ᩥ᭩ศ㢮ࡢᡭἲ୍⯡⥺ᙧࣔࢹࣝࢆ⏝࠸ࡓ
ⱥㄒࣛࢸࣥࢢ࠾ࡅࡿᩥἲⓗㄗࡾࡢᙳ㡪
▼ 㞝㝯㸦᪩✄⏣Ꮫ Ꮫ⥲ྜ◊✲ࢭࣥࢱ࣮㸧 ㏆⸨ ᝆ㸦᪩✄⏣Ꮫ ࢢ࣮ࣟࣂ࢚ࣝࢹࣗࢣ࣮ࢩࣙࣥࢭࣥࢱ࣮㸧
ᮏ◊✲࡛ࡣ㸪ᩥ᭩ศ㢮ࡢᡭἲ୍⯡⥺ᙧࣔࢹࣝࢆ⏝࠸࡚ⱥㄒࣛࢸࣥࢢ࠾ࡅࡿᩥἲⓗㄗࡾࡢ ᙳ㡪ࡘ࠸࡚ㄪᰝࡋࡓ㸬᪥ᮏேⱥㄒᏛ⩦⪅ࡢࢫࣆ࣮࢟ࣥࢢࡸࣛࢸࣥࢢࡢࣃࣇ࢛࣮࣐ࣥࢫࢆ⮬ື᥇ Ⅼࡍࡿヨࡳࡀ㏆ᖺὀ┠ࢆ㞟ࡵ࡚࠸ࡿࡀ㸪ࡢ≉ᚩ㔞ࡀࢀホ౯ᐤࡋ࡚࠸ࡿࡘ࠸࡚ࡣ㸪ࡲ ࡔ༑ศ᫂ࡽࡉࢀ࡚࠸࡞࠸㸬ᮏ◊✲࡛ࡣ㸪ࣃࣇ࢛࣮࣐ࣥࢫࢆ ࡿ୍ࡘࡢᣦᶆ࡛࠶ࡿᩥἲⓗṇ☜ࡉ ↔Ⅼࢆᙜ࡚࡚㸪ࡑࡢᙳ㡪ࢆㄪᰝࡋࡓ㸬ࡑࡢ⤖ᯝ㸪࢚ࢵࢭホ౯ᙳ㡪ࢆ࠼࡚࠸ࡿᩥἲⓗㄗࡾࡣ㸪 ືモࡢㄒᙡ㛵ࡍࡿ࢚࣮ࣛㄒ㡰㛵ࡍࡿ࢚࣮ࣛࡢ✀㢮ࡢᩥἲⓗㄗࡾ࡛࠶ࡗࡓ㸬ⱥㄒᏛ⩦⪅ࡢࣛ ࢸࣥࢢホ౯࠾࠸࡚ࡣ㸪ືモ㛵ࡍࡿㄗࡾࡣ㸪ࡢᩥἲⓗㄗࡾẚ࡚㸪࢚ࢵࢭࡢయⓗホ౯ ࡁ࡞ᙳ㡪ࢆ࠼࡚࠸ࡿࡇࢆ♧၀ࡍࡿ⤖ᯝ࡞ࡗࡓ㸬Investigating effects of accuracy in English writing based on the
technique of essay classification and generalized linear model
Yutaka Ishii (Center for Higher Education Studies, Waseda University) Yusuke Kondo (Global Education Center, Waseda University)
This study investigates effects of grammatical accuracy in English writing based on the technique of essay classification and generalized linear model. In recent years, there has been an increasing interest in automated scoring of learners' production skills such as speaking and writing. However, one major issue in the research concerned what linguistic features contribute to essay grading. This paper assesses the significance of grammatical accuracy in essay scoring. The results showed that verb related errors (errors in word selection) and errors in word order affect learners’ essay evaluation.
