch": 1,
+          "moras": 4,
+          "ms_per_item_median": 1.2639
+        },
+        {
+          "iters": 100,
+          "mean_ms": 2.1173,
+          "median_ms": 2.0918,
+          "p90_ms": 2.2344,
+          "min_ms": 2.0438,
+          "max_ms": 2.4081,
+          "batch": 1,
+          "moras": 8,
+          "ms_per_item_median": 2.0918
+        },
+        {
+          "iters": 100,
+          "mean_ms": 3.9489,
+          "median_ms": 3.9323,
+          "p90_ms": 4.0494,
+          "min_ms": 3.847,
+          "max_ms": 4.1159,
+          "batch": 1,
+          "moras": 16,
+          "ms_per_item_median": 3.9323
+        },
+        {
+          "iters": 100,
+          "mean_ms": 88.3233,
+          "median_ms": 88.2953,
+          "p90_ms": 88.6815,
+          "min_ms": 87.1352,
+          "max_ms": 91.9882,
+          "batch": 32,
+          "moras": 12,
+          "ms_per_item_median": 2.7592
+        }
+      ]
+    },
+    "torch_eager": {
+      "backend": "torch eager CPU",
+      "torch_num_threads": 1,
+      "warmup_iters": 10,
+      "per_shape": [
+        {
+          "iters": 100,
+          "mean_ms": 19.1696,
+          "median_ms": 20.5543,
+          "p90_ms": 23.2695,
+          "min_ms": 12.6302,
+          "max_ms": 25.5071,
+          "batch": 1,
+          "moras": 4,
+          "ms_per_item_median": 20.5543
+        },
+        {
+          "iters": 100,
+          "mean_ms": 18.4129,
+          "median_ms": 18.872,
+          "p90_ms": 26.2903,
+          "min_ms": 10.057,
+          "max_ms": 27.6841,
+          "batch": 1,
+          "moras": 8,
+          "ms_per_item_median": 18.872
+        },
+        {
+          "iters": 100,
+          "mean_ms": 19.1912,
+          "median_ms": 21.4159,
+          "p90_ms": 25.2457,
+          "min_ms": 9.5397,
+          "max_ms": 27.1297,
+          "batch": 1,
+          "moras": 16,
+          "ms_per_item_median": 21.4159
+        },
+        {
+          "iters": 100,
+          "mean_ms": 68.2776,
+          "median_ms": 68.7565,
+          "p90_ms": 71.7727,
+          "min_ms": 58.8458,
+          "max_ms": 73.8272,
+          "batch": 32,
+          "moras": 12,
+          "ms_per_item_median": 2.1486
+        }
+      ]
+    }
+  },
+  "gguf": {
+    "status": "not_requested"
+  },
+  "huggingface_push": {
+    "executed": false,
+    "why": "publishing is a human decision; this script has no upload path",
+    "repo_id": "downstep/downstep-accent",
+    "repo_type": "model",
+    "private": false,
+    "files": [
+      {
+        "path": "artifacts/export/vocab.json",
+        "bytes": 61786,
+        "mib": 0.059,
+        "sha256": "80a6f40f6679a1efb8e31ffad3626659f5ef071f3e20ee1fa55e3f9c20f34af8"
+      },
+      {
+        "path": "artifacts/export/downstep.onnx",
+        "bytes": 140862821,
+        "mib": 134.337,
+        "sha256": "535cb239b7e5ee9b78f254d8d1f4cfa35d773c98dcb78c68494ff324ebdb0c95"
+      },
+      {
+        "path": "artifacts/export/downstep.int8.onnx",
+        "bytes": 36023179,
+        "mib": 34.354,
+        "sha256": "76dc3085a79052726dad6bb2cd1e1efe03ecd71973b8beac3adbd090b2a31b52"
+      }
+    ],
+    "total_bytes": 176947786,
+    "total_mib": 168.751,
+    "equivalent_command": "huggingface-cli upload downstep/downstep-accent <export-dir> . --repo-type=model"
+  },
+  "exit_code": 1
+}

model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc20409727cbc84b297d1215b124ce10791e04f8663d025420fe618df849055f
+size 140258487

training.json

@@ -0,0 +1,122 @@
+{
+  "args": {
+    "bench": "benchmarks",
+    "size": "base",
+    "epochs": 12,
+    "batch_size": 256,
+    "lr": 0.0003,
+    "weight_decay": 0.01,
+    "warmup": 300,
+    "augment": 0.25,
+    "device": "auto",
+    "out": "checkpoints/base",
+    "seed": 0,
+    "max_steps": 0,
+    "dev_limit": 3000
+  },
+  "device": "mps",
+  "git_commit": null,
+  "parameters": 35054595,
+  "best_dev_nucleus_exact": 0.6557333333333333,
+  "wall_clock_seconds": 4925.1,
+  "history": [
+    {
+      "epoch": 0,
+      "step": 350,
+      "loss": 1.3525640233925411,
+      "dev_nucleus_exact": 0.6162666666666666,
+      "dev_boundary_f1": 0.9794019933554816,
+      "seconds": 465.5
+    },
+    {
+      "epoch": 1,
+      "step": 700,
+      "loss": 0.7800240731239318,
+      "dev_nucleus_exact": 0.6346666666666667,
+      "dev_boundary_f1": 0.999333777481679,
+      "seconds": 908.4
+    },
+    {
+      "epoch": 2,
+      "step": 1050,
+      "loss": 0.6399015586716789,
+      "dev_nucleus_exact": 0.6461333333333333,
+      "dev_boundary_f1": 0.9986666666666667,
+      "seconds": 1333.5
+    },
+    {
+      "epoch": 3,
+      "step": 1400,
+      "loss": 0.5727078880582537,
+      "dev_nucleus_exact": 0.652,
+      "dev_boundary_f1": 1.0,
+      "seconds": 1727.4
+    },
+    {
+      "epoch": 4,
+      "step": 1750,
+      "loss": 0.5232000116791043,
+      "dev_nucleus_exact": 0.6490666666666667,
+      "dev_boundary_f1": 1.0,
+      "seconds": 2122.2
+    },
+    {
+      "epoch": 5,
+      "step": 2100,
+      "loss": 0.482598266686712,
+      "dev_nucleus_exact": 0.6514666666666666,
+      "dev_boundary_f1": 1.0,
+      "seconds": 2517.7
+    },
+    {
+      "epoch": 6,
+      "step": 2450,
+      "loss": 0.4508444060598101,
+      "dev_nucleus_exact": 0.6528,
+      "dev_boundary_f1": 1.0,
+      "seconds": 2921.6
+    },
+    {
+      "epoch": 7,
+      "step": 2800,
+      "loss": 0.4187164889914649,
+      "dev_nucleus_exact": 0.6525333333333333,
+      "dev_boundary_f1": 1.0,
+      "seconds": 3324.3
+    },
+    {
+      "epoch": 8,
+      "step": 3150,
+      "loss": 0.39189900176865716,
+      "dev_nucleus_exact": 0.6546666666666666,
+      "dev_boundary_f1": 1.0,
+      "seconds": 3718.0
+    },
+    {
+      "epoch": 9,
+      "step": 3500,
+      "loss": 0.3702780089208058,
+      "dev_nucleus_exact": 0.6538666666666667,
+      "dev_boundary_f1": 1.0,
+      "seconds": 4119.3
+    },
+    {
+      "epoch": 10,
+      "step": 3850,
+      "loss": 0.3531410438673837,
+      "dev_nucleus_exact": 0.6554666666666666,
+      "dev_boundary_f1": 1.0,
+      "seconds": 4519.3
+    },
+    {
+      "epoch": 11,
+      "step": 4200,
+      "loss": 0.34747300050088337,
+      "dev_nucleus_exact": 0.6557333333333333,
+      "dev_boundary_f1": 1.0,
+      "seconds": 4924.5
+    }
+  ],
+  "n_train_examples": 89430,
+  "n_constructed": 17886
+}

vocab.json

@@ -0,0 +1,6137 @@
+{
+ "format": "downstep-vocab-1",
+ "pad": "<pad>",
+ "unk": "<unk>",
+ "fields": [
+  "mora",
+  "pos_fwd",
+  "pos_bwd",
+  "comp_idx",
+  "comp_boundary",
+  "goshu",
+  "comp_accent",
+  "acon_type",
+  "comp_len",
+  "is_final_comp",
+  "comp_first_char",
+  "comp_last_char",
+  "in_lexicon"
+ ],
+ "tables": {
+  "mora": [
+   "ー",
+   "ン",
+   "イ",
+   "シ",
+   "カ",
+   "マ",
+   "ガ",
+   "ク",
+   "ミ",
+   "タ",
+   "キ",
+   "ト",
+   "ワ",
+   "ノ",
+   "コ",
+   "ナ",
+   "ヤ",
+   "サ",
+   "オ",
+   "ラ",
+   "ダ",
+   "リ",
+   "ツ",
+   "チ",
+   "ニ",
+   "ハ",
+   "モ",
+   "セ",
+   "ア",
+   "ジ",
+   "ヒ",
+   "エ",
+   "ケ",
+   "ス",
+   "ッ",
+   "バ",
+   "ロ",
+   "ウ",
+   "ド",
+   "ホ",
+   "ゴ",
+   "フ",
+   "ル",
+   "チョ",
+   "テ",
+   "ヨ",
+   "ジョ",
+   "ズ",
+   "ショ",
+   "ザ",
+   "ヲ",
+   "メ",
+   "ブ",
+   "ソ",
+   "デ",
+   "ゲ",
+   "ム",
+   "ネ",
+   "ギ",
+   "ボ",
+   "レ",
+   "グ",
+   "ビ",
+   "ベ",
+   "キョ",
+   "ユ",
+   "ジュ",
+   "ギョ",
+   "シュ",
+   "チュ",
+   "ゼ",
+   "ゾ",
+   "シャ",
+   "ヘ",
+   "ビョ",
+   "パ",
+   "ヌ",
+   "プ",
+   "ポ",
+   "キュ",
+   "リョ",
+   "リュ",
+   "チャ",
+   "ミョ",
+   "ペ",
+   "ジャ",
+   "ピ",
+   "ニュ",
