{"id":2645,"date":"2022-06-28T18:27:07","date_gmt":"2022-06-28T10:27:07","guid":{"rendered":"https:\/\/www.csslayer.info\/wordpress\/?p=2645"},"modified":"2022-06-30T00:26:04","modified_gmt":"2022-06-29T16:26:04","slug":"libime-intro-1","status":"publish","type":"post","link":"https:\/\/www.csslayer.info\/wordpress\/fcitx-dev\/libime-intro-1\/","title":{"rendered":"libime \u539f\u7406\u4ecb\u7ecd\uff08\u4e00\uff09"},"content":{"rendered":"\n<p><strong>2022\/06\/29\uff1a\u611f\u8c22\u4e00\u4e9b\u670b\u53cb\u7684\u6307\u6b63\uff0c\u66f4\u65b0\u4e86\u5173\u4e8e rime \u7684\u4e00\u70b9\u4fe1\u606f\u3002<\/strong><\/p>\n\n\n\n<p>\u5199\u5728\u5f00\u5934\u7684\u8bdd\uff1a\u8bb8\u591a\u5e74\u524d\uff0c\u5728\u6211\u624d\u521a\u521a\u5f00\u59cb\u7528 linux \u7684\u65f6\u5019\uff0c\u521a\u597d\u8d76\u4e0a\u4e86 Sunpinyin \u7684\u8bde\u751f\u3002\u90a3\u4e2a\u65f6\u5019 Yong Sun \u5199\u4e86 Sunpinyin \u4ee3\u7801\u5bfc\u8bfb\u3002\u5c3d\u7ba1\u90a3\u4e2a\u65f6\u5019\u56e0\u4e3a\u77e5\u8bc6\u7684\u50a8\u5907\u4e0d\u591f\u4e30\u5bcc\uff0c\u57fa\u672c\u4e0a\u6ca1\u6709\u80fd\u591f\u7406\u89e3\u6587\u7ae0\u7684\u5185\u5bb9\uff0c\u4f46\u662f\u73b0\u5728\u56de\u8fc7\u5934\u6765\u7684\u8bdd\uff0c\u7ec8\u4e8e\u5728 2017 \u5e74\u7684\u65f6\u5019\uff0c\u81ea\u5df1\u4ece\u5934\u5199\u4e86\u4e00\u4e2a\u7c7b\u4f3c\u7b97\u6cd5\u7684\u8f93\u5165\u6cd5\u5f15\u64ce\uff0c\u6765\u4ee3\u66ff\u65e9\u5df2\u8fc7\u65f6\u7684 Fcitx 4 \u4e2d\u7684\u9ed8\u8ba4\u62fc\u97f3\u548c\u7801\u8868\u3002<\/p>\n\n\n\n<p>\u867d\u7136\u4e5f\u8bb8\u6709\u4eba\u4f1a\u6709\u7591\u95ee\uff0c\u4e3a\u4ec0\u4e48\u8981\u4ece\u5934\u5199\u8fd9\u6837\u4e00\u4e2a\u8f93\u5165\u6cd5\u5f15\u64ce\uff0c\u96be\u9053 Sunpinyin \u4e0d\u591f\u597d\u5417\uff1f\u6211\u5f88\u4e45\u4e4b\u524d\u7684\u4e00\u7bc7\u535a\u5ba2\u5176\u5b9e\u5177\u4f53\u505a\u51fa\u4e86\u4e00\u4e9b<a href=\"https:\/\/www.csslayer.info\/wordpress\/fcitx-dev\/a-new-pinyin-input-method-that-might-be-slightly-better-than-sunpinyin\/\" data-type=\"URL\" data-id=\"https:\/\/www.csslayer.info\/wordpress\/fcitx-dev\/a-new-pinyin-input-method-that-might-be-slightly-better-than-sunpinyin\/\">\u529f\u80fd\u4e0a\u7684\u89e3\u91ca<\/a>\u3002\u8fd9\u91cc\u6211\u4eec\u518d\u91cd\u65b0\u4ece\u4ee3\u7801\u89d2\u5ea6\u6765\u89e3\u91ca\u4e00\u4e0b libime \u7684\u6765\u9f99\u53bb\u8109\u3002<\/p>\n\n\n\n<p>libime \u76ee\u524d\u7684\u5e93\u5206\u4e3a 4 \u4e2a\u90e8\u5206\uff0c\u5176\u4e2d\u7ca4\u62fc\u5355\u72ec\u5728\u4e00\u4e2aGit \u4ee3\u7801\u4ed3\u5e93\u91cc\uff0c\u7ca4\u62fc\u5927\u90e8\u5206\u7684\u4ee3\u7801\u662f\u548c\u62fc\u97f3\u7c7b\u4f3c\u7684\u3002\u4e4b\u540e\u4e5f\u4e0d\u4f1a\u7279\u522b\u4ecb\u7ecd\u7ca4\u62fc\u7684\u90e8\u5206\u3002<\/p>\n\n\n\n<p>\u53e6\u5916\u4e00\u4e9b\u539f\u56e0\u662f\uff0c\u5176\u4e2d\u5f88\u591a\u6838\u5fc3\u7684\u4ee3\u7801\u5bf9\u6211\u81ea\u5df1\u6765\u8bf4\u4e5f\u5df2\u7ecf\u4e0d\u5f88\u719f\u6089\u4e86\uff0c\u4e5f\u662f\u8d81\u6b64\u673a\u4f1a\u8bb0\u5f55\u4e00\u4e0b\u7ffb\u65b0\u4e00\u4e0b\u81ea\u5df1\u7684\u8bb0\u5fc6\u3002\u53e6\u5916\u6211\u76f8\u4fe1\u8fd9\u4e9b\u7ecf\u5386\u7684\u90e8\u5206\uff0c\u5bf9\u4e8e\u4e86\u89e3 Linux \u8f93\u5165\u6cd5\u548c Fcitx \u7684\u53d1\u5c55\u8fc7\u7a0b\u7684\u4e5f\u662f\u5f88\u6709\u8da3\u7684\uff0c\u5f53\u6210\u4e00\u4e2a\u6545\u4e8b\u6765\u770b\u5927\u6982\u4e5f\u4e0d\u9519\uff1f<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u6838\u5fc3\u90e8\u5206\uff08Core\uff09<\/h2>\n\n\n\n<p>\u5728\u89e3\u8bf4 Core \u4e4b\u524d\uff0c\u5c31\u8981\u8bf4\u5230\u4e3a\u4ec0\u4e48\u8981\u5199 libime \u8fd9\u4ef6\u4e8b\u3002\u800c\u8fd9\u5c31\u4e0d\u5f97\u4e0d\u8bf4\u8bf4\u4e4b\u524d\u7684\u51e0\u4e2a\u524d\u8f88\u4eec\uff0c\u6709\u4e9b\u6211\u53ea\u6709\u5927\u81f4\u7684\u4e86\u89e3\uff0c\u6240\u4ee5\u53ef\u80fd\u6709\u9519\u8bef\u3002\u9996\u5148\u662f fcitx4 \u7684\u62fc\u97f3\u8f93\u5165\uff0c\u5b83\u7684\u57fa\u672c\u539f\u7406\u5c31\u662f Forward maximum matching\uff0c\u7b80\u5355\u6765\u8bf4\u5c31\u662f\u6839\u636e\u4f60\u5f53\u524d\u7684\u8f93\u5165\uff0c\u627e\u5230\u8bcd\u5e93\u4e2d\u6700\u957f\u80fd\u5339\u914d\u5b83\u7684\u5185\u5bb9\u3002\u8fd9\u4e2a\u7b97\u6cd5\u672c\u8eab\u5728\u6c49\u5b57\u5206\u8bcd\u7684\u9886\u57df\u4e5f\u6709\u5e94\u7528\uff0c\u4f18\u70b9\u5f53\u7136\u5c31\u662f\u5feb\uff0c\u7f3a\u70b9\u4e5f\u5f88\u660e\u663e\uff0c\u5c31\u662f\u7ed9\u51fa\u7684\u63d0\u793a\u4e0d\u591f\u51c6\u786e\uff0c\u6ca1\u6709\u53e5\u5b50\u7684\u6982\u5ff5\u3002\u800c Sunpinyin \uff0c\u662f\u4e00\u4e2a\u57fa\u4e8eN-gram\u7edf\u8ba1\u8bed\u8a00\u6a21\u578b\u7684\u8f93\u5165\u6cd5\u3002\u5b83\u6839\u636e\u62fc\u97f3\u5339\u914d\u5230\u7684\u8bcd\u6765\u8ba1\u7b97\u4e00\u4e2a\u6700\u7ec8\u53e5\u5b50\u7684\u6982\u7387\uff0c\u4ece\u4e2d\u5c06\u7ed3\u679c\u6309\u7167\u6982\u7387\u9ad8\u4f4e\u8fdb\u884c\u6392\u5e8f\u3002<\/p>\n\n\n\n<p>\uff08\u66f4\u6b63\u4e00\u4e0b\u5173\u4e8e Rime \u7684\u4fe1\u606f\uff09 Rime \u51e0\u5e74\u524d\u786e\u5b9e\u6ca1\u6709\u8bed\u8a00\u6a21\u578b\uff0c\u6216\u8005\u8bf4\u53ea\u662f\u7c7b\u4f3c Unigarm \u7684\u7b97\u6cd5 \uff0c<a href=\"https:\/\/github.com\/rime\/librime\/commit\/9248a6bc78b1ef772255d17077da19783daa52aa\" data-type=\"URL\" data-id=\"https:\/\/github.com\/rime\/librime\/commit\/9248a6bc78b1ef772255d17077da19783daa52aa\">2019 \u5e74\u5de6\u53f3\u52a0\u5165\u4e86 \u4e00\u4e2a bigarm \u7684\u5b9e\u73b0<\/a>\u3002\u7136\u800c\u6839\u636e\u641c\u7d22\u5bf9\u5e94\u7684\u6587\u4ef6\u540d\uff0c\u4f3c\u4e4e\u6ca1\u6709\u54ea\u4e2a Linux \u53d1\u884c\u7248\u6253\u5305\u4e86\u5bf9\u5e94\u7684\u6570\u636e\u2026\u6240\u4ee5\u53ef\u80fd\u6839\u672c\u4e5f\u6ca1\u4ec0\u4e48\u4eba\u5b9e\u9645\u7528\u4e0a\u4e86 rime \u7684\u8fd9\u4e2a\u529f\u80fd\u3002<\/p>\n\n\n\n<p>\u4f46 Sunpinyin \u4f5c\u4e3a\u5b9e\u73b0\u8f93\u5165\u6cd5\u6765\u8bf4\uff0c\u7f3a\u5c11\u5f88\u591a\u4e30\u5bcc\u529f\u80fd\u65b9\u9762\u7684 API\uff0c\u4f8b\u5982\u8bcd\u5e93\u53ea\u6709\u4e00\u4e2a\u3002\u4e5f\u6ca1\u6709\u7279\u522b\u65b9\u4fbf\u7684\u4fee\u6539\u8bcd\u5e93\u7684\u65b9\u5f0f\u3002\u53e6\u5916\u8bcd\u5e93\u672c\u8eab\u8fd8\u662f\u91c7\u7528 sqlite \u8fdb\u884c\u5b58\u50a8\uff0c\u5bf9\u8f93\u5165\u6cd5\u6765\u8bf4\u5e76\u6ca1\u6709\u5f88\u5927\u7684\u5fc5\u8981\u3002<\/p>\n\n\n\n<p>\u56e0\u6b64\uff0clibime \u7684\u6838\u5fc3\u76ee\u6807\u5c31\u662f\uff0c\u5b9e\u73b0\u4e00\u4e2a\u7c7b\u4f3c Sunpinyin \u7684\u57fa\u4e8e N-gram \u7684\u8f93\u5165\u6cd5\u5f15\u64ce\uff0c\u5e76\u4e14\u9700\u8981\u53ef\u4ee5\u65b9\u4fbf\u5730\u652f\u6301\u6211\u9700\u8981\u7684\u529f\u80fd\uff0c\u5305\u62ec\u591a\u8bcd\u5e93\u7b49\u529f\u80fd\u3002<\/p>\n\n\n\n<p>\u4f5c\u4e3a Fcitx 4 \u7684\u7cbe\u795e\u7eed\u4f5c\uff0c\u6211\u8fd8\u5e0c\u671b\u5b83\u80fd\u591f\u5360\u7528\u5c3d\u91cf\u5c11\u7684\u8d44\u6e90\u3002<\/p>\n\n\n\n<p>\u4f5c\u4e3a\u6838\u5fc3\u90e8\u5206\uff0c\u6211\u4e3b\u8981\u60f3\u8981\u5305\u62ec\u7684\u6570\u636e\u7ed3\u6784\u5305\u62ec\u4e24\u4e2a\uff0c<a href=\"https:\/\/zh.wikipedia.org\/wiki\/Trie\">Trie<\/a> \u548c\u8bed\u8a00\u6a21\u578b\u3002\u5982\u679c\u67e5\u770b\u65e9\u671f\u7684\u5386\u53f2\u7684\u8bdd\uff0c\u53ef\u4ee5\u770b\u5230\u597d\u51e0\u79cd\u5df2\u7ecf\u5220\u6389\u7684 Trie \u7684\u4ee3\u7801\u7684\u5b9e\u73b0\u3002\u5982\u679c\u5b66\u4e60\u8fc7\u6570\u636e\u7ed3\u6784\u4ee3\u7801\u7684\u8bdd\uff0c\u53ef\u80fd\u4f1a\u77e5\u9053 Trie \u662f\u4e00\u4e2a\u8868\u793a\u6811\u7684\u6570\u636e\u7ed3\u6784\uff0c\u9002\u5408\u5b58\u50a8\u8bb8\u591a\u5b57\u7b26\u4e32\u3002\u5f53\u4f60\u5728 trie \u4e0a\u6e38\u8d70\u7684\u65f6\u5019\uff0c\u6839\u636e\u5b57\u7b26\u4e32\u5f53\u524d\u5b57\u6bcd\u7684\u4f4d\u7f6e\u5c31\u53ef\u4ee5\u627e\u5230\u5176\u4ed6\u5339\u914d\u8fd9\u4e2a\u524d\u7f00\u7684\u8282\u70b9\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"500\" height=\"469\" src=\"https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247.png\" alt=\"\" class=\"wp-image-2646\" srcset=\"https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247.png 500w, https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-300x281.png 300w\" sizes=\"auto, (max-width: 500px) 100vw, 500px\" \/><\/figure>\n\n\n\n<p>\u5176\u4e2d\u6211\u5bf9\u4e8e\u5b83\u7684\u6570\u636e\u7ed3\u6784\u529f\u80fd\u7684\u8981\u6c42\u4e3b\u8981\u662f\u8fd9\u6837\u7684\uff1a<\/p>\n\n\n\n<p>1\u3001\u53ef\u4ee5\u5728\u8fd0\u884c\u65f6\u4fee\u6539\uff08\u6709\u4e00\u4e9b Trie \u7684\u5b9e\u73b0\u662f\u65e0\u6cd5\u52a8\u6001\u4fee\u6539\u7684\uff09<br>2\u3001\u5185\u5b58\u5360\u7528\u7d27\u51d1<\/p>\n\n\n\n<p>\u5728\u5c1d\u8bd5\u4e86\u591a\u4e2a\u5b9e\u73b0\u4e4b\u540e\uff0c\u6700\u540e\u91c7\u7528\u4e86 double array trie \u4f5c\u4e3a Trie \u7684\u5b9e\u73b0\uff0c\u91c7\u7528\u4e86 <a href=\"http:\/\/www.tkl.iis.u-tokyo.ac.jp\/~ynaga\/cedar\/\" data-type=\"URL\" data-id=\"http:\/\/www.tkl.iis.u-tokyo.ac.jp\/~ynaga\/cedar\/\">cedar<\/a> \u7684\u4ee3\u7801\u4f5c\u4e3a\u5b9e\u73b0\uff0c\u4f46\u628a\u5176\u4e2d\u53ef\u4ee5\u7528 C++  \u6807\u51c6\u5e93\u66ff\u4ee3\u7684\u90e8\u5206\u90fd\u8fdb\u884c\u4e86\u66ff\u4ee3\u4ece\u800c\u7cbe\u7b80\u4e86\u8bb8\u591a\u4e0d\u9700\u8981\u7684\u4ee3\u7801\u3002\u8fd9\u5176\u4e2d\u8fd8\u8e29\u4e86\u4e00\u4e2a\u5751\uff0c\u7b80\u5355\u6765\u8bf4\u5c31\u662f std::vector \u66ff\u4ee3\u4e4b\u540e\u7684\u6027\u80fd\u4e0d\u5982\u539f\u672c\u7684\u6027\u80fd\uff0c\u6700\u7ec8\u53d1\u73b0\u539f\u56e0\u662f C++ \u7684 allocator \u6ca1\u6709\u80fd\u5bf9\u5e94 realloc \u7684\u51fd\u6570\u3002\u5728\u53ef\u4ee5\u539f\u5730 realloc \u7684\u65f6\u5019\uff0c\u53ef\u4ee5\u8282\u7701\u5f88\u591a\u64cd\u4f5c\u3002\u56e0\u6b64\u53c8\u5b9e\u73b0\u4e86\u4e00\u4e2a naivevector \uff0c\u628a std::vector \u91cc\u9762\u5b9e\u73b0\u66ff\u6362\u6210\u91c7\u7528 malloc \u548c realloc \u7684\uff0c\u4ece\u800c\u83b7\u5f97\u4e86\u548c\u539f\u7248\u7c7b\u4f3c\u7684\u6027\u80fd\u3002<\/p>\n\n\n\n<p>\u800c\u8bed\u8a00\u6a21\u578b\u6211\u5219\u627e\u5230\u4e86 <a href=\"https:\/\/kheafield.com\/code\/kenlm\/\" data-type=\"URL\" data-id=\"https:\/\/kheafield.com\/code\/kenlm\/\">kenlm<\/a> \u4f5c\u4e3a\u5b9e\u73b0\uff0c\u5728\u5176\u4e0a\u5c01\u88c5\u4e86\u5bf9\u5e94\u7684 API \u65b9\u4fbf\u8c03\u7528\u3002\u5bf9\u5e94\u7684\u5c31\u5c01\u88c5\u5728 LanguageModel  \u7c7b\u4e2d\u3002kenlm \u63d0\u4f9b\u4e86\u591a\u79cd\u5b58\u50a8\u65b9\u5f0f\u8fd9\u91cc libime \u76f4\u63a5\u91c7\u7528\u4e86\u4e00\u79cd\u5185\u5b58\u5360\u7528\u8f83\u5c0f\u7684\u9009\u9879\u3002<\/p>\n\n\n\n<p>\u5230\u6b64 Core \u4e2d\u4e24\u4e2a\u6838\u5fc3\u6570\u636e\u7ed3\u6784\u7684\u95ee\u9898\u5c31\u90fd\u89e3\u51b3\u4e86\u3002<\/p>\n\n\n\n<p>kenlm \u63d0\u4f9b\u7684\u6838\u5fc3\u529f\u80fd\u5c31\u662f\u7ed9\u51fa\u4e00\u4e2a\u5f53\u524d\u72b6\u6001\uff0c\u67e5\u8be2\u5bf9\u5e94\u7684\u8bcd\u7684\u6982\u7387\u3002<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">    LanguageModel model(LIBIME_BINARY_DIR \"\/data\/sc.lm\");\n    State state(model.nullState()), out_state(model.nullState());\n    std::string word;\n    float sum = 0.0f;\n    while (std::cin &gt;&gt; word) {\n        float s;\n        WordNode w(word, model.index(word));\n        std::cout &lt;&lt; w.idx() &lt;&lt; \" \" &lt;&lt; (s = model.score(state, w, out_state))\n                  &lt;&lt; '\\n';\n        std::cout &lt;&lt; \"Prob\" &lt;&lt; std::pow(10, s) &lt;&lt; '\\n';\n        state = out_state;\n        sum += s;\n    }<\/pre>\n\n\n\n<p>State \u4fdd\u5b58\u4e86\u5f53\u524d\u7684\u72b6\u6001\u4fe1\u606f\uff0c\u672c\u8d28\u4e0a\u5c31\u662f\u5b58\u50a8\u4e4b\u524d\u51fa\u73b0\u7684 WordNode \u6709\u54ea\u4e9b\uff0c\u4ece\u800c\u8ba1\u7b97\u5bf9\u5e94\u7684\u4e0b\u4e00\u4e2a\u8bcd\u51fa\u73b0\u7684\u6982\u7387\u3002\u6982\u7387\u7684\u6570\u503c\u90fd\u91c7\u7528\u5bf9\u6570\u3002\u56e0\u6b64\u53ea\u9700\u8981\u5c06\u83b7\u5f97\u7684\u6982\u7387\u7b80\u5355\u76f8\u52a0\uff0c\u5373\u53ef\u5f97\u5230\u6574\u4f53\u7684\u6982\u7387\u3002<\/p>\n\n\n\n<p>\u5728\u6709\u4e86 Trie \u548c\u8bed\u8a00\u6a21\u578b\u4e4b\u540e\uff0cCore \u6700\u540e\u8981\u63d0\u4f9b\u7684\u6838\u5fc3\u529f\u80fd\u5c31\u662f\u5229\u7528 Trie \u548c\u6a21\u578b\u8fdb\u884c\u641c\u7d22\uff0c\u6765\u83b7\u5f97\u6700\u7ec8\u7684\u5339\u914d\u7684\u53e5\u5b50\u7684\u7ed3\u679c\uff0c\u8fd9\u4e2a\u8fc7\u7a0b\u5728 NLP \u4e2d\u4e00\u822c\u79f0\u505a decode\u3002\u5bf9\u5e94\u7684\u5b9e\u73b0\u5728 libime \u91cc\u5c31\u653e\u5728\u4e86 Decoder \u91cc\u3002\u8fd9\u91cc\u91c7\u7528\u7684\u7b97\u6cd5\u5c31\u662f <a href=\"https:\/\/zh.wikipedia.org\/wiki\/%E7%BB%B4%E7%89%B9%E6%AF%94%E7%AE%97%E6%B3%95\" data-type=\"URL\" data-id=\"https:\/\/zh.wikipedia.org\/wiki\/%E7%BB%B4%E7%89%B9%E6%AF%94%E7%AE%97%E6%B3%95\">Viterbi<\/a>\u3002\u5f53\u7136\uff0c\u4e3a\u4e86\u7b80\u5316\u505a\u4e86\u5bf9\u5e94\u7684\u526a\u679d\uff0c\u6574\u4f53\u7684\u7b97\u6cd5\u4e00\u822c\u88ab\u79f0\u505a Beam