㸯㸬ࡣࡌࡵ
ᮏ◊✲࡛ࡣ㸪ᩥ᭩ศ㢮ࡢᡭἲ୍⯡⥺ᙧࣔࢹ ࣝࢆ⏝࠸࡚ⱥㄒࣛࢸࣥࢢ࠾ࡅࡿᩥἲⓗㄗ ࡾࡢᙳ㡪ࡘ࠸࡚ㄪᰝࡋࡓ㸬᪥ᮏேⱥㄒᏛ⩦⪅ࡢ ࢫࣆ࣮࢟ࣥࢢࡸࣛࢸࣥࢢࡢࣃࣇ࢛࣮࣐ࣥࢫ ࢆ⮬ື᥇Ⅼࡍࡿヨࡳࡀ㏆ᖺὀ┠ࢆ㞟ࡵ࡚࠸ࡿࡀ㸪 ࡢ≉ᚩ㔞ࡀࢀホ౯ᐤࡋ࡚࠸ࡿ ࡘ࠸࡚ࡣ㸪ࡲࡔ༑ศ᫂ࡽࡉࢀ࡚࠸࡞࠸㸬ᮏ ◊✲࡛ࡣ㸪ࣃࣇ࢛࣮࣐ࣥࢫࢆ ࡿ୍ࡘࡢᣦᶆ࡛࠶ ࡿᩥἲⓗṇ☜ࡉ↔Ⅼࢆᙜ࡚࡚㸪ࡑࡢᙳ㡪ࢆㄪᰝ ࡍࡿ㸬㸰㸬ඛ⾜◊✲
ࡇࢀࡲ࡛ࡢࣛࢸࣥࢢ◊✲ࡢከࡃࡣ㸪➨ゝ ㄒࡋ࡚ࡢⱥㄒࣛࢸࣥࢢࢆᑐ㇟ࡋ࡚⾜ࢃ ࢀ࡚ࡁ࡚࠾ࡾ㸪᪥ᮏேⱥㄒᏛ⩦⪅ࢆᑐ㇟ࡋࡓ◊ ✲ࡣ࠶ࡲࡾ⾜ࢃࢀ࡚ࡇ࡞ࡗࡓ㸬ࡲࡓ᪥ᮏ࠾ࡅ ࡿⱥㄒࣛࢸࣥࢢ◊✲ࡣࡁࡃศࡅ࡚ࡘ ࡢၥ㢟ࡀᏑᅾࡍࡿゝࢃࢀ࡚ࡁࡓ㸬୍ࡘࡣ㸪ࣛ ࢸࣥࢢࡢホ౯ࡣ㸪ᩍᖌࡗ࡚㸪ࡁ࡞㈇ᢸ࡛ ࠶ࡿ࠸࠺ࡇ㸪ホᐃ⪅㛫ࡢホ౯ࡢ୍⮴ࡢၥ 㢟࡛࠶ࡿ㸬 ୖグࡢၥ㢟ࢆゎỴࡍࡿࡓࡵ㸪㏆ᖺὀ┠ࢆ㞟ࡵ ࡚࠸ࡿࡢࡀࣛࢸࣥࢢࡢ⮬ືホ౯࡛࠶ࡿ㸬ࣛ ࢸࣥࢢࡢ⮬ືホ౯࠾࠸࡚ࡣ㸪ࡉࡲࡊࡲ࡞ゝㄒ ⓗᣦᶆࡀ⏝࠸ࡽࢀࡿ㸬௦⾲ࡋ࡚Educational Testing Service㸦ETS㸧ࡢ e-rater ࡣ㸪ୗグ 12 ಶࡢኚᩘࢆ⏝࠸࡚࠸ࡿ㸬 1. ⥲ㄒᩘᑐࡍࡿᩥἲ࢚࣮ࣛࡢྜ 2. ⥲ㄒᩘᑐࡍࡿㄒࡢ⏝ἲࡘ࠸࡚ࡢ࢚ࣛ ࣮ࡢྜ 3. ⥲ㄒᩘᑐࡍࡿᡭ㡰ࡢ࢚࣮ࣛࡢྜ 4. ⥲ㄒᩘᑐࡍࡿࢫࢱࣝࡘ࠸࡚ࡢ࢚࣮ࣛ ࡢྜ 5. ᚲせࡉࢀࡿㄯヰせ⣲ࡢᩘ 6. ㄯヰせ⣲࠾ࡅࡿᖹᆒㄒᩘ 7. సᩥࢆ 6 Ⅼἲ࡛᥇Ⅼࡍࡿ㝿ㄒᙡࡢ㢮ఝ ᗘࡀ୍␒㏆࠸Ⅼᩘ 8. ᭱㧗Ⅼࢆྲྀࡗࡓసᩥࡢㄒᙡࡢ㢮ఝᗘ 9. Type-Token Ratio 10. ㄒᙡࡢᅔ㞴ᗘ11. ᖹᆒ༢ㄒ㛗 12. ⥲ㄒᩘ