+   "ティ",
+   "ヒョ",
+   "ディ",
+   "ヒャ",
+   "キャ",
+   "ファ",
+   "フォ",
+   "フィ",
+   "ニャ",
+   "ミャ",
+   "ニョ",
+   "ギュ",
+   "フェ",
+   "ウィ",
+   "チェ",
+   "ウェ",
+   "ヒュ",
+   "ジェ",
+   "トゥ",
+   "ウォ",
+   "ビャ",
+   "リャ",
+   "ギャ",
+   "ヅ",
+   "デュ",
+   "ビュ",
+   "シェ",
+   "ピュ",
+   "ピョ",
+   "ツェ",
+   "ピャ",
+   "ミュ",
+   "ツァ",
+   "ヴィ",
+   "ヂ",
+   "ドゥ",
+   "ヴァ",
+   "ヴォ"
+  ],
+  "pos_fwd": [
+   "0",
+   "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"
+  ],
+  "pos_bwd": [
+   "0",
+   "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"
+  ],
+  "comp_idx": [
+   "0",
+   "1",
+   "2",
+   "3",
+   "4",
+   "5",
+   "6",
+   "7",
+   "8",
+   "9",
+   "10",
+   "11",
+   "12",
+   "13",
+   "14",
+   "15"
+  ],
+  "comp_boundary": [
+   "0",
+   "1"
+  ],
+  "goshu": [
+   "和",
+   "漢",
+   "固",
+   "外",
+   "混",
+   "記号"
+  ],
+  "comp_accent": [
+   "0",
+   "1",
+   "2",
+   "3",
+   "4",
+   "5",
+   "6",
+   "7",
+   "8",
+   "8+"
+  ],
+  "acon_type": [
+   "C2",
+   "C1",
+   "C3",
+   "C4",
+   "名詞%F1",
+   "P2",
+   "動詞%F2@0",
+   "C5",
+   "P1",
+   "動詞F4@0",
+   "動詞%F1",
+   "名詞%F1/動詞%F2@0/形容詞%F2@0",
+   "動詞%F2@0/形容詞%F1",
+   "名詞%F1/動詞%F1/形容詞%F1",
+   "動詞%F4@0",
+   "動詞%F5/名詞%F1",
+   "動詞%F2@1",
+   "名詞%F2@0/動詞%F2@0/形容詞%F2@0",
+   "動詞%F1/形容詞%F1",
+   "動詞%F2@0/形容詞%F1/名詞%F5",
+   "動詞%F2@1/形容詞%F2@1",
+   "動詞%F4@1/形容詞%F4@1",
+   "動詞%F2@0/名詞%F1/形容詞%F1",
+   "動詞%F2@1/形容詞%F2@1/名詞%F2@0",
+   "名詞%F2@1/動詞%F2@1/形容詞%F2@1",
+   "名詞%F1/動詞%F2@0/形容詞%F1"
+  ],
+  "comp_len": [
+   "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"
+  ],
+  "is_final_comp": [
+   "0",
+   "1"
+  ],
+  "comp_first_char": [
+   "大",
+   "東",
+   "西",
+   "山",
+   "新",
+   "中",
+   "上",
+   "南",
+   "北",
+   "町",
+   "川",
+   "下",
+   "高",
+   "小",
+   "三",
+   "が",
+   "の",
+   "は",
+   "と",
+   "日",
+   "か",
+   "一",
+   "工",
+   "し",
+   "を",
+   "本",
+   "い",
+   "八",
+   "学",
+   "長",
+   "短",
+   "平",
+   "松",
+   "神",
+   "二",
+   "ア",
+   "病",
+   "さ",
+   "あ",
+   "島",
+   "金",
+   "シ",
+   "峠",
+   "天",
+   "田",
+   "こ",
+   "石",
+   "水",
+   "信",
+   "ス",
+   "宮",
+   "き",
+   "通",
+   "ノ",
+   "お",
+   "岩",
+   "カ",
+   "台",
+   "た",
+   "四",
+   "五",
+   "線",
+   "赤",
+   "白",
+   "吉",
+   "せ",
+   "御",
+   "電",
+   "前",
+   "十",
+   "千",
+   "太",
+   "内",
+   "福",
+   "く",
+   "野",
+   "コ",
+   "原",
+   "も",
+   "製",
+   "じ",
+   "市",
+   "木",
+   "サ",
+   "マ",
+   "エ",
+   "岳",
+   "六",
+   "ま",
+   "沢",
+   "な",
+   "商",
+   "富",
+   "リ",
+   "フ",
+   "セ",
+   "ち",
+   "森",
+   "ト",
+   "花",
+   "清",
+   "テ",
+   "丸",
+   "国",
+   "ひ",
+   "イ",
+   "安",
+   "オ",
+   "女",
+   "タ",
+   "七",
+   "つ",
+   "み",
+   "横",
+   "バ",
+   "ラ",
+   "谷",
+   "ク",
+   "け",
+   "黒",
+   "だ",
+   "崎",
+   "浜",
+   "青",
+   "古",
+   "河",
+   "そ",
+   "豊",
+   "藤",
+   "ど",
+   "う",
+   "ハ",
+   "池",
+   "や",
+   "え",
+   "橋",
+   "朝",
+   "て",
+   "深",
+   "広",
+   "久",
+   "名",
+   "海",
+   "向",
+   "ば",
+   "鼻",
+   "産",
+   "出",
+   "よ",
+   "伊",
+   "生",
+   "公",
+   "佐",
+   "会",
+   "岡",
+   "銀",
+   "レ",
+   "道",
+   "荒",
+   "美",
+   "九",
+   "立",
+   "今",
+   "人",
+   "キ",
+   "す",
+   "総",
+   "江",
+   "建",
+   "パ",
+   "加",
+   "倉",
+   "若",
+   "井",
+   "和",
+   "口",
+   "ふ",
+   "鳥",
+   "甲",
+   "ウ",
+   "ビ",
+   "ホ",
+   "鹿",
+   "り",
+   "屋",
+   "明",
+   "ジ",
+   "チ",
+   "ガ",
+   "ご",
+   "ロ",
+   "元",
+   "ダ",
+   "京",
+   "春",
+   "浦",
+   "ほ",
+   "ぶ",
+   "竹",
+   "不",
+   "土",
+   "坂",
+   "城",
+   "デ",
+   "桜",
+   "岬",
+   "文",
+   "プ",
+   "地",
+   "相",
+   "住",
+   "戸",
+   "関",
+   "梅",
+   "滝",
+   "寺",
+   "メ",
+   "真",
+   "嵯",
+   "別",
+   "船",
+   "入",
+   "物",
+   "馬",
+   "瀬",
+   "後",
+   "薬",
+   "ソ",
+   "稲",
+   "常",
+   "手",
+   "津",
+   "げ",
+   "ブ",
+   "モ",
+   "緑",
+   "奥",
+   "に",
+   "多",
+   "ぼ",
+   "化",
+   "羽",
+   "郷",
+   "ド",
+   "塩",
+   "阿",
+   "ニ",
+   "草",
+   "堂",
+   "ミ",
+   "玉",
+   "半",
+   "わ",
+   "開",
+   "鉄",
+   "グ",
+   "ゆ",
+   "万",
+   "食",
+   "正",
+   "越",
+   "法",
+   "観",
+   "乙",
+   "有",
+   "飯",
+   "徳",
+   "丘",
+   "め",
+   "ベ",
+   "研",
+   "愛",
+   "宇",
+   "尾",
+   "丁",
+   "保",
+   "ケ",
+   "武",
+   "永",
+   "熊",
+   "協",
+   "旭",
+   "団",
+   "村",
+   "ボ",
+   "合",
+   "興",
+   "柳",
+   "ナ",
+   "泉",
+   "畑",
+   "百",
+   "放",
+   "ね",
+   "鷹",
+   "む",
+   "県",
+   "見",
+   "桃",
+   "矢",
+   "社",
+   "光",
+   "自",
+   "ん",
+   "ポ",
+   "共",
+   "経",
+   "堀",
+   "志",
+   "月",
+   "沖",
+   "ざ",
+   "空",
+   "宝",
+   "湾",
+   "亀",
+   "片",
+   "ツ",
+   "音",
+   "駅",
+   "湯",
+   "無",
+   "里",
+   "浅",
+   "外",
+   "塚",
+   "ネ",
+   "牛",
+   "門",
+   "紫",
+   "米",
+   "労",
+   "機",
+   "林",
+   "沼",
+   "ぐ",
+   "精",
+   "ヤ",
+   "焼",
+   "幌",
+   "鶴",
+   "厚",
+   "温",
+   "成",
+   "栄",
+   "遠",
+   "所",
+   "根",
+   "ゴ",
+   "ワ",
+   "細",
+   "秋",
+   "ら",
+   "唐",
+   "へ",
+   "仁",
+   "目",
+   "れ",
+   "勝",
+   "心",
+   "方",
+   "ぎ",
+   "行",
+   "竜",
+   "ル",
+   "ろ",
+   "流",
+   "ペ",
+   "連",
+   "栗",
+   "ヒ",
+   "家",
+   "萩",
+   "坊",
+   "杉",
+   "教",
+   "ユ",
+   "作",
+   "医",
+   "免",
+   "番",
+   "須",
+   "風",
+   "仙",
+   "っ",
+   "切",
+   "札",
+   "近",
+   "農",
+   "書",
+   "柏",
+   "で",
+   "分",
+   "園",
+   "香",
+   "昭",
+   "板",
+   "火",
+   "駒",
+   "篠",
+   "都",
+   "布",
+   "ピ",
+   "ぜ",
+   "聖",
+   "運",
+   "茶",
+   "ず",
+   "ぞ",
+   "知",
+   "び",
+   "峰",
+   "印",
+   "仲",
+   "引",
+   "発",
+   "院",
+   "時",
+   "龍",
+   "館",
+   "末",
+   "笹",
+   "飛",
+   "子",
+   "早",
+   "之",
+   "づ",
+   "王",
+   "魚",
+   "桂",
+   "油",
+   "鬼",
+   "世",
+   "初",
+   "掛",
+   "親",
+   "交",
+   "足",
+   "軽",
+   "幸",
+   "修",
+   "星",
+   "砂",
+   "第",
+   "間",
+   "笠",
+   "渡",
+   "重",
+   "落",
+   "尻",
+   "曽",
+   "色",
+   "貝",
+   "鳴",
+   "弁",
+   "波",
+   "蔵",
+   "磯",
+   "組",
+   "港",
+   "ぱ",
+   "動",
+   "芦",
+   "室",
+   "樫",
+   "折",
+   "頭",
+   "錦",
+   "取",
+   "技",
+   "付",
+   "源",
+   "烏",
+   "造",
+   "酒",
+   "雲",
+   "楽",
+   "菊",
+   "年",
+   "角",
+   "ゲ",
+   "ム",
+   "猪",
+   "網",
+   "箱",
+   "衣",
+   "丹",
+   "湖",
+   "築",
+   "全",
+   "事",
+   "帝",
+   "牧",
+   "権",
+   "桑",
+   "円",
+   "那",
+   "静",
+   "妙",
+   "雪",
+   "ゼ",
+   "��",
+   "飾",
+   "代",
+   "紙",
+   "販",
+   "べ",
+   "ヨ",
+   "割",
+   "柴",
+   "車",
+   "気",
+   "醍",
+   "駐",
+   "芝",
+   "同",
+   "菅",
+   "調",
+   "伏",
+   "場",
+   "犬",
+   "庄",
+   "蒲",
+   "裏",
+   "又",
+   "形",
+   "登",
+   "筑",
+   "弥",
+   "芸",
+   "証",
+   "郡",
+   "粟",
+   "柿",
+   "局",
+   "祭",
+   "ヘ",
+   "指",
+   "舟",
+   "回",
+   "奈",
+   "蓮",
+   "鈴",
+   "打",
+   "比",
+   "民",
+   "ザ",
+   "味",
+   "直",
+   "体",
+   "彦",
+   "能",
+   "黄",
+   "植",
+   "先",
+   "寿",
+   "恵",
+   "好",
+   "陸",
+   "楠",
+   "寒",
+   "釜",
+   "盆",
+   "次",
+   "袋",
+   "函",
+   "宗",
+   "来",
+   "毛",
+   "玄",
+   "舞",
+   "独",
+   "鍋",
+   "忠",
+   "数",
+   "呉",
+   "宿",
+   "丙",
+   "定",
+   "増",
+   "役",
+   "曲",
+   "男",
+   "最",
+   "葛",
+   "殿",
+   "喜",
+   "雨",
+   "垣",
+   "鋼",
+   "段",
+   "押",
+   "守",
+   "科",
+   "ぬ",
+   "博",
+   "坪",
+   "茂",
+   "養",
+   "勧",
+   "嵐",
+   "左",
+   "猿",
+   "虫",
+   "積",
+   "釣",
+   "当",
+   "紅",
+   "善",
+   "双",
+   "基",
+   "蘇",
+   "的",
+   "表",
+   "鎌",
+   "阪",
+   "雄",
+   "理",
+   "曙",
+   "栃",
+   "境",
+   "店",
+   "琴",
+   "貯",
+   "利",
+   "用",
+   "面",
+   "備",
+   "熱",
+   "者",
+   "弘",
+   "航",
+   "鴨",
+   "条",
+   "ぷ",
+   "専",
+   "巻",
+   "浄",
+   "窪",
+   "部",
+   "身",
+   "鵜",
+   "兵",
+   "刈",
+   "振",
+   "衛",
+   "反",
+   "樒",
+   "麻",
+   "吹",
+   "脇",
+   "夕",
+   "支",
+   "様",
+   "紀",
+   "友",
+   "実",
+   "材",
+   "盛",
+   "姫",
+   "思",
+   "旅",
+   "萱",
+   "鏡",
+   "岐",