search\u3002\uff08\u975eNLP \u4e13\u4e1a\uff0c\u7528\u8bcd\u53ef\u80fd\u6709\u4e0d\u51c6\u786e\u7684\u5730\u65b9\uff09<\/p>\n\n\n\n<p>\u800c\u5728\u6b64\u4e4b\u524d\uff0c\u6211\u4eec\u5148\u56de\u5230\u6211\u4eec\u9700\u8981\u5b9e\u73b0\u7684\u8f93\u5165\u6cd5\u4e0a\u6765\uff0c\u8fd9\u6837\u624d\u80fd\u66f4\u597d\u7684\u8bb2\u89e3\u8fd9\u91cc\u7b97\u6cd5\u662f\u600e\u4e48\u8d77\u4f5c\u7528\u7684\u3002<\/p>\n\n\n\n<p>\u9996\u5148\u6211\u4eec\u8981\u63d0\u5230\u7684\u5c31\u662f\u8fd9\u4e2a\u7c7b\uff0cSegmentGraph\u3002\u5b83\u8868\u793a\u7684\u662f\u5c06\u5b57\u7b26\u4e32\u5207\u5206\u7684\u65b9\u5f0f\u7684\u56fe\u3002\u5176\u4e2d\u6bcf\u4e2a\u53ef\u4ee5\u5207\u5206\u7684\u8282\u70b9\u90fd\u662f\u56fe\u4e2d\u7684\u4e00\u4e2a\u8282\u70b9\u3002\u5bf9\u5e94\u5230\u62fc\u97f3\u5f53\u4e2d\uff0c\u5c31\u662f\u62fc\u97f3\u8fdb\u884c\u5207\u5206\u7684\u65b9\u5f0f\u3002\u5207\u5206\u7684\u65b9\u5f0f\u662f\u5b8c\u5168\u7684\u3002\u5bf9\u4e8e\u4e00\u4e2a\u6700\u590d\u6742\u7684\u60c5\u51b5\u6765\u8bf4\uff0c\u4f8b\u5982\uff08anana\uff09\uff0c\u53ef\u80fd\u7684\u5207\u5206\u5c31\u662f\u8fd9\u6837\u7684\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"960\" height=\"201\" src=\"https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-1.png\" alt=\"\" class=\"wp-image-2647\" srcset=\"https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-1.png 960w, https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-1-300x63.png 300w, https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-1-768x161.png 768w\" sizes=\"auto, (max-width: 960px) 100vw, 960px\" \/><\/figure>\n\n\n\n<p>\u5bf9\u5e94\u8868\u793a\u4e86 a na na\uff0can a na\uff0can an a \u4e09\u79cd\u5207\u5206\u65b9\u5f0f\u3002<\/p>\n\n\n\n<p>decoder \u7684\u8f93\u5165\u662f\u4e00\u4e2a Segment Graph\uff0c\u800c\u8f93\u51fa\u5219\u662f\u4e00\u4e2a Lattice\uff08\u7f51\u683c\uff09\u3002\u8fd9\u91cc\u53d1\u751f\u4e86\u4ec0\u4e48\u5462\uff1f\u7b80\u5355\u6765\u8bf4\uff0c\u5c31\u662f\u628a Segment Graph \u7684\u4e0a\u6bcf\u4e00\u6761\u8def\u5f84\uff0c\u90fd\u653e\u5230\u8bcd\u5178\u4e0a\u53bb\u5339\u914d\u3002\u5177\u4f53\u91c7\u7528\u7684\u5339\u914d\u65b9\u5f0f\u7531\u8bcd\u5178\u51b3\u5b9a\u3002\u4f8b\u5982\uff0c\u5bf9\u4e8e\u4e0a\u9762\u7684\u4f8b\u5b50\uff0ca na na \u8fd9\u79cd\u5207\u5206\uff0c\u5c31\u8981\u5339\u914d a na na\uff0ca na\uff0ca\uff0cna na\uff0cna \u51e0\u79cd\u4e0d\u540c\u7684\u8def\u5f84\u3002\u9488\u5bf9\u8fd9\u79cd\u8def\u5f84\u6784\u5efa\u51fa\u6765\u7684 Lattice \u5c31\u7c7b\u4f3c\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"960\" height=\"301\" src=\"https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-2.png\" alt=\"\" class=\"wp-image-2648\" srcset=\"https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-2.png 960w, https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-2-300x94.png 300w, https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-2-768x241.png 768w\" sizes=\"auto, (max-width: 960px) 100vw, 960px\" \/><figcaption>\u5f53\u7136\uff0c\u8fd9\u4ec5\u4ec5\u662f\u9488\u5bf9 a na na \u8fd9\u4e00\u79cd\u5207\u5206\uff0c\u5176\u4ed6\u7684\u5207\u5206\u8fd8\u4f1a\u5728\u4e0a\u9762\u52a0\u4e0a\u4e0d\u540c\u7684\u8fb9\u3002\u800c\u4e0a\u56fe\u7684\u6bcf\u6761\u8fb9\u4e0a\u6807\u6ce8\u7684\u62fc\u97f3\uff0c\u5728\u4f7f\u7528 