㸦Burstein, Chodorow & Leacock, 2004㸧 ୖグࡢኚᩘࡢ୰࡛యࡢ3 ศࡢ 1 ࢆ༨ࡵࡿࡢࡀᩥ ἲⓗㄗࡾ࡛࠶ࡿ㸬ୖグࢆ㚷ࡳࡿ㸪ࣛࢸࣥࢢ ࡢホ౯ࡣᩥἲⓗㄗࡾࡁࡃᐤࡍࡿ⪃࠼ࡽ ࢀࡿ㸬୍᪉࡛㸪Jones㸦2006㸧ࡣ㸪ࣛࢸࣥࢢ ࠾ࡅࡿ46 ⟠ᡤࡢᩥἲⓗㄗࡾࢆゞṇࡋ㸪ࡑࢀࢆ ࠶ࡿ⮬ື᥇Ⅼࢩࢫࢸ࣒ホ౯ࡉࡏࡓࡇࢁ㸪ゞṇ ๓ゞṇᚋࡢࢫࢥࡀࡃྠࡌ࡛࠶ࡗࡓ࠸࠺ ࡇࢆሗ࿌ࡋ࡚࠸ࡿ㸬ࡇࡢ◊✲ࡣ㸪Ꮫ⩦⪅ࡢᩥἲ ⓗㄗࡾࡀࣛࢸࣥࢢࡢホ౯࠾࠸࡚ᐤࡋ࡞ ࠸ྍ⬟ᛶࢆ♧၀ࡍࡿ◊✲࡛࠶ࡿ㸬ࡑࡇ࡛ᮏ◊✲࡛ ࡣ㸪ᩥἲⓗㄗࡾ࠸࠺ኚᩘࡢࡳ࡛ࢀࣛࢸ ࣥࢢࡢホ౯ࢆண ࡍࡿࡇࡀ࡛ࡁࡿࢆ᳨ウ ࡍࡿ㸬 ᮏ◊✲ࡀ㸪ᩥἲⓗㄗࡾ↔Ⅼࢆᙜ࡚ࡿ⌮⏤ࡋ ࡚ࡣ㸪Ⅼࡀᣲࡆࡽࢀࡿ㸬୍Ⅼ┠ࡣ㸪⮬↛ゝㄒฎ ⌮ࡢᩍ⫱ᛂ⏝࡛㏆ᖺ┒ࢇᩥἲⓗㄗࡾࡢ⮬ື᳨ ฟ㛵ࡍࡿඹ㏻ࢱࢫࢡ࡞ࡀ⾜ࢃࢀ࡚࠸ࡿ࠸ ࠺Ⅼ࡛࠶ࡿ㸬࠼ࡤ㸪ࡇࢀࡲ࡛ୗグ⾲1 ࡢࡼ࠺ ࡞ඹ㏻ࢱࢫࢡࡀ⾜ࢃࢀ࡚࠸ࡿ㸬 ⾲-1 ⮬↛ゝㄒฎ⌮ࡢᩍ⫱ᛂ⏝ࡢඹ㏻ࢱࢫࢡ ㆟ྡ ࢱࢫࢡ
Helping Our Own 㸦HOO㸧2011
ㄽᩥࡢᩥἲⓗㄗࡾゞṇ Helping Our Own
㸦HOO㸧2012 ๓⨨モ㝈ᐃモࡢᩥἲ ⓗㄗࡾゞṇ Native Language Identification Shared Task ⱥసᩥࡽⱥㄒᏛ⩦⪅ ࡢẕㄒ᥎ᐃ CoNLL 2013 㝈ᐃモ㸪๓⨨モ㸪ᩘ㸪 ືモࡢᙧ㸪୍⮴㸪ࢫ࣌ ࣝ㸪ྃㄞⅬࡢᩥἲⓗㄗ ࡾゞṇ ㄗࡾ᳨ฟ࣭ゞṇ࣮࣡ࢡ ࢩࣙࢵࣉ 2012 ๓⨨モㄗࡾ࣭ືモ㸦 ㄒ-ືモࡢ୍⮴㸧ㄗࡾࡢ 2 ࡘࡢࢺࣛࢵࢡຍ࠼㸪 ㄗࡾࡢ✀㢮ࢆ㝈ᐃࡋ࡞ ࠸ࣃࣟࢵࢺࢺࣛࢵࢡ ୖグࡢࡼ࠺࡞◊✲࠾ࡅࡿ▱ぢࡣ㸪እᅜㄒᩍ⫱◊ ✲ࡢⓎᒎࡁࡃᐤࡍࡿࡶࡢ࡛࠶ࡿ⪃࠼ࡽ ࢀࡿ㸬ࡋࡋ࡞ࡀࡽ㸪እᅜㄒᩍ⫱◊✲⪅ࡗ࡚ ࡶ᭱ࡶ㛵ᚰࡀ࠶ࡿࡢࡣ㸪࠺࠸ࡗࡓᩥἲⓗㄗࡾࡀ ࣛࢸࣥࢢࡢホ౯ᐤࡍࡿ࠸࠺Ⅼ࡛࠶ ࡿ㸬ࡑࡢ◊✲ᡂᯝࡣ㸪ᩍᐊ࠾ࡅࡿࣛࢸࣥࢢ ᣦᑟ࡞ࡶ⏕ࡍࡇࡀྍ⬟࡛࠶ࡿ㸬 Ⅼ┠ࡣ㸪እᅜㄒᩍ⫱ࡢά⏝ࡀ㐍ࡵࡽࢀ࡚࠸ ࡿ࣮ࣚࣟࢵࣃゝㄒඹ㏻ཧ↷ᯟ࠾ࡅࡿᩥἲⓗṇ ☜ࡉࡢグ㏙ࡢ⢭⦓࡛࠶ࡿ㸬ୗグ⾲2 ࡣ㸪ᩥἲⓗ ㄗࡾᑐࡍࡿ⬟ຊグ㏙ᩥ࡛࠶ࡿࡀ㸪ࡑࡢグ㏙ࢆぢ ࡚ࡳࡿ㸪ẁ㝵ࡈࡢㄗࡾࡢ⛬ᗘࡘ࠸࡚₍↛ ࡋࡓグ㏙ࡀ࡞ࡉࢀ࡚࠸ࡿࡢࡳ࡛࠶ࡿ㸬Ꮫ⩦⪅ࡀ⩦ ⇍ᗘࡈ࠾ࡋࡸࡍ࠸ᩥἲⓗㄗࡾࢆ≉ᐃࡍࡿ ࡇࡣ㸪ࡇ࠺ࡋࡓ⬟ຊグ㏙ᩥࡢ⢭⦓ࡶᛂ⏝ࡍ ࡿࡇࡀྍ⬟࡛࠶ࡿ㸬 ⾲-2 ࣮ࣚࣟࢵࣃゝㄒඹ㏻ཧ↷ᯟࡢᩥἲⓗṇ☜ࡉࡢ ⬟ຊグ㏙ᩥ ࣞ࣋ࣝ ᩥἲⓗṇ☜ࡉ C2 㸦࠼ࡤ㸪ࡇࢀࡽゝ࠺ࡇࢆ⪃࠼࡚࠸ ࡿࡸ㸪ேࡢᛂࢆࣔࢽࢱ࣮ࡋ࡚࠸ࡿ ࡼ࠺࡞࠸ࡗࡓ㸧ࡢࡇὀពࢆᡶ ࡗ࡚࠸ࡿ࡛ࡶ㸪」㞧࡞ゝⴥࡘ࠸࡚ᖖ 㧗࠸ᩥἲ㥑ຊࢆ⥔ᣢࡋ࡚࠸ࡿ㸬 C1 ᖖ㧗࠸ᩥἲⓗṇ☜ࡉࢆ⥔ᣢࡍࡿ㸬ㄗࡾࡣᑡ࡞ࡃ㸪ぢࡘࡅࡿࡇࡣ㞴ࡋ࠸㸬 B2 㧗࠸ᩥἲ㥑ຊࡀ࠶ࡿ㸬ࡣࠕゝ࠸㛫 㐪࠸ࠖࡸ㸪ᩥᵓ㐀࡛ࡢഅ↛㉳ࡇࡋࡓㄗࡾ ࡸல⣽࡞ഛࡀぢࡽࢀࡿሙྜࡀ࠶ࡿࡀ㸪 ࡑࡢᩘࡣᑡ࡞ࡃ㸪ᚋ࡛ぢ┤ࡏࡤゞṇ࡛ࡁ ࡿࡶࡢࡀከ࠸㸬 ẚ㍑ⓗ㧗࠸ᩥἲ㥑ຊࡀぢࡽࢀࡿ㸬ㄗゎ ࡘ࡞ࡀࡿࡼ࠺࡞㛫㐪࠸ࡣ≢ࡉ࡞࠸㸬 B1 㥆ᰁࡳࡢ࠶ࡿ≧ἣ࡛ࡣ㸪ྜṇ☜ࢥ࣑ ࣗࢽࢣ࣮ࢩࣙࣥࢆ⾜࠺ࡇࡀ࡛ࡁࡿ㸬ከ ࡃࡢሙྜ㧗࠸࡛ࣞ࣋ࣝࡢ㥑⬟ຊࡀ࠶ ࡿࡀ㸪ẕㄒࡢᙳ㡪ࡀ᫂ࡽ࡛࠶ࡿ㸬ㄗࡾ ࡶぢࡽࢀࡿࡀ㸪ᮏேࡀ㏙ࡼ࠺ࡋ࡚࠸ ࡿࡇࡣ᫂ࡽศࡿ㸬 ẚ㍑ⓗண ྍ⬟࡞≧ἣ࡛㸪㢖⦾ࢃࢀ ࡿࠕ⧞ࡾ㏉ࡋࠖࡸࣃࢱ࣮ࣥࡢࣞࣃ࣮ࢺࣜ ࣮ࢆ㸪ྜṇ☜࠺ࡇࡀ࡛ࡁࡿ㸬 A2 ࠸ࡃࡘࡢ༢⣧࡞ᩥἲᵓ㐀ࢆṇࡋࡃ ࠺ࡇࡀ࡛ࡁࡿࡀ㸪౫↛ࡋ࡚Ỵࡲࡗ࡚ ≢ࡍᇶᮏⓗ࡞㛫㐪࠸ࡀ࠶ࡿ Ɇ ࠼ࡤ㸪 ไࢆΰྠࡋࡓࡾ㸪ᛶ࣭ᩘ࣭᱁࡞ࡢ୍ ⮴ࢆᛀࢀࡓࡾࡍࡿഴྥࡀ࠶ࡿ㸬ࡋࡋ㸪 ᮏேࡀఱࢆゝ࠾࠺ࡋ࡚࠸ࡿࡢࡣࡓ ࠸࡚࠸ࡢሙྜ᫂ࡽ࡛࠶ࡿ㸬 A1 