+   "管",
+   "諸",
+   "駿",
+   "助",
+   "区",
+   "極",
+   "納",
+   "辻",
+   "ぽ",
+   "特",
+   "荘",
+   "仏",
+   "情",
+   "糸",
+   "在",
+   "扇",
+   "湊",
+   "遊",
+   "楢",
+   "敷",
+   "樋",
+   "力",
+   "性",
+   "洗",
+   "追",
+   "針",
+   "変",
+   "延",
+   "記",
+   "首",
+   "鴻",
+   "計",
+   "桧",
+   "参",
+   "並",
+   "岸",
+   "庁",
+   "陣",
+   "触",
+   "言",
+   "競",
+   "筋",
+   "読",
+   "塔",
+   "瓜",
+   "節",
+   "右",
+   "服",
+   "球",
+   "夜",
+   "将",
+   "府",
+   "消",
+   "瓦",
+   "鞍",
+   "仕",
+   "種",
+   "鷲",
+   "ヴ",
+   "鷺",
+   "橘",
+   "溝",
+   "舘",
+   "晴",
+   "棚",
+   "穂",
+   "幾",
+   "検",
+   "鍛",
+   "乳",
+   "児",
+   "持",
+   "返",
+   "健",
+   "等",
+   "芥",
+   "皇",
+   "綾",
+   "ズ",
+   "仮",
+   "品",
+   "払",
+   "歯",
+   "慶",
+   "症",
+   "辺",
+   "鉱",
+   "防",
+   "る",
+   "礼",
+   "裁",
+   "堤",
+   "張",
+   "意",
+   "梶",
+   "浮",
+   "淡",
+   "灯",
+   "狩",
+   "ゾ",
+   "伝",
+   "榎",
+   "照",
+   "由",
+   "肥",
+   "器",
+   "座",
+   "ン",
+   "帯",
+   "洋",
+   "夫",
+   "政",
+   "楯",
+   "紡",
+   "華",
+   "街",
+   "諏",
+   "豆",
+   "串",
+   "悪",
+   "英",
+   "釧",
+   "呼",
+   "現",
+   "穴",
+   "乱",
+   "差",
+   "戦",
+   "歌",
+   "売",
+   "妻",
+   "順",
+   "ギ",
+   "主",
+   "泊",
+   "酸",
+   "受",
+   "檜",
+   "潟",
+   "織",
+   "財",
+   "貿",
+   "賀",
+   "庭",
+   "洞",
+   "済",
+   "炭",
+   "話",
+   "冷",
+   "対",
+   "拓",
+   "狐",
+   "育",
+   "隠",
+   "鮎",
+   "居",
+   "影",
+   "瑞",
+   "菱",
+   "逢",
+   "渕",
+   "鉢",
+   "夢",
+   "業",
+   "毎",
+   "為",
+   "薩",
+   "陽",
+   "領",
+   "与",
+   "寝",
+   "屯",
+   "群",
+   "胡",
+   "警",
+   "逆",
+   "夏",
+   "弓",
+   "棒",
+   "椿",
+   "淑",
+   "芳",
+   "路",
+   "急",
+   "才",
+   "氷",
+   "突",
+   "藪",
+   "軍",
+   "骨",
+   "余",
+   "党",
+   "類",
+   "式",
+   "泥",
+   "麦",
+   "企",
+   "強",
+   "資",
+   "進",
+   "勘",
+   "臼",
+   "茨",
+   "選",
+   "顔",
+   "巣",
+   "廻",
+   "景",
+   "留",
+   "苦",
+   "試",
+   "障",
+   "少",
+   "投",
+   "磐",
+   "祖",
+   "紋",
+   "薄",
+   "集",
+   "我",
+   "桐",
+   "活",
+   "休",
+   "淀",
+   "菖",
+   "血",
+   "剣",
+   "壱",
+   "更",
+   "格",
+   "使",
+   "劇",
+   "可",
+   "問",
+   "官",
+   "菜",
+   "難",
+   "柱",
+   "牟",
+   "甘",
+   "眼",
+   "粉",
+   "餅",
+   "君",
+   "応",
+   "毘",
+   "狂",
+   "端",
+   "葉",
+   "往",
+   "念",
+   "獅",
+   "霞",
+   "ぴ",
+   "ぺ",
+   "吾",
+   "垂",
+   "壬",
+   "拍",
+   "櫛",
+   "潮",
+   "綜",
+   "義",
+   "腰",
+   "鐵",
+   "雁",
+   "乗",
+   "包",
+   "士",
+   "待",
+   "般",
+   "荷",
+   "辰",
+   "創",
+   "師",
+   "湘",
+   "牡",
+   "虚",
+   "逓",
+   "釈",
+   "革",
+   "苗",
+   "満",
+   "漆",
+   "亜",
+   "制",
+   "帽",
+   "昌",
+   "未",
+   "羅",
+   "腹",
+   "走",
+   "俣",
+   "声",
+   "尼",
+   "屈",
+   "感",
+   "灘",
+   "猫",
+   "琵",
+   "環",
+   "背",
+   "苅",
+   "陶",
+   "ヌ",
+   "嘉",
+   "報",
+   "滑",
+   "股",
+   "茅",
+   "薮",
+   "貴",
+   "冠",
+   "単",
+   "壇",
+   "如",
+   "繰",
+   "藻",
+   "講",
+   "列",
+   "抜",
+   "揚",
+   "染",
+   "母",
+   "要",
+   "語",
+   "介",
+   "替",
+   "朱",
+   "構",
+   "洪",
+   "私",
+   "耳",
+   "達",
+   "漢",
+   "蜂",
+   "蟹",
+   "覚",
+   "貫",
+   "霧",
+   "兼",
+   "図",
+   "希",
+   "摂",
+   "旧",
+   "株",
+   "槻",
+   "模",
+   "泰",
+   "皆",
+   "祝",
+   "紺",
+   "綱",
+   "蒸",
+   "蛇",
+   "塗",
+   "峯",
+   "弾",
+   "暗",
+   "結",
+   "老",
+   "荻",
+   "輪",
+   "遅",
+   "奉",
+   "始",
+   "字",
+   "嶋",
+   "巾",
+   "恋",
+   "滋",
+   "脳",
+   "辛",
+   "雑",
+   "庚",
+   "洲",
+   "渚",
+   "省",
+   "笛",
+   "縄",
+   "迎",
+   "鳩",
+   "恩",
+   "掘",
+   "梨",
+   "続",
+   "軒",
+   "非",
+   "願",
+   "兜",
+   "弟",
+   "歩",
+   "絵",
+   "虎",
+   "賞",
+   "起",
+   "周",
+   "埼",
+   "寄",
+   "房",
+   "欠",
+   "止",
+   "灰",
+   "狭",
+   "甚",
+   "着",
+   "砥",
+   "硝",
+   "秩",
+   "臨",
+   "談",
+   "論",
+   "跡",
+   "過",
+   "郵",
+   "予",
+   "卸",
+   "帆",
+   "干",
+   "撫",
+   "祇",
+   "篭",
+   "虹",
+   "髪",
+   "借",
+   "判",
+   "剤",
+   "枚",
+   "案",
+   "状",
+   "糠",
+   "肉",
+   "請",
+   "雷",
+   "令",
+   "優",
+   "写",
+   "墨",
+   "測",
+   "燈",
+   "筒",
+   "芽",
+   "苫",
+   "転",
+   "遣",
+   "乾",
+   "孝",
+   "巡",
+   "底",
+   "揖",
+   "敬",
+   "智",
+   "杭",
+   "湧",
+   "綿",
+   "装",
+   "買",
+   "踊",
+   "送",
+   "鉾",
+   "営",
+   "慈",
+   "樹",
+   "決",
+   "看",
+   "筆",
+   "練",
+   "芋",
+   "蓬",
+   "銭",
+   "幕",
+   "改",
+   "散",
+   "斎",
+   "画",
+   "童",
+   "茜",
+   "詰",
+   "違",
+   "鎮",
+   "治",
+   "絶",
+   "菩",
+   "袴",
+   "迫",
+   "頓",
+   "兄",
+   "勇",
+   "奇",
+   "姉",
+   "懸",
+   "昆",
+   "標",
+   "煙",
+   "鍾",
+   "韓",
+   "頼",
+   "丈",
+   "俵",
+   "卵",
+   "床",
+   "康",
+   "廃",
+   "得",
+   "拝",
+   "柄",
+   "盲",
+   "繊",
+   "繋",
+   "郭",
+   "以",
+   "刑",
+   "収",
+   "庵",
+   "浪",
+   "渋",
+   "溜",
+   "竪",
+   "蓋",
+   "輸",
+   "倶",
+   "壷",
+   "掃",
+   "整",
+   "椎",
+   "死",
+   "破",
+   "笑",
+   "素",
+   "蜻",
+   "鍵",
+   "ヅ",
+   "供",
+   "冬",
+   "屏",
+   "巽",
+   "招",
+   "汽",
+   "編",
+   "被",
+   "設",
+   "評",
+   "貧",
+   "銚",
+   "鳶",
+   "鼎",
+   "兎",
+   "堅",
+   "崩",
+   "彌",
+   "斉",
+   "料",
+   "枯",
+   "炎",
+   "稗",
+   "良",
+   "伸",
+   "再",
+   "惣",
+   "斜",
+   "核",
+   "桶",
+   "沓",
+   "点",
+   "粒",
+   "終",
+   "舎",
+   "菓",
+   "蓄",
+   "蕨",
+   "蛍",
+   "貞",
+   "週",
+   "鋳",
+   "離",
+   "堺",
+   "摺",
+   "播",
+   "曾",
+   "望",
+   "盤",
+   "負",
+   "醤",
+   "隅",
+   "厳",
+   "因",
+   "幹",
+   "復",
+   "昇",
+   "昼",
+   "普",
+   "校",
+   "梵",
+   "椥",
+   "氏",
+   "泣",
+   "淵",
+   "箕",
+   "複",
+   "附",
+   "伯",
+   "傍",
+   "祓",
+   "簾",
+   "縮",
+   "習",
+   "解",
+   "霊",
+   "位",
+   "嶺",
+   "巨",
+   "忍",
+   "柔",
+   "燕",
+   "硫",
+   "禁",
+   "秀",
+   "紛",
+   "統",
+   "縫",
+   "職",
+   "袖",
+   "許",
+   "責",
+   "陰",
+   "値",
+   "処",
+   "岱",
+   "怪",
+   "旗",
+   "枕",
+   "榊",
+   "沙",
+   "純",
+   "耕",
+   "芹",
+   "葦",
+   "護",
+   "透",
+   "露",
+   "額",
+   "僧",
+   "壁",
+   "客",
+   "嶽",
+   "恒",
+   "燃",
+   "珍",
+   "畜",
+   "苑",
+   "蘭",
+   "閉",
+   "命",
+   "孫",
+   "排",
+   "澄",
+   "畳",
+   "税",
+   "糖",
+   "胸",
+   "鐘",
+   "鯖",
+   "咲",
+   "型",
+   "密",
+   "承",
+   "杓",
+   "棟",
+   "議",
+   "質",
+   "込",
+   "靴",
+   "韮",
+   "冨",
+   "救",
+   "映",
+   "昨",
+   "枝",
+   "桟",
+   "槙",
+   "父",
+   "瓢",
+   "皮",
+   "羊",
+   "腺",
+   "葭",
+   "速",
+   "麗",
+   "倭",
+   "傘",
+   "則",
+   "嫁",
+   "帰",
+   "度",
+   "微",
+   "抵",
+   "提",
+   "損",
+   "攻",
+   "暑",
+   "期",
+   "杖",
+   "桔",
+   "椙",
+   "汁",
+   "浴",
+   "渓",
+   "碧",
+   "礁",
+   "置",
+   "貸",
+   "配",
+   "銅",
+   "亭",
+   "厨",
+   "失",
+   "巴",
+   "斑",
+   "暮",
+   "濁",
+   "爪",
+   "畠",
+   "縁",
+   "誠",
+   "銘",
+   "飲",
+   "鳳",
+   "勢",
+   "召",
+   "司",
+   "尺",
+   "抱",
+   "拾",
+   "梁",
+   "楊",
+   "樟",
+   "缶",
+   "聞",
+   "肩",
+   "蛭",
+   "霜",
+   "鞆",
+   "頂",
+   "鶏",
+   "塞",
+   "廿",
+   "戻",
+   "檀",
+   "混",
+   "申",
+   "移",
+   "絞",
+   "考",
+   "菰",
+   "萌",
+   "誉",
+   "讃",
+   "迷",
+   "逃",
+   "丑",
+   "偏",
+   "堰",
+   "婦",
+   "宣",
+   "慰",
+   "抗",
+   "旦",
+   "���",
+   "汐",
+   "汗",
+   "注",
+   "版",
+   "犯",
+   "狼",
+   "琉",
+   "異",
+   "療",
+   "継",
+   "蔦",
+   "蚊",
+   "蝙",
+   "釘",
+   "鍬",
+   "階",
+   "際",
+   "頃",
+   "騒",
+   "他",
+   "個",
+   "尋",
+   "添",
+   "献",
+   "答",
+   "繁",
+   "蒔",
+   "藁",
+   "預",
+   "ッ",
+   "乃",
+   "侍",
+   "俊",
+   "占",
+   "卯",
+   "叩",
+   "吐",
+   "圧",
+   "婆",
+   "寛",
+   "審",
+   "層",
+   "彼",
+   "律",
+   "忙",
+   "懐",
+   "扱",
+   "捻",
+   "敦",
+   "敵",
+   "斬",
+   "栂",
+   "殺",
+   "皐",
+   "稚",
+   "聚",
+   "膳",
+   "蛸",
+   "躑",
+   "雀",
+   "鼠",
+   "具",
+   "卒",
+   "季",
+   "戒",
+   "故",
+   "痛",
+   "祥",
+   "算",
+   "給",
+   "維",
+   "艶",
+   "誘",
+   "辞",
+   "鉛",
+   "仰",
+   "偕",
+   "刻",
+   "即",
