From: Ian Jackson Date: Mon, 29 Jun 2009 16:43:51 +0000 (+0100) Subject: New Word dictionary and new disambiguation algorithm X-Git-Tag: 1.9.2~91 X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~yarrgweb/git?p=ypp-sc-tools.db-test.git;a=commitdiff_plain;h=ac65228e40fa375c829b46607fb4941ff11376e9;ds=sidebyside New Word dictionary and new disambiguation algorithm --- diff --git a/pctb/README.charset b/pctb/README.charset index e1fd3ff..c57f2f5 100644 --- a/pctb/README.charset +++ b/pctb/README.charset @@ -38,7 +38,8 @@ the entry you have just made. Upper vs lower case - important note regarding `l' and `I' ---------------------------------------------------------- -We maintain separate dictionaries for upper and lower case. At the +We maintain separate dictionaries for upper case (Upper), lower case +(Lower), and (initial portions of) mid-phrase words (Word). At the beginning of each cell in the table, we expect uppercase; in the middle of a word we expect lowercase; and, unfortunately, after an inter-word gap, we are not sure. @@ -47,10 +48,16 @@ This is troublesome because `l' and `I' look identical on the screen. So any time we see a word starting with `l' or `I', the program has to ask about it. +After an interword gap, we first search for a Word entry in the +dictionary. If there is a match we use it. Otherwise we search both +the uppercase and lowercase dictionaries; if one matches and the other +doesn't, or one matches a wider character than the other, we use it. +If that fails to resolve the ambiguity we must ask. + *Do not* make an entry in the character set dictionary mapping `vertical stick' to `l' or `I'. Instead, select enough of the whole word in question that no word would start with the other letter, and enter the -whole word or part of it as a new glyph. +whole word or part of it as a new glyph as a new Word. For example, in the supplied dictionary there is already a glyph for `Iron'; this is OK because there are no words which start `lron'. @@ -72,12 +79,11 @@ for the uppercase letter in the lowercase dictionary. Instead, make a new glyph for the last letter of the previous word plus the (unusually narrow) inter-word space, and end that entry with -\x20 (yes, type \ x 20). +a literal space ` '. For example, you might find that `yG' is treated as `y' and the G doesn't get matched. Select the `y' -region of the bitmap and type `y\x20' into the string box. -Sorry for this rather poor UI! +region of the bitmap and type `y ' into the string box. Overlapping characters - ligatures @@ -101,9 +107,10 @@ However since you are not guaranteed to actually get an OCR query at all if the dictionary contains errors, you shouldn't rely on this. If you think you have made mistakes answering OCR queries (for -example, the recognised data is wrong), you should download a fresh -copy of charset-15.txt from - http://www.chiark.greenend.org.uk/~ijackson/ypp-sc-tools/master/pctb/charset-15.txt +example, the recognised data is wrong), you should delete the file +#local-char*#.txt, which contains your local updates. It will then +only use the centrally provided (and vetted) master file (which is +automatically updated when you run the PCTB client, by default). Enabling interactive character set update diff --git a/pctb/TODO b/pctb/TODO index 1e8e4b1..8088a10 100644 --- a/pctb/TODO +++ b/pctb/TODO @@ -1,3 +1,5 @@ +add UI option to dictionary-manager to make user specify which dictionary + to add multi-context entries to install/test dictionary upload/approval onboard ship island name horizontal trim write real uploader diff --git a/pctb/common.c b/pctb/common.c index 3cb4323..bc1cb8d 100644 --- a/pctb/common.c +++ b/pctb/common.c @@ -88,8 +88,14 @@ int dbfile_scanf(const char *fmt, ...) { } void dbfile_assertfail(const char *file, int line, const char *m) { - fatal("Error in database file %s at byte %ld:\n" - " Requirement not met at %s:%d:\n" - " %s", - path,(long)ftell(dbfile), file,line, m); + if (dbfile) + fatal("Error in dictionary file %s at byte %ld:\n" + " Requirement not met at %s:%d:\n" + " %s", + path,(long)ftell(dbfile), file,line, m); + else + fatal("Semantic error in dictionaries:\n" + " Requirement not met at %s:%d:\n" + " %s", + file,line, m); } diff --git a/pctb/ocr.c b/pctb/ocr.c index fe264e8..f610445 100644 --- a/pctb/ocr.c +++ b/pctb/ocr.c @@ -42,6 +42,7 @@ typedef struct DatabaseNode { typedef struct { OcrReader *rd; + OcrCellType ct; int w; Pixcol *cols; int x; @@ -52,30 +53,21 @@ typedef struct { int rx; } FindCharResults; +enum { ct_Lower, ct_Upper, ct_Word, ct_Digit }; static const char *context_names[]= { - "Lower", - "Upper", - "Digit" + "Lower", /* bit 0, value 001 */ + "Upper", /* bit 1, value 002 */ + "Word", /* bit 2, value 004 */ + "Digit", /* bit 3, value 010 */ }; struct OcrCellTypeInfo { /* bitmaps of indices into context_names: */ unsigned initial, nextword, midword; int space_spaces; const char *name; + int (*findchar_select)(const FindCharArgs *fca, + const FindCharResults results[]); }; -const struct OcrCellTypeInfo ocr_celltype_number= { - 4,4,4, - .space_spaces= 5, - .name= "number" -}; -const struct OcrCellTypeInfo ocr_celltype_text= { - .initial=2, /* Uppercase */ - .nextword=3, /* Either */ - .midword=1, /* Lower only */ - .space_spaces= 4, - .name= "text" -}; - #define NCONTEXTS (sizeof(context_names)/sizeof(context_names[0])) @@ -340,9 +332,12 @@ static DatabaseNode *findchar(const FindCharArgs *fca, int *match_rx) { nmatches++; } if (nmatches==1) { - debugf( " unambiguous"); + debugf(" unique"); } else { - match=-1; + debugf(" ambiguous"); + match= !fca->ct->findchar_select ? -1 : + fca->ct->findchar_select(fca,results); + debugf(" resolved %s", match<0 ? "" : context_names[match]); } if (match<0) return 0; @@ -351,6 +346,33 @@ static DatabaseNode *findchar(const FindCharArgs *fca, int *match_rx) { return results[ctxi].match; } +static int findchar_select_text(const FindCharArgs *fca, + const FindCharResults results[]) { + if (fca->ctxmap != 017) return -1; + + dbassert(! results[ct_Digit].match ); + if (results[ct_Word].match) return ct_Word; + if (results[ct_Lower].rx > results[ct_Upper].rx) return ct_Lower; + if (results[ct_Upper].rx > results[ct_Lower].rx) return ct_Upper; + return -1; +} + +const struct OcrCellTypeInfo ocr_celltype_number= { + 010,010,010, + .space_spaces= 5, + .name= "number", + .findchar_select= 0 +}; +const struct OcrCellTypeInfo ocr_celltype_text= { + .initial= 012, /* Digit|Upper */ + .nextword= 017, /* Digit|Upper|Lower|Word */ + .midword= 014, /* Digit|Lower */ + .space_spaces= 4, + .name= "text", + .findchar_select= findchar_select_text +}; + + const char *ocr_celltype_name(OcrCellType ct) { return ct->name; } OcrResultGlyph *ocr(OcrReader *rd, OcrCellType ct, int w, Pixcol cols[]) { @@ -359,6 +381,7 @@ OcrResultGlyph *ocr(OcrReader *rd, OcrCellType ct, int w, Pixcol cols[]) { FindCharArgs fca; fca.rd= rd; + fca.ct= ct; fca.w= w; fca.cols= cols; fca.x= -1;