Pinyin \u8bcd\u5178\u8fdb\u884c decoder \u65f6\u5c31\u4f1a\u53bb\u8bcd\u5178\u4e2d\u8fdb\u884c\u5339\u914d\u3002\u6bcf\u4e00\u6761\u8fb9\u90fd\u5c06\u5bf9\u5e94\u7684 SegmentGraph \u4e2d\u7684 SegmentGraphNode\uff0c\u6269\u5c55\u4e3a\u5bf9\u5e94\u7684\u8bcd\u7684 LatticeNode\u3002<\/figcaption><\/figure>\n\n\n\n<p>\u8fd9\u91cc\u6211\u4eec\u7a0d\u5fae\u6362\u4e00\u4e2a\u522b\u7684\u4f8b\u5b50\u6765\u8fdb\u884c\u89e3\u91ca\u3002\u4f8b\u5982\u62fc\u97f3\u5185\u5207\u5206\u7684\u7ecf\u5178\u4f8b\u5b50xian\uff08xi an\uff09\u3002\u5728 xian \u8fd9\u4e2a\u524d\u7f00\u6240\u5bf9\u5e94\u7684\u8282\u70b9\uff0c\u6211\u4eec\u5c31\u6709\u901a\u8fc7 xian \u4f5c\u4e3a\u6574\u4f53\u5339\u914d\u7684\u300c\u5148\uff0c\u7ebf\uff0c\u53bf\uff0c\u73b0\u2026\u2026\u300d\u3002\u800c\u5728 xi \u8fd9\u4e2a\u524d\u7f00\u8282\u70b9\uff0c\u5219\u53ea\u6709 xi\uff08\u897f\uff0c\u559c\uff0c\u7cfb\uff0c\u6d17\u2026\u2026\uff09\uff0c\u7136\u540e\u8fd9\u4e9b xi \u7684\u8282\u70b9\u53c8\u901a\u8fc7 an \u53ef\u4ee5\u5230\u8fbe xian \u8fd9\u4e2a SegmentGraphNode\u3002\u4f8b\u5982\u6709 an \u300c\u4ffa\uff0c\u6309\uff0c\u5b89\uff0c\u2026\u2026\uff09\u3002\u6700\u540e\uff0c\u8fd8\u6709\u76f4\u63a5\u901a\u8fc7 xi an \u4f5c\u4e3a\u6574\u8bcd\u5230\u8fbe xian \u7684\uff08\u897f\u5b89\uff09\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"960\" height=\"720\" src=\"https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-3.png\" alt=\"\" class=\"wp-image-2649\" srcset=\"https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-3.png 960w, https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-3-300x225.png 300w, https:\/\/www.csslayer.info\/wordpress\/wp-content\/uploads\/2022\/06\/\u56fe\u7247-3-768x576.png 768w\" sizes=\"auto, (max-width: 960px) 100vw, 960px\" \/><figcaption>\u5c06 SegmentGraph \u7ecf\u8fc7\u8bcd\u5178\u5339\u914d\u8f6c\u6362\u6210 Lattice \u4e4b\u540e\u7684\u56fe\u5c31\u5982\u56fe\u6240\u793a\u3002\u5f53\u7136\uff0c\u56e0\u4e3a\u5b83\u5728 Segment Graph \u4e0a\u662f\u6709\u8054\u7cfb\u7684\uff0c\u5e76\u4e0d\u9700\u8981\u628a\u5176\u4e2d\u7684\u6bcf\u4e00\u6761\u8fb9\u90fd\u5728\u5185\u5b58\u5f53\u4e2d\u76f4\u63a5\u5c55\u793a\u51fa\u6765\u3002\u53ea\u9700\u8981\u7ba1\u7406\u597d Lattice Node \u548c Segment Graph Node \u4e4b\u95f4\u7684\u6620\u5c04\u5173\u7cfb\u5373\u53ef\u3002<\/figcaption><\/figure>\n\n\n\n<p>\u5728\u5efa\u7acb\u597d Lattice \u4e4b\u540e\uff0c\u5c31\u662f\u7ecf\u5178\u7684 Beam search \u53d1\u6325\u4f5c\u7528\u7684\u5730\u65b9\u4e86\u3002\u901a\u8fc7\u52a8\u6001\u89c4\u5212\u7684\u65b9\u5f0f\u8ba1\u7b97\u6bcf\u6761\u53ef\u80fd\u8def\u5f84\u7684\u6982\u7387\uff08\u6216\u79f0\u4f5cscore\uff09\uff0c\u5728\u67d0\u4e9b\u60c5\u51b5\u4e0b\u4f1a\u51cf\u5c11\u53ef\u80fd\u7684\u641c\u7d22\u7ed3\u679c\uff0c\u5426\u5219\u53ef\u80fd\u641c\u7d22\u7a7a\u95f4\u5de8\u5927\u3002<\/p>\n\n\n\n<p>\u8fd9\u91cc\u7684 Beam search \u6574\u4f53\u6765\u8bf4\u5206\u6210\u4e09\u4e2a\u6b65\u9aa4\uff0c\u7b2c\u4e00\u6b65\u5c31\u662f\u4e0a\u9762\u8bf4\u7684\u6784\u5efa Lattice\uff08\u7f51\u683c\uff09\uff0c\u7b2c\u4e8c\u6b65 forward