+   "存",
+   "帳",
+   "杏",
+   "柚",
+   "橿",
+   "漁",
+   "瀧",
+   "舌",
+   "菌",
+   "蕎",
+   "術",
+   "退",
+   "適",
+   "隊",
+   "鴬",
+   "鵠",
+   "鶯",
+   "鼈",
+   "伴",
+   "倒",
+   "刀",
+   "娘",
+   "宅",
+   "揃",
+   "束",
+   "樽",
+   "潜",
+   "煮",
+   "熟",
+   "眠",
+   "籠",
+   "締",
+   "胃",
+   "臥",
+   "荏",
+   "裾",
+   "閑",
+   "鯨",
+   "側",
+   "号",
+   "嵌",
+   "巧",
+   "拳",
+   "探",
+   "摩",
+   "撮",
+   "断",
+   "易",
+   "此",
+   "涌",
+   "激",
+   "硬",
+   "章",
+   "肝",
+   "腕",
+   "途",
+   "酔",
+   "魯",
+   "似",
+   "伽",
+   "便",
+   "危",
+   "宏",
+   "淳",
+   "然",
+   "煎",
+   "珊",
+   "禅",
+   "粗",
+   "量",
+   "震",
+   "亡",
+   "但",
+   "低",
+   "俄",
+   "停",
+   "冶",
+   "嫌",
+   "寸",
+   "展",
+   "崇",
+   "己",
+   "接",
+   "是",
+   "暁",
+   "暦",
+   "杜",
+   "柑",
+   "桝",
+   "液",
+   "減",
+   "澤",
+   "澱",
+   "炙",
+   "瓶",
+   "砕",
+   "稽",
+   "約",
+   "翁",
+   "脂",
+   "脚",
+   "葵",
+   "蟻",
+   "誕",
+   "證",
+   "賢",
+   "蹴",
+   "軌",
+   "轟",
+   "邸",
+   "飴",
+   "鱒",
+   "冥",
+   "刺",
+   "匂",
+   "幽",
+   "戎",
+   "推",
+   "盗",
+   "耐",
+   "茗",
+   "詞",
+   "豪",
+   "軟",
+   "魔",
+   "争",
+   "何",
+   "債",
+   "吟",
+   "呂",
+   "委",
+   "姥",
+   "嬉",
+   "幻",
+   "廣",
+   "捨",
+   "暴",
+   "杯",
+   "榛",
+   "欧",
+   "殻",
+   "求",
+   "演",
+   "漬",
+   "煥",
+   "瑠",
+   "痴",
+   "皿",
+   "竿",
+   "箸",
+   "翠",
+   "肺",
+   "膝",
+   "芙",
+   "蝿",
+   "規",
+   "討",
+   "詩",
+   "酉",
+   "項",
+   "餡",
+   "劣",
+   "噺",
+   "囃",
+   "封",
+   "採",
+   "曇",
+   "歴",
+   "沈",
+   "浚",
+   "湿",
+   "準",
+   "畝",
+   "睦",
+   "瞳",
+   "碁",
+   "螟",
+   "補",
+   "裸",
+   "訓",
+   "謝",
+   "贔",
+   "超",
+   "依",
+   "働",
+   "勤",
+   "呑",
+   "員",
+   "啓",
+   "執",
+   "墓",
+   "循",
+   "必",
+   "悠",
+   "愁",
+   "憲",
+   "戊",
+   "扶",
+   "晶",
+   "没",
+   "渦",
+   "爺",
+   "癖",
+   "監",
+   "称",
+   "箇",
+   "粕",
+   "糀",
+   "絹",
+   "罪",
+   "脱",
+   "艦",
+   "茹",
+   "蒟",
+   "蓑",
+   "蓼",
+   "藍",
+   "蜘",
+   "蝦",
+   "説",
+   "讀",
+   "辨",
+   "郎",
+   "鍔",
+   "闇",
+   "降",
+   "陵",
+   "雍",
+   "霹",
+   "饅",
+   "駕",
+   "鬢",
+   "鮨",
+   "鯛",
+   "鰻",
+   "鱈",
+   "ぢ",
+   "奄",
+   "州",
+   "幡",
+   "従",
+   "恐",
+   "畔",
+   "益",
+   "確",
+   "粥",
+   "緞",
+   "鎧",
+   "鮫",
+   "鯉",
+   "凌",
+   "削",
+   "升",
+   "只",
+   "固",
+   "圏",
+   "宍",
+   "容",
+   "尊",
+   "尚",
+   "席",
+   "幅",
+   "径",
+   "披",
+   "挿",
+   "敏",
+   "斗",
+   "晒",
+   "晩",
+   "桁",
+   "玩",
+   "痩",
+   "砲",
+   "磁",
+   "礒",
+   "窯",
+   "系",
+   "緒",
+   "縞",
+   "萬",
+   "象",
+   "轡",
+   "逸",
+   "錫",
+   "録",
+   "鍜",
+   "限",
+   "隈",
+   "雌",
+   "題",
+   "顕",
+   "鱗",
+   "鴛",
+   "鶉",
+   "傷",
+   "垢",
+   "幼",
+   "序",
+   "征",
+   "揺",
+   "撞",
+   "擬",
+   "曹",
+   "残",
+   "浸",
+   "獣",
+   "腎",
+   "茄",
+   "頸",
+   "騎",
+   "伺",
+   "佑",
+   "催",
+   "充",
+   "凝",
+   "凸",
+   "到",
+   "叡",
+   "各",
+   "喧",
+   "喪",
+   "塀",
+   "塙",
+   "妹",
+   "學",
+   "寧",
+   "尖",
+   "就",
+   "巌",
+   "弔",
+   "徒",
+   "徹",
+   "悟",
+   "抑",
+   "掖",
+   "會",
+   "朴",
+   "栽",
+   "楼",
+   "槍",
+   "櫃",
+   "櫓",
+   "欽",
+   "沿",
+   "潅",
+   "猩",
+   "疾",
+   "痙",
+   "眉",
+   "瞬",
+   "竃",
+   "糞",
+   "綺",
+   "胎",
+   "茎",
+   "葬",
+   "蚕",
+   "蛯",
+   "蝶",
+   "蟷",
+   "蠣",
+   "衝",
+   "裕",
+   "襟",
+   "託",
+   "誤",
+   "醸",
+   "陳",
+   "麓",
+   "黎",
+   "互",
+   "儲",
+   "凡",
+   "剥",
+   "壮",
+   "寂",
+   "尽",
+   "届",
+   "履",
+   "庫",
+   "廊",
+   "忘",
+   "惜",
+   "愚",
+   "憎",
+   "捧",
+   "既",
+   "柘",
+   "槇",
+   "歳",
+   "毀",
+   "泳",
+   "渥",
+   "炒",
+   "狸",
+   "痘",
+   "盧",
+   "祈",
+   "秤",
+   "稼",
+   "綴",
+   "致",
+   "酢",
+   "鋲",
+   "鑵",
+   "隔",
+   "飼",
+   "佼",
+   "係",
+   "俗",
+   "吸",
+   "噴",
+   "圓",
+   "均",
+   "完",
+   "屁",
+   "彫",
+   "懲",
+   "拭",
+   "挙",
+   "擦",
+   "朗",
+   "果",
+   "枢",
+   "椅",
+   "椋",
+   "樺",
+   "滅",
+   "濡",
+   "燐",
+   "爛",
+   "牢",
+   "率",
+   "玖",
+   "珠",
+   "班",
+   "祢",
+   "禿",
+   "窓",
+   "竈",
+   "紐",
+   "翌",
+   "聴",
+   "薦",
+   "蝋",
+   "螺",
+   "袈",
+   "認",
+   "遺",
+   "鉤",
+   "銃",
+   "錬",
+   "雙",
+   "雛",
+   "餓",
+   "髷",
+   "魂",
+   "鰹",
+   "鴫",
+   "副",
+   "午",
+   "吊",
+   "唯",
+   "國",
+   "夷",
+   "射",
+   "捕",
+   "搦",
+   "涙",
+   "界",
+   "疋",
+   "礎",
+   "葎",
+   "蒜",
+   "詫",
+   "鯵",
+   "ゑ",
+   "任",
+   "例",
+   "其",
+   "凄",
+   "勅",
+   "勉",
+   "厄",
+   "含",
+   "哲",
+   "喰",
+   "契",
+   "奴",
+   "妖",
+   "帖",
+   "帷",
+   "庖",
+   "庶",
+   "扼",
+   "挟",
+   "揉",
+   "擲",
+   "朽",
+   "李",
+   "棲",
+   "樅",
+   "橡",
+   "櫻",
+   "沫",
+   "溪",
+   "溶",
+   "漏",
+   "潰",
+   "濱",
+   "煉",
+   "熾",
+   "獨",
+   "疱",
+   "程",
+   "窶",
+   "筍",
+   "粂",
+   "粘",
+   "索",
+   "紬",
+   "署",
+   "芭",
+   "苔",
+   "苛",
+   "莇",
+   "菎",
+   "葡",
+   "蕗",
+   "薪",
+   "藺",
+   "蛋",
+   "袁",
+   "褄",
+   "覗",
+   "訴",
+   "貰",
+   "賛",
+   "迢",
+   "遍",
+   "遼",
+   "醗",
+   "錣",
+   "鍼",
+   "頌",
+   "饂",
+   "鬱",
+   "鵯",
+   "麒",
+   "麺",
+   "齊",
+   "龜",
+   "厩",
+   "悲",
+   "慣",
+   "戯",
+   "扁",
+   "択",
+   "歓",
+   "渉",
+   "濃",
+   "秘",
+   "笊",
+   "簡",
+   "蜜",
+   "誇",
+   "踏",
+   "陀",
+   "隆",
+   "飽",
+   "乞",
+   "倍",
+   "像",
+   "儀",
+   "典",
+   "凧",
+   "刷",
+   "券",
+   "剛",
+   "哀",
+   "唄",
+   "喝",
+   "圃",
+   "埋",
+   "姿",
+   "孔",
+   "寮",
+   "悔",
+   "惚",
+   "撥",
+   "架",
+   "樗",
+   "欲",
+   "派",
+   "淋",
+   "潤",
+   "炉",
+   "煩",
+   "皺",
+   "籤",
+   "糊",
+   "脊",
+   "腸",
+   "艷",
+   "蒼",
+   "蕊",
+   "薔",
+   "薗",
+   "蟇",
+   "訳",
+   "譲",
+   "賎",
+   "距",
+   "跳",
+   "躁",
+   "酪",
+   "錠",
+   "雫",
+   "顎",
+   "馳",
+   "髭",
+   "鮮",
+   "鯲",
+   "鰺",
+   "麟",
+   "佩",
+   "俳",
+   "偽",
+   "匹",
+   "呵",
+   "咄",
+   "坩",
+   "埠",
+   "塘",
+   "姶",
+   "恨",
+   "括",
+   "控",
+   "換",
+   "擂",
+   "操",
+   "於",
+   "曳",
+   "机",
+   "栖",
+   "棋",
+   "棧",
+   "椚",
+   "殴",
+   "毒",
+   "氣",
+   "汚",
+   "汲",
+   "涎",
+   "癒",
+   "睨",
+   "硯",
+   "碓",
+   "筏",
+   "箒",
+   "箪",
+   "篝",
+   "簀",
+   "綬",
+   "胴",
+   "臭",
+   "菟",
+   "著",
+   "蛙",
+   "誣",
+   "豕",
+   "贈",
+   "轆",
+   "邪",
+   "鋏",
+   "鏑",
+   "閻",
+   "隼",
+   "雇",
+   "颪",
+   "鰯",
+   "麝",
+   "了",
+   "亢",
+   "享",
+   "侃",
+   "倦",
+   "僕",
+   "億",
+   "兢",
+   "冗",
+   "凍",
+   "凹",
+   "剃",
+   "劉",
+   "勲",
+   "叔",
+   "叫",
+   "叶",
+   "咸",
+   "唱",
+   "嗅",
+   "嘲",
+   "囚",
+   "坐",
+   "坦",
+   "塒",
+   "塵",
+   "妄",
+   "婿",
+   "媒",
+   "嬢",
+   "害",
+   "導",
+   "尿",
+   "峡",
+   "巷",
+   "弦",
+   "弱",
+   "彩",
+   "忌",
+   "快",
+   "怒",
+   "恰",
+   "慷",
+   "憂",
+   "憚",
+   "戴",
+   "托",
+   "拡",
+   "拵",
+   "撰",
+   "敗",
+   "杣",
+   "枠",
+   "枡",
+   "椹",
+   "楔",
+   "楡",
+   "概",
+   "檮",
+   "殖",
+   "洟",
+   "渾",
+   "爽",
+   "牽",
+   "犀",
+   "猛",
+   "猟",
+   "疣",
+   "疫",
+   "睡",
+   "祐",
+   "禽",
+   "秦",
+   "穏",
+   "篁",
+   "簗",
+   "級",
+   "累",
+   "紹",
+   "絃",
+   "緋",
+   "翰",
+   "耶",
+   "聟",
+   "聨",
+   "肌",
+   "肖",
+   "脹",
+   "腔",
+   "腫",
+   "臣",
+   "茱",
+   "葺",
+   "蒋",
+   "蔀",
+   "薫",
+   "薯",
+   "虻",
+   "蝉",
+   "蝸",
+   "蟋",
+   "蟠",
+   "袞",
+   "褌",
+   "褒",
+   "視",
+   "訝",
+   "誅",
+   "諤",
+   "謹",
+   "識",
+   "趣",
+   "軸",
+   "邑",
+   "錆",
+   "錢",
+   "錺",
+   "鐙",
+   "鐺",
+   "鑑",
+   "闊",
+   "除",
+   "雅",
+   "頚",
+   "驚",
+   "黍",
+   "仇",
+   "佛",
+   "捌",
+   "棄",
+   "漫",
+   "澁",
+   "範",
+   "簿",
+   "肘",
+   "苧",
+   "蜃",
+   "融",
+   "訪",
+   "躬",
+   "錯",
+   "鞘",
+   "ヲ",
+   "丼",
+   "也",
+   "亂",
+   "亦",
+   "佳",
+   "併",
+   "侘",
+   "候",
+   "僻",
+   "准",
+   "凜",
+   "刃",
+   "刊",
+   "勿",
+   "卓",
+   "卿",
+   "及",
+   "咽",
+   "哺",
+   "喋",
+   "嘆",
+   "堆",
