langdrift取证与调优: MISS日志附被拒译文样本,ja假名阈值4%→2%+至少2假名(治汉字密集合法日语误杀),langdrift重试收敛为1次(0.7高温无益)

This commit is contained in:
liwenhao2 2026-09-15 11:21:07 +08:00
parent 8abad1eb78
commit 4ce45cb4b9
1 changed files with 9 additions and 4 deletions

View File

@ -1964,7 +1964,9 @@ class MainActivity : AppCompatActivity() {
isEcho(s2.text, tr, echoJaccardTh(s2.text)) -> isEcho(s2.text, tr, echoJaccardTh(s2.text)) ->
Log.i(TAG, "MISS sid=" + s2.id + " reason=seg-echo") Log.i(TAG, "MISS sid=" + s2.id + " reason=seg-echo")
!targetLangOk(tr) -> !targetLangOk(tr) ->
Log.i(TAG, "MISS sid=" + s2.id + " reason=seg-drift") Log.i(TAG, "MISS sid=" + s2.id + " reason=seg-drift" +
" sample=" + tr.take(24).replace("
", " "))
else -> if (s2.trans != tr) { else -> if (s2.trans != tr) {
s2.trans = tr s2.trans = tr
s2.line?.let { touched.add(it) } s2.line?.let { touched.add(it) }
@ -2073,8 +2075,10 @@ class MainActivity : AppCompatActivity() {
Log.i(TAG, "MISS sid=" + sid + " reason=echo") Log.i(TAG, "MISS sid=" + sid + " reason=echo")
!targetLangOk(tr) -> { !targetLangOk(tr) -> {
val att = bumpAttempt(sid) val att = bumpAttempt(sid)
Log.i(TAG, "MISS sid=" + sid + " reason=langdrift attempt=" + att) Log.i(TAG, "MISS sid=" + sid + " reason=langdrift attempt=" +
if (att <= 2) drift.add(Pair(sid, src)) att + " sample=" + tr.take(24).replace("
", " "))
if (att <= 1) drift.add(Pair(sid, src))
} }
else -> good[sid] = tr else -> good[sid] = tr
} }
@ -2277,8 +2281,9 @@ class MainActivity : AppCompatActivity() {
if (tr.length < 12) return true if (tr.length < 12) return true
return when (uiLanguage()) { return when (uiLanguage()) {
"ja" -> { "ja" -> {
val kana = tr.count { it.code in 0x3040..0x30FF }
val latin = tr.count { it in 'a'..'z' || it in 'A'..'Z' } val latin = tr.count { it in 'a'..'z' || it in 'A'..'Z' }
tr.count { it.code in 0x3040..0x30FF } * 25 >= tr.length && kana >= 2 && kana * 50 >= tr.length &&
latin * 3 <= tr.length latin * 3 <= tr.length
} }
"zh" -> { "zh" -> {