search\uff0c\u7b2c\u4e09\u6b65 backward search\u3002<\/p>\n\n\n\n<p>\u7b2c\u4e8c\u6b65 forward search \u7684\u57fa\u672c\u6b65\u9aa4\u5c31\u662f\uff0c\u5e7f\u5ea6\u4f18\u5148\u904d\u5386 Segment Graph \u4e0a\u7684\u6240\u6709\u8282\u70b9\uff0c\u6216\u8005\u6362\u53e5\u8bdd\u8bf4\uff0c\u5c31\u662f\u904d\u5386\u6240\u6709\u53ef\u80fd\u7684\u5207\u5206\u70b9\uff08\u4e0d\u8bba\u5207\u5206\u8def\u5f84\u4e3a\u4f55\uff09\uff0c\u7136\u540e\u8ba1\u7b97\u5207\u5206\u70b9 Segment Graph Node \u5bf9\u5e94\u7684 Lattice Node \u7684\u6700\u9ad8\u5f97\u5206\uff0c\u8ba1\u7b97\u65b9\u6cd5\u5c31\u662f\u627e\u5230 Lattice Node \u5bf9\u5e94\u7684 Segment Graph Node \u7684\u7236 Segment Graph Node\u3002\u8fd9\u4e2a\u8282\u70b9\u662f\u552f\u4e00\u7684\uff0c\u56e0\u4e3a Lattice Node \u662f\u901a\u8fc7\u8ba1\u7b97\u67d0\u6bb5\u62fc\u97f3\u5339\u914d\u7684\u8bcd\u627e\u5230\u7684\u3002\u7136\u540e\u9488\u5bf9\u8fd9\u4e2a\u7236 Segment Graph Node \u5185\u7684 Lattice Node \u8fdb\u884c\u904d\u5386\uff0c\u8ba1\u7b97\u5176\u4e2d\u7684\u6700\u9ad8\u5206\u3002\u7b80\u5355\u6765\u8bf4\u5c31\u662f latticeNode.score = latticeNodeFrom.score + languageModel.score(latticeNodeFrom.state, latticeNode.word) \u3002\u8fd9\u6837\u4e0d\u65ad\u6c42\u89e3\uff0c\u76f4\u5230\u6c42\u51fa Lattice End Node \u7684\u5f97\u5206\u3002<\/p>\n\n\n\n<p>\u800c Backward Search \u7684\u4e3b\u8981\u76ee\u7684\u662f\u8ba1\u7b97 N-Best score\uff0c\u5982\u679c\u53ea\u9700\u8981\u4e00\u4e2a\u7ed3\u679c\u7684\u8bdd\uff0c\u90a3\u663e\u800c\u6613\u89c1\u5c31\u662f Lattice End Node \u7684\u6700\u9ad8\u5206\u5bf9\u5e94\u8def\u5f84\u3002\u4f46\u600e\u4e48\u6c42\u89e3\u6b21\u4e00\u7ea7\u7684\u5f97\u5206\u7684\u8def\u5f84\u5462\uff1f<\/p>\n\n\n\n<p>\u7b80\u5355\u6765\u8bf4\u5c31\u662f\u4ece\u7ed3\u5c3e\u5012\u63a8\u56de\u53bb\u3002\u6700\u9ad8\u5206\u7684 Lattice Node \u7684\u7236 Segment Graph Node \u7684\u5176\u4ed6\u5019\u9009\u66ff\u6362\uff0c\u6765\u5bf9\u6bd4\u83b7\u5f97\u66f4\u597d\u7684\u7ed3\u679c\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>2022\/06\/29\uff1a\u611f\u8c22\u4e00\u4e9b\u670b\u53cb\u7684\u6307\u6b63\uff0c\u66f4\u65b0\u4e86\u5173\u4e8e rime \u7684\u4e00\u70b9\u4fe1\u606f\u3002 \u5199\u5728\u5f00\u5934\u7684\u8bdd\uff1a\u8bb8\u591a\u5e74\u524d\uff0c\u5728\u6211\u624d\u521a\u521a\u5f00\u59cb\u7528 linux \u7684\u65f6\u5019\uff0c\u521a\u597d\u8d76\u4e0a\u4e86 Sunpinyin \u7684\u8bde\u751f\u3002\u90a3\u4e2a\u65f6\u5019 Yong Sun \u5199\u4e86 Sunpinyin \u4ee3\u7801\u5bfc\u8bfb\u3002\u5c3d\u7ba1\u90a3\u4e2a\u65f6\u5019\u56e0\u4e3a\u77e5\u8bc6\u7684\u50a8\u5907\u4e0d\u591f\u4e30\u5bcc\uff0c\u57fa\u672c\u4e0a\u6ca1\u6709\u80fd\u591f\u7406\u89e3\u6587\u7ae0\u7684\u5185\u5bb9\uff0c\u4f46\u662f\u73b0\u5728\u56de\u8fc7\u5934\u6765\u7684\u8bdd\uff0c\u7ec8\u4e8e\u5728 2017 \u5e74\u7684\u65f6\u5019\uff0c\u81ea\u5df1\u4ece\u5934\u5199\u4e86\u4e00\u4e2a\u7c7b\u4f3c\u7b97\u6cd5\u7684\u8f93\u5165\u6cd5\u5f15\u64ce\uff0c\u6765\u4ee3\u66ff\u65e9\u5df2\u8fc7\u65f6\u7684 Fcitx 4 \u4e2d\u7684\u9ed8\u8ba4\u62fc\u97f3\u548c\u7801\u8868\u3002 \u867d\u7136\u4e5f\u8bb8\u6709\u4eba\u4f1a\u6709\u7591\u95ee\uff0c\u4e3a\u4ec0\u4e48\u8981\u4ece\u5934\u5199\u8fd9\u6837\u4e00\u4e2a\u8f93\u5165\u6cd5\u5f15\u64ce\uff0c\u96be\u9053 Sunpinyin \u4e0d\u591f\u597d\u5417\uff1f\u6211\u5f88\u4e45\u4e4b\u524d\u7684\u4e00\u7bc7\u535a\u5ba2\u5176\u5b9e\u5177\u4f53\u505a\u51fa\u4e86\u4e00\u4e9b\u529f\u80fd\u4e0a\u7684\u89e3\u91ca\u3002\u8fd9\u91cc\u6211\u4eec\u518d\u91cd\u65b0\u4ece\u4ee3\u7801\u89d2\u5ea6\u6765\u89e3\u91ca\u4e00\u4e0b libime \u7684\u6765\u9f99\u53bb\u8109\u3002 libime \u76ee\u524d\u7684\u5e93\u5206\u4e3a 4 \u4e2a\u90e8\u5206\uff0c\u5176\u4e2d\u7ca4\u62fc\u5355\u72ec\u5728\u4e00\u4e2aGit \u4ee3\u7801\u4ed3\u5e93\u91cc\uff0c\u7ca4\u62fc\u5927\u90e8\u5206\u7684\u4ee3\u7801\u662f\u548c\u62fc\u97f3\u7c7b\u4f3c\u7684\u3002\u4e4b\u540e\u4e5f\u4e0d\u4f1a\u7279\u522b\u4ecb\u7ecd\u7ca4\u62fc\u7684\u90e8\u5206\u3002 \u53e6\u5916\u4e00\u4e9b\u539f\u56e0\u662f\uff0c\u5176\u4e2d\u5f88\u591a\u6838\u5fc3\u7684\u4ee3\u7801\u5bf9\u6211\u81ea\u5df1\u6765\u8bf4\u4e5f\u5df2\u7ecf\u4e0d\u5f88\u719f\u6089\u4e86\uff0c\u4e5f\u662f\u8d81\u6b64\u673a\u4f1a\u8bb0\u5f55\u4e00\u4e0b\u7ffb\u65b0\u4e00\u4e0b\u81ea\u5df1\u7684\u8bb0\u5fc6\u3002\u53e6\u5916\u6211\u76f8\u4fe1\u8fd9\u4e9b\u7ecf\u5386\u7684\u90e8\u5206\uff0c\u5bf9\u4e8e\u4e86\u89e3 Linux \u8f93\u5165\u6cd5\u548c Fcitx \u7684\u53d1\u5c55\u8fc7\u7a0b\u7684\u4e5f\u662f\u5f88\u6709\u8da3\u7684\uff0c\u5f53\u6210\u4e00\u4e2a\u6545\u4e8b\u6765\u770b\u5927\u6982\u4e5f\u4e0d\u9519\uff1f \u6838\u5fc3\u90e8\u5206\uff08Core\uff09 \u5728\u89e3\u8bf4 Core \u4e4b\u524d\uff0c\u5c31\u8981\u8bf4\u5230\u4e3a\u4ec0\u4e48\u8981\u5199 libime \u8fd9\u4ef6\u4e8b\u3002\u800c\u8fd9\u5c31\u4e0d\u5f97\u4e0d\u8bf4\u8bf4\u4e4b\u524d\u7684\u51e0\u4e2a\u524d\u8f88\u4eec\uff0c\u6709\u4e9b\u6211\u53ea\u6709\u5927\u81f4\u7684\u4e86\u89e3\uff0c\u6240\u4ee5\u53ef\u80fd\u6709\u9519\u8bef\u3002\u9996\u5148\u662f fcitx4 &hellip; <a href=\"https:\/\/www.csslayer.info\/wordpress\/fcitx-dev\/libime-intro-1\/\">Continue reading <span class=\"meta-nav\">&rarr;<\/span><\/a><\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[44],"tags":[],"class_list":["post-2645","post","type-post","status-publish","format-standard","hentry","category-fcitx-dev"],"_links":{"self":[{"href":"https:\/\/www.csslayer.info\/wordpress\/wp-json\/wp\/v2\/posts\/2645","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.csslayer.info\/wordpress\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.csslayer.info\/wordpress\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.csslayer.info\/wordpress\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.csslayer.info\/wordpress\/wp-json\/wp\/v2\/comments?post=2645"}],"version-history":[{"count":6,"href":"https:\/\/www.csslayer.info\/wordpress\/wp-json\/wp\/v2\/posts\/2645\/revisions"}],"predecessor-version":[{"id":2659,"href":"https:\/\/www.csslayer.info\/wordpress\/wp-json\/wp\/v2\/posts\/2645\/revisions\/2659"}],"wp:attachment":[{"href":"https:\/\/www.csslayer.info\/wordpress\/wp-json\/wp\/v2\/media?parent=2645"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.csslayer.info\/wordpress\/wp-json\/wp\/v2\/categories?post=2645"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.csslayer.info\/wordpress\/wp-json\/wp\/v2\/tags?post=2645"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}