+   "填",
+   "塾",
+   "壺",
+   "夙",
+   "奨",
+   "奪",
+   "姪",
+   "威",
+   "屠",
+   "崑",
+   "彰",
+   "息",
+   "懇",
+   "戌",
+   "抒",
+   "担",
+   "挾",
+   "据",
+   "掴",
+   "掻",
+   "握",
+   "搗",
+   "斐",
+   "施",
+   "柵",
+   "梯",
+   "棘",
+   "榾",
+   "槽",
+   "欅",
+   "殊",
+   "汀",
+   "洒",
+   "洛",
+   "浩",
+   "涸",
+   "淮",
+   "淺",
+   "滓",
+   "澗",
+   "炊",
+   "烟",
+   "燒",
+   "爆",
+   "狢",
+   "猜",
+   "獺",
+   "玲",
+   "珪",
+   "瑛",
+   "瑪",
+   "略",
+   "疎",
+   "疲",
+   "矛",
+   "矧",
+   "砒",
+   "砺",
+   "碇",
+   "碍",
+   "碑",
+   "磧",
+   "礪",
+   "祟",
+   "票",
+   "祺",
+   "禊",
+   "稀",
+   "稻",
+   "穿",
+   "窃",
+   "策",
+   "筵",
+   "箋",
+   "籐",
+   "糧",
+   "絣",
+   "緩",
+   "縦",
+   "纏",
+   "翼",
+   "腓",
+   "腱",
+   "臍",
+   "舗",
+   "舜",
+   "艇",
+   "茣",
+   "茸",
+   "莫",
+   "菫",
+   "葱",
+   "蘆",
+   "蛻",
+   "褐",
+   "覇",
+   "謂",
+   "謄",
+   "貢",
+   "賊",
+   "遁",
+   "避",
+   "鈍",
+   "鉦",
+   "鋪",
+   "錨",
+   "鐡",
+   "閼",
+   "陛",
+   "隙",
+   "雉",
+   "靨",
+   "韻",
+   "頬",
+   "饗",
+   "鬘",
+   "鰊",
+   "鰐",
+   "鴇",
+   "麿",
+   "鼓",
+   "鼾",
+   "ヱ",
+   "句",
+   "史",
+   "喬",
+   "捲",
+   "授",
+   "揮",
+   "曰",
+   "毅",
+   "涼",
+   "煤",
+   "詠",
+   "輿",
+   "駄",
+   "鮪",
+   "鰍",
+   "ヂ",
+   "仄",
+   "侭",
+   "侯",
+   "侵",
+   "兀",
+   "允",
+   "凱",
+   "剝",
+   "剰",
+   "劍",
+   "功",
+   "匙",
+   "厘",
+   "叙",
+   "吃",
+   "吻",
+   "咬",
+   "嗣",
+   "嘗",
+   "噛",
+   "囂",
+   "圭",
+   "埓",
+   "培",
+   "墳",
+   "夾",
+   "奏",
+   "奮",
+   "姐",
+   "姚",
+   "姨",
+   "娑",
+   "孟",
+   "孤",
+   "宋",
+   "宴",
+   "宵",
+   "寅",
+   "察",
+   "尸",
+   "峨",
+   "峻",
+   "庸",
+   "彗",
+   "彷",
+   "恥",
+   "悉",
+   "慇",
+   "慧",
+   "抛",
+   "抹",
+   "抽",
+   "搬",
+   "撃",
+   "斧",
+   "旁",
+   "族",
+   "旺",
+   "曖",
+   "朔",
+   "枇",
+   "柊",
+   "栓",
+   "桓",
+   "梗",
+   "梛",
+   "條",
+   "棺",
+   "椰",
+   "椴",
+   "榧",
+   "檳",
+   "櫂",
+   "欄",
+   "欣",
+   "殉",
+   "汎",
+   "沸",
+   "泡",
+   "淞",
+   "溢",
+   "漸",
+   "潔",
+   "濤",
+   "瀋",
+   "瀑",
+   "瀝",
+   "熔",
+   "狒",
+   "猯",
+   "猶",
+   "琅",
+   "琺",
+   "甜",
+   "甥",
+   "疥",
+   "癇",
+   "眺",
+   "瞭",
+   "磊",
+   "磨",
+   "示",
+   "秒",
+   "穀",
+   "笙",
+   "筌",
+   "箔",
+   "籾",
+   "粛",
+   "絡",
+   "絲",
+   "綽",
+   "縒",
+   "縟",
+   "縹",
+   "罠",
+   "罨",
+   "耀",
+   "耗",
+   "聯",
+   "脈",
+   "膏",
+   "膨",
+   "膿",
+   "臆",
+   "臈",
+   "舐",
+   "艱",
+   "董",
+   "蓉",
+   "蓴",
+   "蔚",
+   "蕪",
+   "薇",
+   "薙",
+   "藩",
+   "藷",
+   "藹",
+   "蚤",
+   "蜀",
+   "蜉",
+   "蜑",
+   "衆",
+   "裂",
+   "覿",
+   "觜",
+   "誓",
+   "諌",
+   "諜",
+   "謙",
+   "貼",
+   "賓",
+   "賽",
+   "贅",
+   "跼",
+   "蹄",
+   "這",
+   "遥",
+   "邦",
+   "邯",
+   "郁",
+   "鄒",
+   "鄧",
+   "酌",
+   "醜",
+   "鈎",
+   "鈑",
+   "鉈",
+   "鋤",
+   "鐚",
+   "鑓",
+   "闘",
+   "頁",
+   "頑",
+   "馮",
+   "駆",
+   "駈",
+   "験",
+   "魁",
+   "魍",
+   "鮭",
+   "鹹",
+   "鹽",
+   "齋",
+   "龕",
+   "ゐ",
+   "伐",
+   "佃",
+   "倫",
+   "假",
+   "冊",
+   "剪",
+   "劃",
+   "励",
+   "叢",
+   "否",
+   "呪",
+   "唇",
+   "唾",
+   "喇",
+   "喫",
+   "囲",
+   "坑",
+   "堕",
+   "壊",
+   "嬲",
+   "嬶",
+   "孀",
+   "孚",
+   "寡",
+   "廉",
+   "忽",
+   "怨",
+   "恕",
+   "憩",
+   "把",
+   "捉",
+   "捏",
+   "措",
+   "掬",
+   "掲",
+   "攝",
+   "敝",
+   "旋",
+   "晦",
+   "暇",
+   "曝",
+   "朧",
+   "枸",
+   "栢",
+   "棗",
+   "椀",
+   "楓",
+   "楕",
+   "櫟",
+   "歪",
+   "汜",
+   "汝",
+   "沃",
+   "沐",
+   "涅",
+   "濯",
+   "炬",
+   "焚",
+   "焜",
+   "燧",
+   "爬",
+   "爾",
+   "琢",
+   "疏",
+   "疑",
+   "痺",
+   "癡",
+   "皓",
+   "瞼",
+   "砧",
+   "礫",
+   "秣",
+   "稔",
+   "穢",
+   "糺",
+   "紳",
+   "絆",
+   "絽",
+   "肴",
+   "臘",
+   "至",
+   "舮",
+   "舳",
+   "艀",
+   "茘",
+   "萠",
+   "蒙",
+   "薊",
+   "衞",
+   "袷",
+   "覆",
+   "課",
+   "謳",
+   "譜",
+   "讚",
+   "貨",
+   "貽",
+   "跨",
+   "蹐",
+   "較",
+   "輻",
+   "迦",
+   "逐",
+   "釦",
+   "鍍",
+   "鏤",
+   "隣",
+   "零",
+   "餌",
+   "騙",
+   "鮑",
+   "鷁",
+   "麹",
+   "黙",
+   "ァ",
+   "ヰ",
+   "云",
+   "伍",
+   "佝",
+   "俑",
+   "倹",
+   "儘",
+   "凛",
+   "効",
+   "卜",
+   "厭",
+   "吏",
+   "呈",
+   "告",
+   "啄",
+   "喉",
+   "嘘",
+   "噓",
+   "嚢",
+   "團",
+   "坏",
+   "域",
+   "墾",
+   "壕",
+   "壜",
+   "央",
+   "奠",
+   "妾",
+   "婚",
+   "嬌",
+   "岑",
+   "崖",
+   "嵩",
+   "廠",
+   "弐",
+   "怖",
+   "悦",
+   "惇",
+   "慢",
+   "摘",
+   "數",
+   "晃",
+   "曜",
+   "柯",
+   "棹",
+   "楚",
+   "楷",
+   "槌",
+   "樂",
+   "樓",
+   "橇",
+   "沁",
+   "洩",
+   "渇",
+   "渺",
+   "漉",
+   "澪",
+   "瀉",
+   "瀞",
+   "炸",
+   "烈",
+   "熨",
+   "爰",
+   "爵",
+   "猴",
+   "獏",
+   "畦",
+   "疳",
+   "癌",
+   "祿",
+   "禮",
+   "禹",
+   "稜",
+   "窮",
+   "符",
+   "篇",
+   "篤",
+   "紆",
+   "紗",
+   "絎",
+   "緊",
+   "緯",
+   "縛",
+   "繩",
+   "罌",
+   "羲",
+   "翔",
+   "聘",
+   "聡",
+   "聰",
+   "肢",
+   "膜",
+   "膵",
+   "與",
+   "舵",
+   "舷",
+   "萵",
+   "蔓",
+   "藝",
+   "虞",
+   "蛾",
+   "蜷",
+   "蝕",
+   "衡",
+   "裡",
+   "諾",
+   "譚",
+   "豚",
+   "賃",
+   "贋",
+   "赫",
+   "趺",
+   "輔",
+   "轌",
+   "述",
+   "逗",
+   "錐",
+   "鎬",
+   "鞠",
+   "頤",
+   "頴",
+   "馨",
+   "鬚",
+   "魏",
+   "魑",
+   "鮒",
+   "鰕",
+   "鰤",
+   "鱶",
+   "鵬",
+   "麵",
+   "亞",
+   "価",
+   "妃",
+   "寐",
+   "嶼",
+   "巳",
+   "弧",
+   "梓",
+   "疽",
+   "租",
+   "繪",
+   "鎖",
+   "顧"
+  ],
+  "comp_last_char": [
+   "山",
+   "田",
+   "ん",
+   "川",
+   "町",
+   "う",
+   "東",
+   "野",
+   "島",
+   "い",
+   "原",
+   "ン",
+   "り",
+   "南",
+   "西",
+   "条",
+   "本",
+   "ー",
+   "の",
+   "院",
+   "学",
+   "業",
+   "上",
+   "大",
+   "崎",
+   "と",
+   "下",
+   "く",
+   "北",
+   "が",
+   "ル",
+   "し",
+   "き",
+   "寺",
+   "ス",
+   "谷",
+   "内",
+   "は",
+   "沢",
+   "子",
+   "を",
+   "ト",
+   "中",
+   "屋",
+   "園",
+   "井",
+   "線",
+   "木",
+   "前",
+   "通",
+   "台",
+   "ク",
+   "ノ",
+   "イ",
+   "新",
+   "岡",
+   "峠",
+   "橋",
+   "地",
+   "つ",
+   "口",
+   "丘",
+   "津",
+   "ウ",
+   "道",
+   "浦",
+   "所",
+   "宮",
+   "郷",
+   "浜",
+   "岳",
+   "み",
+   "庫",
+   "平",
+   "石",
+   "水",
+   "戸",
+   "校",
+   "和",
+   "ム",
+   "っ",
+   "松",
+   "京",
+   "合",
+   "ド",
+   "ち",
+   "ら",
+   "部",
+   "会",
+   "場",
+   "倉",
+   "坂",
+   "路",
+   "け",
+   "森",
+   "生",
+   "見",
+   "立",
+   "行",
+   "瀬",
+   "神",
+   "里",
+   "十",
+   "日",
+   "別",
+   "尾",
+   "塚",
+   "ま",
+   "方",
+   "代",
+   "畑",
+   "池",
+   "間",
+   "門",
+   "リ",
+   "保",
+   "泉",
+   "れ",
+   "根",
+   "か",
+   "光",
+   "等",
+   "村",
+   "堂",
+   "鼻",
+   "ア",
+   "江",
+   "城",
+   "林",
+   "郎",
+   "工",
+   "人",
+   "た",
+   "金",
+   "め",
+   "ラ",
+   "丸",
+   "央",
+   "草",
+   "岩",
+   "手",
+   "市",
+   "産",
+   "国",
+   "事",
+   "タ",
+   "賀",
+   "館",
+   "ツ",
+   "え",
+   "品",
+   "之",
+   "物",
+   "元",
+   "敷",
+   "ろ",
+   "王",
+   "海",
+   "機",
+   "丁",
+   "プ",
+   "三",
+   "形",
+   "沼",
+   "滝",
+   "送",
+   "高",
+   "ズ",
+   "月",
+   "峰",
+   "越",
+   "一",
+   "吉",
+   "葉",
+   "尻",
+   "岬",
+   "成",
+   "お",
+   "阪",
+   "目",
+   "幡",
+   "後",
+   "グ",
+   "港",
+   "良",
+   "明",
+   "家",
+   "な",
+   "羽",
+   "栄",
+   "番",
+   "気",
+   "こ",
+   "向",
+   "設",
+   "陽",
+   "楽",
+   "社",
+   "キ",
+   "マ",
+   "蔵",
+   "カ",
+   "色",
+   "茂",
+   "州",
+   "分",
+   "ぎ",
+   "ロ",
+   "て",
+   "洋",
+   "出",
+   "チ",
+   "科",
+   "び",
+   "名",
+   "げ",
+   "藤",
+   "幌",
+   "じ",
+   "湾",
+   "せ",
+   "甲",
+   "シ",
+   "頭",
+   "船",
+   "竹",
+   "鉄",
+   "字",
+   "府",
+   "る",
+   "黒",
+   "信",
+   "店",
+   "聞",
+   "天",
+   "小",
+   "河",
+   "柳",
+   "ジ",
+   "文",
+   "六",
+   "居",
+   "ね",
+   "免",
+   "ブ",
+   "波",
+   "オ",
+   "ッ",
+   "庄",
+   "知",
+   "取",
+   "馬",
+   "さ",
+   "発",
+   "花",
+   "鳥",
+   "ミ",
+   "長",
+   "コ",
+   "ケ",
+   "ず",