Ꮫ⩦῭ࡳࡢࣞࣃ࣮ࢺ࣮ࣜࡘ࠸࡚㸪࠸ࡃ ࡘࡢ㝈ࡽࢀࡓ༢⣧࡞ᩥἲᵓ㐀ࡸᵓᩥ ࢆ࠺ࡇࡣ࡛ࡁࡿ㸬 ᪥ᮏேⱥㄒᏛ⩦⪅ࡢࣛࢸࣥࢢホ౯ᩥἲ ⓗㄗࡾࡢ㛵ಀᛶࢆ᳨ウࡋࡓࡶࡢࡋ࡚Kitamura 㸦2011㸧ࡀᏑᅾࡍࡿ㸬Kitamura㸦2011㸧࡛ࡣ㸪 Ỵᐃᮌศᯒࢆ⏝࠸࡚㸪ᩥἲⓗㄗࡾࡢ࢚ࢵࢭホ౯ ࡢᙳ㡪ࢆㄪᰝࡋࡓ㸬ࡑࡢ⤖ᯝ㸪ࢺࣆࢵࢡࡼࡗ ࡚ࡣᩥἲⓗㄗࡾࡀ࢚ࢵࢭホ౯ᐤࡍࡿࡇ ࡀ♧ࡉࢀࡓ㸬ࡋࡋ࡞ࡀࡽ㸪ᑐ㇟ࡉࢀࡓᩥἲⓗ
ㄗࡾࡣㄒືモࡢ୍⮴㸦He have been living there since June.㸧㸪ືモࡢᙧ㸦I can’t skiing well, but… 㸧 㸪 ࡞ ᩥ 㸦 Because people’s interesting thing is not the same.㸧ࡢ୕✀㢮࡛ ࠶ࡗࡓࡓࡵ㸪ᮏ◊✲࡛ࡣ㸪20 ಶࡢᩥἲⓗㄗࡾࢆ ᑐ㇟ࡋ㸪᳨ウࢆヨࡳࡿ㸬
Tang and Liu㸦2005㸧࡛ࡣ㸪ᩥ᭩ศ㢮ࢱࢫࢡ ࡛Ⰻࡃ⏝࠸ࡽࢀࡿ⣲ᛶ㑅ᢥࡋ࡚㸪Information Gain㸪Chi-squared test㸪Odds ratio㸪Bi-Nominal Separation ࢆ ᣲ ࡆ ࡚ ࠸ ࡿ 㸬 ᮏ ◊ ✲ ࡛ ࡣ 㸪 Bi-Nominal Separation ࢆ⏝࠸ࡿ㸬
㸱㸬ࢹ࣮ࢱ
ᮏ ◊ ✲ ࡛ ⏝ ࠸ ࡓ ࢹ ࣮ ࢱ ࡣ 㸪Konan-JIEM learner corpus ࡤ ࢀ ࡿ ゝ ㄒ ㈨ ※ ࡛ ࠶ ࡿ 㸦Nagata, Whittaker, & Sheinman, 2011㸧㸬ࡇ ࡢࢥ࣮ࣃࢫࡣ㸪⏥༡Ꮫᩍ⫱ ᐃ◊✲ᡤ㸦JIEM㸧 ࡀඹྠ࡛㞟ࡋ㸪ࣀࢸ࣮ࢩࣙࣥࢆ⾜ࡗࡓࢥ࣮ࣃ ࢫ࡛࠶ࡿ㸬᪥ᮏேⱥㄒᏛ⩦⪅ࡢ170 ࢚ࢵࢭࡽ ᡂࡾ㸪ࡲࡓ㸪ᩥἲㄗࡾሗရモ㸭ྃሗࡀேᡭ ࡛ࡉࢀ࡚࠸ࡿ㸬◊✲┠ⓗࡢࡓࡵ࡛࠶ࢀࡤゝㄒ ㈨※༠ࢆ㏻ࡌ࡚㸪㉎ධࡀྍ⬟࡛࠶ࡿ㸬ᴫせࢆ⾲ 3 㸪࢚࣮ࣛࢱࢢࡢ✀㢮ࢆ⾲ 4 ♧ࡍ㸬 ⾲-3 ࢹ࣮ࢱࡢᴫせ ࢚ࢵࢭࡢᩘ 170 ࢚ࢵࢭࢆ᭩࠸ࡓᏛ⏕ࡢᩘ 10 ⥲ᩥᩘ 2409 ⥲༢ㄒᩘ 19285 ␗࡞ࡾㄒᩘ 2054 ᡂ⏣㸦2013㸧࡛ࡣ㸪Konan-JIEM learner corpus ࡢಶࠎࡢ࢚ࢵࢭ᪥ᮏࡢᏛ࡛ⱥㄒᩍ ⫱ᚑࡋ࡚࠸ࡿᩍဨࡀホ౯ࢆ࠼࡚࠸ࡿ㸬ᡂ⏣ 㸦2013㸧࡛ࡣࡍ࡚ࡢ࢚ࢵࢭࢆ 3 ேࡢᩍဨࡀ ホ౯ࡋ࡚ඃࢀࡓࡶࡢࡑ࠺࡛࡞࠸ࡶࡢศ㢮ࡋ ࡚࠸ࡿ㸬 ᮏ◊✲ࡣ㸪ᡂ⏣㸦2013㸧ࡢホ౯ࢆ᥇⏝ࡋ㸪 య࡛ 170 ࠶ࡿ࢚ࢵࢭࡽࣛࣥࢲ࣒ඃࢀࡓ ࡶࡢ50 ࡑ࠺࡛࡞࠸ࡶࡢ 50 ࢆ㑅ࡧ 100 ࡢ࢚ࢵ ࢭࢆศᯒࡢᑐ㇟ࡋࡓ㸬 ⾲-4 ࢚࣮ࣛࢱࢢࡢ✀㢮 ࢱࢢ ෆᐜ n_num ྡモ-༢」࢚࣮ࣛ ຍ⟬㸪ྍ⟬࢚࣮ࣛ n_lxc ྡモ-ㄒᙡ㑅ᢥ࢚࣮ࣛ n_o ྡモ-ࡑࡢࡢ࢚࣮ࣛ pn ௦ྡモ㛵ࡍࡿ࢚࣮ࣛ v_agr ືモ-ே⛠࣭ᩘࡢ୍⮴ v_tns ືモ-ไ࢚࣮ࣛ v_lxc ືモ-ㄒᙡ㑅ᢥ࢚࣮ࣛ v_o ືモ-ࡑࡢࡢ࢚࣮ࣛ mo ຓືモ㛵ࡍࡿ࢚࣮ࣛ aj ᙧᐜモ㛵ࡍࡿ࢚࣮ࣛ av モ㸦ྃ㸧㛵ࡍࡿ࢚࣮ࣛ prp ๓⨨モ㛵ࡍࡿ࢚࣮ࣛ at ෙモ㛵ࡍࡿ࢚࣮ࣛ con ᥋⥆モ㛵ࡍࡿ࢚࣮ࣛ rel 㛵ಀモ㛵ࡍࡿ࢚࣮ࣛ itr ၥモ㛵ࡍࡿ࢚࣮ࣛ o_lxc ㄒ௨ୖࡽᡂࡿᡂ࡛ྃࡢ ㄒᙡ㑅ᢥ࣑ࢫ ord ㄒ㡰࢚࣮ࣛ uk ≉ᐃ⬟࡞࢚࣮ࣛ ᵓᡂୖࡢ⮴ⓗ࡞࣑ࢫ f ࣇࣛࢢ࣓ࣥࢺ㸦᩿∦㸪ᮍࡢᩥ➼㸧
㸲㸬ศᯒ