+   "わ",
+   "入",
+   "岸",
+   "都",
+   "俣",
+   "宿",
+   "心",
+   "堀",
+   "フ",
+   "済",
+   "広",
+   "二",
+   "来",
+   "皇",
+   "古",
+   "法",
+   "八",
+   "士",
+   "造",
+   "ナ",
+   "師",
+   "者",
+   "や",
+   "車",
+   "白",
+   "軒",
+   "音",
+   "潟",
+   "乙",
+   "久",
+   "作",
+   "極",
+   "穂",
+   "薬",
+   "福",
+   "ビ",
+   "切",
+   "又",
+   "ば",
+   "売",
+   "多",
+   "御",
+   "治",
+   "力",
+   "す",
+   "術",
+   "玉",
+   "雲",
+   "武",
+   "際",
+   "化",
+   "油",
+   "土",
+   "組",
+   "も",
+   "面",
+   "秦",
+   "佐",
+   "器",
+   "報",
+   "安",
+   "関",
+   "奥",
+   "友",
+   "連",
+   "風",
+   "命",
+   "団",
+   "袋",
+   "ヤ",
+   "永",
+   "住",
+   "動",
+   "巻",
+   "体",
+   "室",
+   "板",
+   "だ",
+   "足",
+   "徳",
+   "笠",
+   "版",
+   "端",
+   "輪",
+   "辺",
+   "レ",
+   "納",
+   "用",
+   "紙",
+   "栖",
+   "美",
+   "属",
+   "積",
+   "殿",
+   "育",
+   "角",
+   "横",
+   "そ",
+   "熊",
+   "牧",
+   "理",
+   "空",
+   "裏",
+   "局",
+   "窪",
+   "正",
+   "砂",
+   "富",
+   "布",
+   "材",
+   "桂",
+   "衛",
+   "開",
+   "ィ",
+   "杉",
+   "渡",
+   "湖",
+   "坊",
+   "民",
+   "須",
+   "ご",
+   "芝",
+   "曲",
+   "ダ",
+   "外",
+   "触",
+   "盤",
+   "電",
+   "ぶ",
+   "星",
+   "垣",
+   "桜",
+   "進",
+   "旭",
+   "百",
+   "世",
+   "勢",
+   "境",
+   "主",
+   "荘",
+   "研",
+   "塩",
+   "寿",
+   "幸",
+   "ニ",
+   "房",
+   "ワ",
+   "坪",
+   "料",
+   "興",
+   "教",
+   "加",
+   "第",
+   "鴨",
+   "流",
+   "に",
+   "陸",
+   "女",
+   "性",
+   "様",
+   "刷",
+   "峨",
+   "数",
+   "屯",
+   "陵",
+   "同",
+   "呂",
+   "急",
+   "梨",
+   "源",
+   "七",
+   "毛",
+   "仏",
+   "宅",
+   "志",
+   "醐",
+   "掛",
+   "菱",
+   "険",
+   "万",
+   "今",
+   "券",
+   "年",
+   "貝",
+   "鋼",
+   "鹿",
+   "五",
+   "寒",
+   "影",
+   "語",
+   "計",
+   "梅",
+   "重",
+   "ギ",
+   "利",
+   "沖",
+   "湯",
+   "竜",
+   "筋",
+   "軍",
+   "宝",
+   "無",
+   "辻",
+   "牛",
+   "愛",
+   "党",
+   "商",
+   "エ",
+   "秋",
+   "若",
+   "虫",
+   "摩",
+   "緑",
+   "千",
+   "類",
+   "郡",
+   "で",
+   "脇",
+   "当",
+   "豊",
+   "半",
+   "械",
+   "縄",
+   "ネ",
+   "ガ",
+   "便",
+   "総",
+   "勝",
+   "経",
+   "妻",
+   "張",
+   "華",
+   "太",
+   "球",
+   "言",
+   "べ",
+   "メ",
+   "盛",
+   "篭",
+   "あ",
+   "丹",
+   "販",
+   "モ",
+   "書",
+   "老",
+   "香",
+   "麻",
+   "丙",
+   "唄",
+   "庁",
+   "磨",
+   "能",
+   "定",
+   "栗",
+   "赤",
+   "段",
+   "首",
+   "養",
+   "バ",
+   "磯",
+   "具",
+   "区",
+   "焼",
+   "築",
+   "割",
+   "度",
+   "朝",
+   "網",
+   "芸",
+   "サ",
+   "洲",
+   "稲",
+   "食",
+   "座",
+   "ゴ",
+   "四",
+   "真",
+   "針",
+   "不",
+   "洞",
+   "豆",
+   "雨",
+   "青",
+   "ひ",
+   "仙",
+   "夫",
+   "助",
+   "湊",
+   "蘭",
+   "峯",
+   "節",
+   "米",
+   "陰",
+   "ぜ",
+   "セ",
+   "実",
+   "輸",
+   "珠",
+   "ょ",
+   "似",
+   "引",
+   "渕",
+   "病",
+   "着",
+   "全",
+   "災",
+   "火",
+   "管",
+   "銀",
+   "矢",
+   "霊",
+   "ど",
+   "仁",
+   "守",
+   "満",
+   "股",
+   "ュ",
+   "穴",
+   "邦",
+   "淵",
+   "舞",
+   "脈",
+   "話",
+   "雪",
+   "骨",
+   "ソ",
+   "念",
+   "相",
+   "荷",
+   "厨",
+   "県",
+   "釜",
+   "堤",
+   "留",
+   "テ",
+   "仲",
+   "峡",
+   "枝",
+   "滑",
+   "亜",
+   "建",
+   "末",
+   "片",
+   "登",
+   "善",
+   "症",
+   "政",
+   "曙",
+   "協",
+   "印",
+   "巣",
+   "界",
+   "ゃ",
+   "岐",
+   "ハ",
+   "英",
+   "近",
+   "持",
+   "易",
+   "活",
+   "帯",
+   "貫",
+   "溝",
+   "紫",
+   "隈",
+   "時",
+   "黄",
+   "詰",
+   "鶴",
+   "九",
+   "務",
+   "声",
+   "専",
+   "束",
+   "舟",
+   "鏡",
+   "営",
+   "塔",
+   "恵",
+   "宗",
+   "嶺",
+   "的",
+   "街",
+   "ぷ",
+   "母",
+   "祭",
+   "種",
+   "箱",
+   "苑",
+   "軽",
+   "濃",
+   "運",
+   "意",
+   "身",
+   "ゆ",
+   "柱",
+   "泊",
+   "現",
+   "磐",
+   "付",
+   "判",
+   "織",
+   "剛",
+   "柄",
+   "父",
+   "ゅ",
+   "味",
+   "先",
+   "我",
+   "有",
+   "飼",
+   "打",
+   "算",
+   "込",
+   "顔",
+   "ぼ",
+   "司",
+   "官",
+   "灯",
+   "荒",
+   "閑",
+   "精",
+   "衣",
+   "ぐ",
+   "働",
+   "干",
+   "灰",
+   "転",
+   "達",
+   "庭",
+   "蓮",
+   "棚",
+   "魚",
+   "吹",
+   "才",
+   "春",
+   "糸",
+   "ボ",
+   "円",
+   "指",
+   "槻",
+   "錦",
+   "奈",
+   "寄",
+   "隅",
+   "む",
+   "夜",
+   "対",
+   "澤",
+   "表",
+   "駒",
+   "状",
+   "雀",
+   "壇",
+   "式",
+   "慶",
+   "粉",
+   "訪",
+   "鮮",
+   "待",
+   "那",
+   "亀",
+   "扇",
+   "染",
+   "楠",
+   "点",
+   "狩",
+   "熱",
+   "舎",
+   "串",
+   "備",
+   "柴",
+   "歳",
+   "糖",
+   "鍋",
+   "防",
+   "令",
+   "清",
+   "照",
+   "究",
+   "論",
+   "護",
+   "両",
+   "羅",
+   "密",
+   "床",
+   "底",
+   "役",
+   "丈",
+   "喜",
+   "折",
+   "歌",
+   "犬",
+   "萩",
+   "刈",
+   "員",
+   "嶋",
+   "応",
+   "朗",
+   "樹",
+   "置",
+   "阜",
+   "雄",
+   "婦",
+   "桑",
+   "画",
+   "籠",
+   "伏",
+   "反",
+   "号",
+   "札",
+   "模",
+   "舘",
+   "ゲ",
+   "パ",
+   "智",
+   "男",
+   "紅",
+   "遠",
+   "陀",
+   "更",
+   "義",
+   "親",
+   "ザ",
+   "好",
+   "族",
+   "権",
+   "琴",
+   "腰",
+   "飯",
+   "唐",
+   "柏",
+   "橘",
+   "績",
+   "ざ",
+   "亭",
+   "将",
+   "早",
+   "離",
+   "労",
+   "素",
+   "落",
+   "集",
+   "介",
+   "曽",
+   "次",
+   "氏",
+   "直",
+   "菓",
+   "識",
+   "情",
+   "條",
+   "棒",
+   "ベ",
+   "払",
+   "桃",
+   "沿",
+   "盟",
+   "縁",
+   "差",
+   "麦",
+   "咲",
+   "尊",
+   "戦",
+   "陣",
+   "餅",
+   "乗",
+   "延",
+   "査",
+   "瓶",
+   "眼",
+   "笹",
+   "篠",
+   "紀",
+   "冠",
+   "圧",
+   "宕",
+   "柿",
+   "走",
+   "朱",
+   "果",
+   "比",
+   "調",
+   "難",
+   "ぱ",
+   "壁",
+   "姫",
+   "枚",
+   "温",
+   "血",
+   "髪",
+   "冶",
+   "告",
+   "庵",
+   "礼",
+   "絵",
+   "茶",
+   "在",
+   "昌",
+   "期",
+   "粒",
+   "酸",
+   "受",
+   "常",
+   "迫",
+   "伊",
+   "俵",
+   "刀",
+   "臣",
+   "量",
+   "デ",
+   "剤",
+   "客",
+   "感",
+   "省",
+   "銭",
+   "匠",
+   "噌",
+   "察",
+   "洗",
+   "畠",
+   "鷹",
+   "ァ",
+   "並",
+   "側",
+   "剣",
+   "想",
+   "筆",
+   "聖",
+   "蔭",
+   "薮",
+   "詞",
+   "左",
+   "深",
+   "潮",
+   "頃",
+   "領",
+   "ふ",
+   "曜",
+   "椿",
+   "漢",
+   "任",
+   "修",
+   "害",
+   "皮",
+   "砲",
+   "綿",
+   "製",
+   "賞",
+   "酒",
+   "夏",
+   "提",
+   "殻",
+   "蓋",
+   "鬼",
+   "龍",
+   "伝",
+   "忠",
+   "歩",
+   "然",
+   "維",
+   "得",
+   "負",
+   "験",
+   "句",
+   "爪",
+   "証",
+   "財",
+   "追",
+   "郭",
+   "鈴",
+   "乳",
+   "彦",
+   "玄",
+   "瓦",
+   "菜",
+   "質",
+   "鼓",
+   "ペ",
+   "予",
+   "児",
+   "刊",
+   "息",
+   "療",
+   "膳",
+   "衡",
+   "装",
+   "辛",
+   "鐵",
+   "桧",
+   "腹",
+   "堅",
+   "層",
+   "斗",
+   "殊",
+   "淀",
+   "灘",
+   "称",
+   "羊",
+   "蒲",
+   "飛",
+   "駅",
+   "麗",
+   "ポ",
+   "位",
+   "炎",
+   "芦",
+   "観",
+   "酵",
+   "ェ",
+   "変",
+   "練",
+   "藪",
+   "託",
+   "副",
+   "呉",
+   "媛",
+   "枕",
+   "添",
+   "公",
+   "君",
+   "學",
+   "宇",
+   "帳",
+   "廻",
+   "榎",
+   "続",
+   "萱",
+   "跡",
+   "よ",
+   "伴",
+   "暮",
+   "猿",
+   "終",
+   "綱",
+   "苗",
+   "諸",
+   "鶏",
+   "夷",
+   "幕",
+   "弾",
+   "旧",
+   "服",
+   "泥",
+   "炭",
+   "狐",
+   "粧",
+   "臼",
+   "議",
+   "鉢",
+   "鐘",
+   "ぺ",
+   "制",
+   "構",
+   "糠",
+   "耕",
+   "遊",
+   "鷺",
+   "休",
+   "兵",
+   "姓",
+   "斯",
+   "植",
+   "椎",
+   "漿",
+   "脳",
+   "腸",
+   "舌",
+   "農",
+   "隊",
+   "鞍",
+   "韓",
+   "ヌ",
+   "ャ",
+   "卵",
+   "塵",
+   "律",
+   "杭",
+   "狗",
+   "由",
+   "肉",
+   "貴",
+   "ぬ",
+   "壱",
+   "夢",
+   "容",
+   "嶽",
+   "最",
+   "核",
+   "鉾",
+   "頂",
+   "ほ",
+   "健",
+   "初",
+   "劇",
+   "消",
+   "烏",
+   "章",
+   "蛉",
+   "視",
+   "附",
+   "非",
+   "鳴",
+   "問",
+   "断",
+   "替",
+   "欠",
+   "猫",
+   "琶",
+   "緬",
+   "脚",
+   "芋",
+   "覚",
+   "記",
+   "速",
+   "順",
+   "鼎",
+   "ヒ",
+   "参",
+   "旗",
+   "樋",
+   "壷",
+   "寸",
+   "射",
+   "帆",
+   "弟",
+   "悪",
+   "放",
+   "斎",
+   "旅",
+   "晃",
+   "案",
+   "窓",
+   "談",
+   "請",
+   "迎",
+   "釈",
+   "題",
+   "乱",
+   "件",
+   "供",
+   "倍",
+   "卸",
+   "忌",
+   "狭",
+   "畔",
+   "祉",
+   "秀",
+   "習",
+   "薄",
+   "蘇",
+   "雁",
+   "乏",
+   "交",
+   "僧",
+   "哲",
+   "塾",
+   "弓",
+   "強",
+   "拝",
+   "摺",
+   "昇",
+   "榑",
+   "樺",
+   "洪",
+   "液",
+   "申",
+   "約",
+   "翁",
+   "荻",
+   "菌",
+   "迦",
+   "鷲",
+   "麺",
+   "価",
+   "儀",
+   "型",
+   "戒",
+   "斐",
+   "施",
+   "檜",
+   "秤",
+   "菅",
+   "頼",
+   "ぽ",
+   "再",
+   "杏",
+   "汁",
+   "為",
+   "畿",
+   "竃",
+   "細",
+   "腐",
+   "苔",
+   "講",
+   "限",
+   "陶",
+   "霧",
+   "鳩",
+   "ゼ",
+   "ピ",
+   "冬",
+   "双",
+   "景",
+   "望",
+   "桶",
+   "樽",
+   "示",
+   "脂",
+   "腺",
+   "釣",
+   "障",
+   "基",
+   "存",
+   "巳",
+   "康",
+   "弥",
+   "恋",
+   "梁",
+   "歯",
+   "汰",
+   "浪",
+   "環",
+   "竺",
+   "胞",
+   "伯",
+   "倒",
+   "堰",
+   "抜",
+   "撲",
+   "減",
+   "索",
+   "侍",
+   "國",
+   "失",
+   "孝",
+   "岱",
+   "巾",
+   "押",
+   "泰",
+   "渓",
+   "竪",
+   "粟",
+   "紋",
+   "膜",
+   "象",
+   "魂",
+   "例",
+   "宜",
+   "杵",
+   "死",
+   "準",
+   "獄",
+   "甘",
+   "盲",
+   "童",
+   "糞",
+   "許",
+   "降",
+   "露",
+   "へ",
+   "乃",
+   "使",
+   "兄",
+   "呑",
+   "巽",
+   "技",
+   "敗",
+   "晶",
+   "標",
+   "礁",
+   "繋",
+   "罐",
+   "艦",
+   "芳",
+   "莱",
+   "葵",
+   "蜂",
+   "融",
+   "輿",
+   "銘",
+   "づ",
+   "像",
+   "喰",
+   "答",
+   "賊",
+   "躅",
+   "鎮",
+   "階",
+   "ホ",
+   "争",
+   "勤",
+   "図",
+   "垂",
+   "堺",
+   "導",
+   "惣",
+   "槌",
+   "珍",
+   "破",
+   "祥",
+   "穀",
+   "肩",
+   "茅",
+   "薩",
+   "袖",
+   "説",
+   "鞆",
+   "願",
+   "魔",
+   "ぴ",
+   "医",
+   "廿",
+   "燕",
+   "起",
+   "鈷",
+   "丑",
+   "共",
+   "増",
+   "寅",
+   "展",
+   "怪",
+   "揚",
+   "撃",
+   "杖",
+   "格",
+   "楼",
+   "殖",
+   "汗",
+   "浮",
+   "益",
+   "祝",
+   "箪",
+   "級",
+   "署",
+   "蒻",
+   "蔦",
+   "蝠",
+   "衰",
+   "解",
+   "読",
+   "蹊",
+   "違",
+   "乾",
+   "勇",
+   "右",
+   "婆",
+   "振",
+   "止",
+   "氷",
+   "測",
+   "煙",
+   "燃",
+   "突",
+   "臓",
+   "蝶",
+   "阿",
+   "除",
+   "霞",
+   "仮",
+   "列",
+   "尺",
+   "巴",
+   "投",
+   "拳",
+   "杯",
+   "梶",
+   "漕",
+   "煥",
+   "税",
+   "竈",
+   "缶",
+   "耳",
+   "職",
+   "背",
+   "菊",
+   "蛸",
+   "讃",
+   "鴻",
+   "鼠",
+   "傷",
+   "后",
+   "寝",
+   "幅",
+   "恩",
+   "胡",
+   "藍",
+   "蛇",
+   "静",
+   "鵜",
+   "典",
+   "包",
+   "単",
+   "圏",
+   "塁",
+   "姉",
+   "尼",
+   "幣",
+   "攻",
+   "昼",
+   "梗",
+   "梯",
+   "検",
+   "瓜",
+   "絡",
+   "給",
+   "罪",
+   "考",
+   "葛",
+   "葬",
+   "辰",
+   "退",
+   "銅",
+   "録",
+   "革",
+   "駄",
+   "騨",
+   "鶯",
+   "削",
+   "威",
+   "宙",
+   "審",
+   "帽",
+   "弁",
+   "態",
+   "敬",
+   "棋",
+   "短",
+   "笑",
+   "簾",
+   "裂",
+   "規",
+   "輝",
+   "閉",
+   "余",
+   "刺",
+   "博",
+   "厩",
+   "尽",
+   "整",
+   "株",
+   "樟",
+   "櫛",
+   "燈",
+   "特",
+   "璃",
+   "略",
+   "皆",
+   "睦",
+   "胸",
+   "腕",
+   "艇",
+   "蓉",
+   "蟹",
+   "誠",
+   "資",
+   "酉",
+   "鍵",
+   "震",
+   "鬚",
+   "兜",
+   "妙",
+   "挙",
+   "箸",
+   "ヨ",
+   "乞",
+   "何",
+   "函",
+   "刑",
+   "叡",
+   "塀",
+   "墨",
+   "夕",
+   "娘",
+   "季",
+   "尋",
+   "少",
+   "尚",
+   "弦",
+   "抗",
+   "択",
+   "暦",
+   "淑",
+   "渉",
+   "渚",
+   "瞳",
+   "筈",
+   "茎",
+   "蒔",
+   "薙",
+   "袴",
+   "謝",
+   "邸",
+   "釘",
+   "鉱",
+   "錐",
+   "鍬",
+   "閤",
+   "隆",
+   "隷",
+   "靴",
+   "響",
+   "飴",
+   "齢",
+   "域",
+   "始",
+   "崩",
+   "幾",
+   "慈",
+   "憲",
+   "斑",
+   "決",
+   "芽",
+   "萌",
+   "螺",
+   "遣",
+   "ぞ",
+   "ユ",
+   "可",
+   "唱",
+   "回",
+   "垢",
+   "妹",
+   "己",
+   "昭",
+   "暗",
+   "栃",
+   "桁",
+   "楯",
+   "櫓",
+   "歴",
+   "泌",
+   "溶",
+   "激",
+   "畝",
+   "竿",
+   "糧",
+   "緒",
+   "肺",
+   "艶",
+   "苅",
+   "苦",
+   "蕊",
+   "遍",
+   "雑",
+   "飲",
+   "麟",
+   "ヴ",
+   "写",
+   "噺",
+   "孫",
+   "屓",
+   "庚",
+   "拠",
+   "架",
+   "毒",
+   "浅",
+   "犀",
+   "瑚",
+   "笛",
+   "簀",
+   "自",
+   "蕨",
+   "兼",
+   "処",
+   "呼",
+   "囚",
+   "奏",
+   "宣",
+   "帖",
+   "廊",
+   "復",
+   "悠",
+   "惑",
+   "戊",
+   "接",
+   "旦",
+   "映",
+   "昧",
+   "朋",
+   "櫃",
+   "残",
+   "派",
+   "牢",
+   "狂",
+   "癖",
+   "砥",
+   "祖",
+   "藁",
+   "蛛",
+   "蛾",
+   "蝉",
+   "衆",
+   "裡",
+   "謀",
+   "賣",
+   "載",
+   "轟",
+   "鉛",
+   "鍔",
+   "鑑",
+   "闇",
+   "靂",
+   "鮨",
+   "鯛",
+   "鴎",
+   "麿",
+   "以",
+   "周",
+   "塙",
+   "嵐",
+   "思",
+   "恒",
+   "晴",
+   "樫",
+   "爺",
+   "磁",
+   "禄",
+   "簡",
+   "繊",
+   "罰",
+   "肥",
+   "補",
+   "買",
+   "返",
+   "選",
+   "鎌",
+   "飾",
+   "与",
+   "傘",
+   "兎",
+   "勲",
+   "咋",
+   "封",
+   "屈",
+   "帝",
+   "席",
+   "懐",
+   "挽",
+   "換",
+   "斉",
+   "昨",
+   "晩",
+   "曇",
+   "杜",
+   "柔",
+   "桝",
+   "榴",
+   "槽",
+   "橡",
+   "欧",
+   "求",
+   "浄",
+   "涌",
+   "澄",
+   "率",
+   "玖",
+   "瑛",
+   "眉",
+   "碗",
+   "笥",
+   "結",
+   "統",
+   "縞",
+   "繁",
+   "翠",
+   "肪",
+   "胃",
+   "臥",
+   "致",
+   "般",
+   "茜",
+   "蕃",
+   "蝿",
+   "裳",
+   "覇",
+   "賢",
+   "踊",
+   "鉋",
+   "鐡",
+   "隠",
+   "霜",
+   "額",
+   "鯖",
+   "鰐",
+   "鶉",
+   "値",
+   "冷",
+   "序",
+   "悲",
+   "捨",
+   "枯",
+   "涎",
+   "漁",
+   "炉",
+   "藻",
+   "逆",
+   "候",
+   "凰",
+   "勘",
+   "厳",
+   "埴",
+   "奄",
+   "奸",
+   "寧",
+   "幼",
+   "彌",
+   "彩",
+   "彰",
+   "採",
+   "攣",
+   "敵",
+   "暁",
+   "朔",
+   "未",
+   "渋",
+   "漉",
+   "漏",
+   "濯",
+   "燭",
+   "献",
+   "琢",
+   "璋",
+   "異",
+   "疋",
+   "盆",
+   "督",
+   "碧",
+   "確",
+   "禽",
+   "程",
+   "窯",
+   "筒",
+   "粕",
+   "粥",
+   "綾",
+   "群",
+   "脱",
+   "舶",
+   "薗",
+   "螂",
+   "裁",
+   "覧",
+   "討",
+   "評",
+   "詮",
+   "認",
+   "貞",
+   "遇",
+   "酘",
+   "釧",
+   "鎚",
+   "鬱",
+   "鯉",
+   "鳶",
+   "鴬",
+   "乎",
+   "仕",
+   "兆",
+   "匂",
+   "嫌",
+   "嬉",
+   "往",
+   "忘",
+   "戎",
+   "損",
+   "机",
+   "李",
+   "汐",
+   "潭",
+   "猪",
+   "獣",
+   "疑",
+   "痕",
+   "砕",
+   "禁",
+   "禰",
+   "私",
+   "筅",
+   "範",
+   "翼",
+   "至",
+   "舗",
+   "薪",
+   "虎",
+   "課",
+   "過",
+   "酌",
+   "鋺",
+   "鎧",
+   "闘",
+   "顧",
+   "鰻",
+   "ョ",
+   "俊",
+   "借",
+   "偏",
+   "刃",
+   "創",
+   "午",
+   "吾",
+   "嘉",
+   "塞",
+   "奇",
+   "如",
+   "婚",
+   "孔",
+   "寛",
+   "慧",
+   "懸",
+   "承",
+   "拓",
+   "栂",
+   "桐",
+   "棟",
+   "槙",
+   "檀",
+   "潤",
+   "瀞",
+   "烟",
+   "爆",
+   "狼",
+   "瑞",
+   "皿",
+   "禅",
+   "系",
+   "編",
+   "聴",
+   "肝",
+   "菰",
+   "葭",
+   "蓑",
+   "薇",
+   "藝",
+   "虹",
+   "裟",
+   "裾",
+   "要",
+   "誕",
+   "轡",
+   "週",
+   "鍾",
+   "雛",
+   "雷",
+   "駐",
+   "髷",
+   "鮫",
+   "鯨",
+   "鱈",
+   "嚢",
+   "壊",
+   "寮",
+   "忍",
+   "暴",
+   "炊",
+   "煮",
+   "畳",
+   "箕",
+   "詩",
+   "豪",
+   "迷",
+   "途",
+   "伽",
+   "促",
+   "俗",
+   "凱",
+   "凶",
+   "刻",
+   "則",
+   "功",
+   "卯",
+   "即",
+   "吟",
+   "啓",
+   "嘆",
+   "嘗",
+   "嘩",
+   "因",
+   "執",
+   "壌",
+   "奴",
+   "媒",
+   "嫁",
+   "履",
+   "巧",
+   "従",
+   "忙",
+   "快",
+   "恥",
+   "恭",
+   "患",
+   "悦",
+   "慎",
+   "慨",
+   "抵",
+   "搬",
+   "擅",
+   "散",
+   "朴",
+   "杓",
+   "榊",
+   "泡",
+   "泳",
+   "浴",
+   "淆",
+   "淳",
+   "漆",
+   "漫",
+   "潜",
+   "濤",
+   "濱",
+   "灸",
+   "猩",
+   "盗",
+   "碆",
+   "祐",
+   "筍",
+   "粂",
+   "紬",
+   "縫",
+   "纏",
+   "耀",
+   "胴",
+   "腫",
+   "舫",
+   "萄",
+   "蕗",
+   "薫",
+   "藩",
+   "蛍",
+   "袁",
+   "貢",
+   "超",
+   "跳",
+   "践",
+   "輌",
+   "遷",
+   "邨",
+   "鋲",
+   "錣",
+   "閣",
+   "阻",
+   "陳",
+   "隣",
+   "鞋",
+   "頌",
+   "飩",
+   "鰌",
+   "鱒",
+   "鳳",
+   "麓",
+   "堵",
+   "微",
+   "慢",
+   "戻",
+   "改",
+   "熟",
+   "独",
+   "眠",
+   "砧",
+   "符",
+   "純",
+   "訓",
+   "豚",
+   "鈍",
+   "鞘",
+   "ゑ",
+   "ゾ",
+   "ヘ",
+   "亥",
+   "伸",
+   "佑",
+   "併",
+   "其",
+   "凧",
+   "卿",
+   "喝",
+   "尿",
+   "崖",
+   "廃",
+   "括",
+   "摂",
+   "敏",
+   "是",
+   "曹",
+   "曾",
+   "樓",
+   "汀",
+   "沓",
+   "泣",
+   "涼",
+   "燐",
+   "猴",
+   "班",
+   "皺",
+   "稽",
+   "箇",
+   "籤",
+   "糊",
+   "紐",
+   "紛",
+   "翌",
+   "腱",
+   "航",
+   "艷",
+   "茸",
+   "蜜",
+   "蟇",
+   "賎",
+   "酔",
+   "鉤",
+   "錠",
+   "項",
+   "顎",
+   "餡",
+   "髭",
+   "ぃ",
+   "互",
+   "俄",
+   "偉",
+   "凍",
+   "勒",
+   "卦",
+   "叉",
+   "咄",
+   "堝",
+   "姑",
+   "慮",
+   "憶",
+   "授",
+   "柑",
+   "椚",
+   "楢",
+   "檎",
+   "氣",
+   "沙",
+   "没",
+   "涙",
+   "畜",
+   "痴",
+   "眄",
+   "祢",
+   "稜",
+   "笏",
+   "筏",
+   "箒",
+   "篝",
+   "而",
+   "臂",
+   "茨",
+   "葎",
+   "葱",
+   "薑",
+   "蚊",
+   "蛙",
+   "裸",
+   "褄",
+   "誉",
+   "誌",
+   "譜",
+   "豕",
+   "貨",
+   "貰",
+   "輛",
+   "轤",
+   "配",
+   "鋏",
+   "鏑",
+   "随",
+   "颪",
+   "鰯",
+   "鰹",
+   "伺",
+   "侃",
+   "侶",
+   "僕",
+   "億",
+   "兢",
+   "効",
+   "勉",
+   "勧",
+   "収",
+   "吊",
+   "吸",
+   "哉",
+   "唳",
+   "唾",
+   "喉",
+   "喚",
+   "坐",
+   "坤",
+   "培",
+   "埼",
+   "塒",
+   "墓",
+   "婿",
+   "嬢",
+   "寂",
+   "尤",
+   "幹",
+   "幽",
+   "庸",
+   "弘",
+   "徒",
+   "徴",
+   "徹",
+   "怒",
+   "怠",
+   "恐",
+   "恤",
+   "悩",
+   "憂",
+   "招",
+   "拾",
+   "掘",
+   "揮",
+   "摘",
+   "撰",
+   "操",
+   "擲",
+   "支",
+   "昆",
+   "暖",
+   "杣",
+   "枠",
+   "柩",
+   "栓",
+   "梵",
+   "榮",
+   "槇",
+   "檗",
+   "欽",
+   "歇",
+   "氈",
+   "汲",
+   "溪",
+   "滅",
+   "瀧",
+   "牆",
+   "犯",
+   "獲",
+   "珂",
+   "琳",
+   "疆",
+   "疣",
+   "疹",
+   "痛",
+   "瘡",
+   "瞭",
+   "瞰",
+   "砿",
+   "碩",
+   "礎",
+   "礬",
+   "稗",
+   "稿",
+   "窮",
+   "箔",
+   "篁",
+   "絃",
+   "絹",
+   "緋",
+   "縮",
+   "纈",
+   "聯",
+   "肌",
+   "胆",
+   "臭",
+   "蒋",
+   "蒸",
+   "蔽",
+   "藉",
+   "蚕",
+   "蛮",
+   "蟀",
+   "蠣",
+   "衷",
+   "複",
+   "褌",
+   "詢",
+   "諤",
+   "豹",
+   "貸",
+   "賓",
+   "蹄",
+   "躬",
+   "軟",
+   "軸",
+   "輔",
+   "轍",
+   "這",
+   "還",
+   "邇",
+   "邑",
+   "鈑",
+   "錘",
+   "錬",
+   "閃",
+   "闢",
+   "隔",
+   "頚",
+   "頸",
+   "顛",
+   "駈",
+   "騎",
+   "鬢",
+   "魁",
+   "鴫",
+   "鹸",
+   "龜",
+   "亞",
+   "亡",
+   "仇",
+   "佛",
+   "侯",
+   "委",
+   "憩",
+   "扁",
+   "於",
+   "狸",
+   "猛",
+   "盧",
+   "筌",
+   "筑",
+   "継",
+   "被",
+   "裕",
+   "鎖",
+   "雌",
+   "ヂ",
+   "丼",
+   "亂",
+   "亦",
+   "企",
+   "低",
+   "佳",
+   "俳",
+   "傭",
+   "債",
+   "優",
+   "准",
+   "凜",
+   "勅",
+   "匹",
+   "卑",
+   "卒",
+   "卓",
+   "厚",
+   "含",
+   "��",
+   "圓",
+   "均",
+   "奠",
+   "奪",
+   "姥",
+   "嬌",
+   "完",
+   "宸",
+   "岑",
+   "巡",
+   "帰",
+   "彼",
+   "径",
+   "悸",
+   "惜",
+   "控",
+   "援",
+   "搦",
+   "撮",
+   "救",
+   "斂",
+   "旋",
+   "既",
+   "柵",
+   "棄",
+   "椋",
+   "楊",
+   "槍",
+   "欅",
+   "欲",
+   "毬",
+   "沈",
+   "注",
+   "洙",
+   "洛",
+   "浩",
+   "涯",
+   "淺",
+   "游",
+   "滓",
+   "滸",
+   "滿",
+   "澗",
+   "爛",
+   "狢",
+   "獺",
+   "瑙",
+   "璽",
+   "痺",
+   "癌",
+   "監",
+   "盾",
+   "矧",
+   "碇",
+   "碑",
+   "磧",
+   "票",
+   "祿",
+   "稀",
+   "策",
+   "筵",
+   "箋",
+   "籍",
+   "籏",
+   "籐",
+   "籾",
+   "紡",
+   "紹",
+   "絣",
+   "綜",
+   "緩",
+   "縦",
+   "耐",
+   "腎",
+   "膝",
+   "臨",
+   "莫",
+   "菫",
+   "葺",
+   "蓙",
+   "蕷",
+   "蘆",
+   "蛻",
+   "蝋",
+   "蝕",
+   "蟻",
+   "襟",
+   "訟",
+   "訳",
+   "診",
+   "試",
+   "諭",
+   "謬",
+   "責",
+   "賜",
+   "赦",
+   "躁",
+   "遅",
+   "邪",
+   "鉦",
+   "銃",
+   "鋸",
+   "錨",
+   "錫",
+   "鍛",
+   "鑼",
+   "靨",
+   "韮",
+   "頬",
+   "餐",
+   "駝",
+   "髄",
+   "鬘",
+   "鮎",
+   "鮒",
+   "鰊",
+   "鴇",
+   "鵑",
+   "鵬",
+   "鼾",
+   "ぢ",
+   "ゥ",
+   "也",
+   "傍",
+   "史",
+   "届",
+   "嵩",
+   "掃",
+   "杁",
+   "毅",
+   "淡",
+   "祇",
+   "絽",
+   "締",
+   "舛",
+   "蔓",
+   "賃",
+   "辞",
+   "鞠",
+   "鮪",
+   "ぇ",
+   "了",
+   "仄",
+   "他",
+   "仰",
+   "但",
+   "侭",
+   "侵",
+   "個",
+   "倫",
+   "倶",
+   "允",
+   "冰",
+   "凡",
+   "凸",
+   "劍",
+   "匙",
+   "却",
+   "厘",
+   "咳",
+   "唇",
+   "喧",
+   "嘯",
+   "囂",
+   "圭",
+   "坦",
+   "埓",
+   "墳",
+   "奔",
+   "套",
+   "姐",
+   "姚",
+   "宴",
+   "峻",
+   "崙",
+   "巨",
+   "廣",
+   "弱",
+   "彿",
+   "怙",
+   "悔",
+   "愾",
+   "憎",
+   "懃",
+   "應",
+   "拭",
+   "拵",
+   "捉",
+   "排",
+   "推",
+   "撞",
+   "攘",
+   "攫",
+   "故",
+   "斧",
+   "昂",
+   "暑",
+   "會",
+   "杷",
+   "柊",
+   "桓",
+   "梛",
+   "棠",
+   "棲",
+   "棺",
+   "椒",
+   "椴",
+   "榔",
+   "榧",
+   "樅",
+   "櫂",
+   "欄",
+   "欣",
+   "歓",
+   "殺",
+   "汞",
+   "沫",
+   "洟",
+   "涛",
+   "渦",
+   "渾",
+   "湘",
+   "湧",
+   "湿",
+   "溜",
+   "溟",
+   "漬",
+   "澱",
+   "濠",
+   "瀑",
+   "煊",
+   "狒",
+   "猶",
+   "琅",
+   "琉",
+   "瑯",
+   "瓏",
+   "甚",
+   "甥",
+   "畏",
+   "痒",
+   "痙",
+   "痩",
+   "瘤",
+   "癇",
+   "癬",
+   "看",
+   "眸",
+   "睛",
+   "硬",
+   "礦",
+   "秒",
+   "穣",
+   "笙",
+   "篦",
+   "籟",
+   "粛",
+   "紳",
+   "絲",
+   "綬",
+   "綽",
+   "緲",
+   "縢",
+   "縣",
+   "纜",
+   "罠",
+   "羹",
+   "耶",
+   "肢",
+   "膊",
+   "膨",
+   "膿",
+   "茗",
+   "荏",
+   "菖",
+   "萬",
+   "萸",
+   "董",
+   "藷",
+   "藹",
+   "蛭",
+   "蜃",
+   "蜑",
+   "蝣",
+   "衝",
+   "襄",
+   "襴",
+   "訶",
+   "諜",
+   "謙",
+   "謨",
+   "貧",
+   "賛",
+   "賽",
+   "辨",
+   "迅",
+   "迹",
+   "遥",
+   "郊",
+   "鄒",
+   "鄧",
+   "鄲",
+   "醜",
+   "鋒",
+   "鋤",
+   "錆",
+   "錢",
+   "鐚",
+   "鑓",
+   "閨",
+   "闊",
+   "闕",
+   "韻",
+   "頓",
+   "顰",
+   "餌",
+   "馮",
+   "馳",
+   "駁",
+   "騰",
+   "鬨",
+   "魎",
+   "鮭",
+   "鱗",
+   "鸞",
+   "黍",
+   "ヱ",
+   "ヲ",
+   "佃",
+   "依",
+   "係",
+   "俎",
+   "偽",
+   "儒",
+   "到",
+   "募",
+   "及",
+   "召",
+   "叭",
+   "否",
+   "囲",
+   "埔",
+   "埜",
+   "奕",
+   "妊",
+   "姜",
+   "姿",
+   "嬶",
+   "崗",
+   "必",
+   "怖",
+   "悟",
+   "慕",
+   "慙",
+   "戟",
+   "捐",
+   "措",
+   "搏",
+   "撓",
+   "擬",
+   "斜",
+   "旛",
+   "晒",
+   "普",
+   "暇",
+   "曳",
+   "朧",
+   "枡",
+   "柚",
+   "棗",
+   "楓",
+   "槃",
+   "櫞",
+   "櫟",
+   "櫻",
+   "汝",
+   "況",
+   "泪",
+   "淋",
+   "渠",
+   "潴",
+   "瀾",
+   "焔",
+   "煕",
+   "燧",
+   "燵",
+   "牌",
+   "痍",
+   "瘍",
+   "癡",
+   "盂",
+   "瞼",
+   "磋",
+   "礫",
+   "祷",
+   "秣",
+   "穢",
+   "簿",
+   "糺",
+   "絆",
+   "絞",
+   "翅",
+   "耆",
+   "肴",
+   "胖",
+   "腑",
+   "膚",
+   "艀",
+   "苧",
+   "蒡",
+   "蓬",
+   "蕉",
+   "薊",
+   "蝗",
+   "袷",
+   "訴",
+   "誤",
+   "誼",
+   "謡",
+   "賦",
+   "跏",
+   "躍",
+   "適",
+   "遺",
+   "避",
+   "邏",
+   "酢",
+   "釦",
+   "銹",
+   "雅",
+   "鞴",
+   "駆",
+   "鬆",
+   "鮑",
+   "鵄",
+   "鵞",
+   "麹",
+   "ぁ",
+   "ぅ",
+   "ぉ",
+   "ヰ",
+   "伍",
+   "佩",
+   "俑",
+   "倹",
+   "僂",
+   "儘",
+   "凛",
+   "凹",
+   "劉",
+   "升",
+   "卉",
+   "卜",
+   "厭",
+   "只",
+   "吏",
+   "呈",
+   "哀",
+   "啄",
+   "喬",
+   "嘘",
+   "噓",
+   "團",
+   "坏",
+   "堆",
+   "塗",
+   "壕",
+   "壜",
+   "壮",
+   "奨",
+   "妾",
+   "姪",
+   "娃",
+   "屁",
+   "崑",
+   "廠",
+   "弐",
+   "弔",
+   "彫",
+   "征",
+   "惇",
+   "惠",
+   "懇",
+   "探",
+   "搗",
+   "敦",
+   "數",
+   "柯",
+   "棹",
+   "楷",
+   "榛",
+   "樂",
+   "橇",
+   "此",
+   "毎",
+   "洒",
+   "混",
+   "渇",
+   "渺",
+   "澪",
+   "瀉",
+   "炸",
+   "烈",
+   "煤",
+   "煩",
+   "爰",
+   "爵",
+   "爾",
+   "猟",
+   "獏",
+   "玲",
+   "珪",
+   "甫",
+   "畦",
+   "疳",
+   "碍",
+   "禮",
+   "禹",
+   "禿",
+   "秘",
+   "篇",
+   "篤",
+   "簗",
+   "紗",
+   "紺",
+   "絶",
+   "緊",
+   "縒",
+   "縛",
+   "繩",
+   "翔",
+   "聘",
+   "聡",
+   "聰",
+   "肘",
+   "膵",
+   "舜",
+   "舵",
+   "舷",
+   "芹",
+   "苣",
+   "苫",
+   "菟",
+   "蒼",
+   "蓄",
+   "薦",
+   "薯",
+   "虚",
+   "虞",
+   "蜷",
+   "蝦",
+   "褐",
+   "詠",
+   "譚",
+   "贋",
+   "赫",
+   "軌",
+   "轌",
+   "述",
+   "透",
+   "酪",
+   "鋳",
+   "鎬",
+   "顕",
+   "馨",
+   "魅",
+   "魏",
+   "鰕",
+   "鰤",
+   "鰺",
+   "鱶",
+   "麵",
+   "ゐ",
+   "ォ",
+   "妃",
+   "孤",
+   "寐",
+   "屠",
+   "嶼",
+   "弧",
+   "梓",
+   "疽",
+   "租",
+   "粗",
+   "繪",
+   "魯"
+  ],
+  "in_lexicon": [
+   "0",
+   "1"
+  ]
+ }
+}