ࡲ ࡎ 㸪 ࢚ ࢵ ࢭ ୰ ࡢ ㄗ ࡾ ࡢ Bi-Normal Separation㸦BNS: Forman, 2003㸧ࢆ⟬ฟࡋ ࡓ㸦⾲ 5㸧㸬BNS ࡣ㸪࠶ࡿ㇟ࡀ࠶ࡿ࢝ࢸࢦ ࣜฟ⌧ࡋࡓ࠺ࡔࡅ࡛࡞ࡃ㸪ࡑࡢ㇟ࡀ ఱᅇฟ⌧ࡋࡓ࠸࠺㢖ᗘࢆ⾲ࡋࡓᣦᶆ࡛࠶ ࡿ㸬ᩥἲⓗㄗࡾࡣࡑࡢ✀㢮ࡔࡅ࡛࡞ࡃ㸪ࡑࡢฟ ⌧㢖ᗘࡀ࢚ࢵࢭࡢホ౯ᙳ㡪ࢆ࠼ࡿ⪃ ࠼ࡽࢀࡿࡓࡵ㸪ᮏ◊✲࡛ᣦᶆࡋ࡚᥇⏝ࡋࡓ㸬 ᮏ◊✲࠾࠸࡚BNS ࡣ㸪ඃࢀࡓ࢚ࢵࢭࡑ ࠺࡛࡞࠸࢚ࢵࢭ࠾࠸࡚ࡑࢀࡒࢀࡢㄗࡾࡀ ฟ⌧ࡋ࡚࠸ࡿ㔞ࡢᕪࢆ♧ࡍࡇ࡞ࡿ㸬 ฟ⌧㢖ᗘࡀᴟ➃ప࠸㸦10 ௨ୗ㸧ࡢᣦᶆࢆ㝖 ࡃ㸪ඃࢀࡓ࢚ࢵࢭࡑ࠺࡛࡞࠸ࡶࡢ࠾࠸ ࡚ࡁ࡞್ࢆ♧ࡍࡶࡢࡣv_lxc㸦ືモࡢㄒᙡ 㛵ࡍࡿ࢚࣮ࣛ㸧 ord㸦ㄒ㡰㛵ࡍࡿ࢚࣮ࣛ㸧 ࡢ2 ✀㢮ࡢᩥἲⓗㄗࡾ࡛࠶ࡗࡓ㸬 ⾲-5 ࢚࣮ࣛࡢ㢖ᗘ Bi-Normal Separation ್ ࢱࢢ ฟ⌧㢖ᗘ BNS Good Poor n_num 52 60 0.18 n_lxc 15 8 0.78 n_o 41 63 0.54 pn 33 38 0.18 v_agr 67 68 0.02 v_tns 40 69 0.68 v_lxc 15 49 1.45 v_o 9 7 0.31 mo 22 40 0.74 aj 28 27 0.05 av 79 125 0.57 prp 122 169 0.41 at 17 18 0.07 con 5 5 0.00rel 2 0 1.00 itr 6 2 1.35 o_lxc 9 8 0.15 ord 20 60 1.35 uk 2 4 0.86 f 0 4 1.00 ḟ㸪ᩥἲⓗㄗࡾࡢ㢖ᗘࢆண ኚᩘ㸪࢚ࢵࢭࡢ ホ౯ࢆᇶ‽ኚᩘࡋ࡚ࣟࢪࢫࢸࢵࢡᅇᖐࢆ⾜ ࠸㸪ࡍ࡚ࡢᩥἲⓗㄗࡾࢆண ኚᩘࡋ࡚ጞࡵ㸪 ࡦࡘࡎࡘኚᩘࢆῶࡽࡋ㸪AIC ࡀᨵၿࡋ࡞ࡃ࡞ࡿ ࡲ࡛⧞ࡾ㏉ࡋࡓ㸬᭱⤊ⓗAIC ࡣ 125.6 ࡞ࡾ㸪 ṧࡗࡓኚᩘࡣ㸪v_lxc㸦ືモࡢㄒᙡ㛵ࡍࡿ࢚ࣛ ࣮㸧ord㸦ㄒ㡰㛵ࡍࡿ࢚࣮ࣛ㸧ࡢ 2 ✀㢮ࡢᩥ ἲⓗㄗࡾ࡛࠶ࡗࡓ㸬௨ୗ v_lxc㸦ືモࡢㄒᙡ 㛵ࡍࡿ࢚࣮ࣛ㸧ord㸦ㄒ㡰㛵ࡍࡿ࢚࣮ࣛ㸧ࡢ 㢖ᗘࢆホ౯ู♧ࡍ㸬ᅗ1 ࠾ࡼࡧ 2 ࡀ♧ࡍࡼ࠺㸪 ホ౯ࡀGood ࡢ࢚ࢵࢭࡢࢇࡇࢀࡽࡢᩥ ἲⓗㄗࡾࡣྵࡲࢀ࡚࠸࡞࠸୍᪉࡛㸪ホ౯Poor ࡢ ࢚ࢵࢭࡢ⣙༙ᩘࡇࢀࡽࡢㄗࡾࡀྵࡲࢀ࡚࠸ ࡿ㸬ࡲࡓ㸪ホ౯Good ࡢ࢚ࢵࢭࡇࢀࡽ 2 ࡘࡢ ࢚࣮ࣛࡀ3 ࡘ௨ୖྵࡲࢀ࡚࠸ࡿࡇࡣ࡞ࡗࡓ㸬 እࡣ࠶ࡿࡀ㸪ࡇࡢഴྥࡣࡢᩥἲⓗㄗࡾ࠾࠸ ࡚ࡶぢࡽࢀࡓ㸬 ᅗ-1 ホ౯ Good ࡢ v_lxc㸦ᕥ㸧 ord㸦ྑ㸧ࡢ㢖ᗘ ᅗ-2 ホ౯ Poor ࡢ v_lxc㸦ᕥ㸧 ord㸦ྑ㸧ࡢ㢖ᗘ ࡇࡢࡼ࠺࡞ࢹ࣮ࢱࡢഴྥࡽ㸪ᩥἲⓗㄗࡾࡢᅇᩘ ホ౯ࡢቃ⏺⥺ࡀᘬࡅࡿ⪃࠼㸪Ỵᐃᮌศᯒࢆ⏝ ࠸࡚ホ౯ࡢண ࢆヨࡳࡓ㸬Ỵᐃᮌศᯒࡼࡿண ⤖ᯝࢆ⾲6 ♧ࡍ㸦㐺ྜ⋡: 0.62㸪⌧⋡: 0.81㸪 ≉␗ᗘ: 0.69㸪ṇ☜ᗘ: 0.74㸪F ್: 0.70㸧㸬ᅗ 3 Ỵᐃᮌศᯒࡢ⤖ᯝࢆ♧ࡍ㸬ࣟࢪࢫࢸࢵࢡᅇᖐศ ᯒࡢ⤖ᯝࡣ␗࡞ࡿࡀ㸪ࡇࡢ⤖ᯝࡶ࢚ࢵࢭホ౯ ࠾࠸࡚ືモࡢㄒᙡ㛵ࡍࡿ࢚࣮ࣛࡀ㔜せ࡞ᩥ ἲⓗㄗࡾ࡛࠶ࡿࡇࢆ♧ࡋ࡚࠸ࡿ㸬 ⾲-6 Ỵᐃᮌศᯒࡢศ㢮⢭ᗘ ┿ࡢ⤖ᯝ ண ⤖ᯝ Good Poor Good 31 19 Poor 7 43 ᅗ-3 Ỵᐃᮌศᯒࡢ⤖ᯝ ᐇ㝿Ꮫ⩦⪅ࡢ⏘ฟࡋࡓⱥᩥࡢࢆୗグ♧ ࡍ㸬
Last, I want to earn much money and I want to <v_lxc crr="travel">trip</v_lxc> <ord crr="everywhere in Japan">in Japan everywhere</ord>. ୖグࡢࡼ࠺㸪ືモ travel ࢆࢃ࡞ࡅࢀࡤ࡞ ࡽ࡞࠸⟠ᡤྡモ trip ࡀ᭩ࢀ࡚࠸ࡿ㸬ࡇ࠺ ࠸ࡗࡓືモࢆ᭩࡞ࡅࢀࡤ࡞ࡽ࡞࠸⟠ᡤྡ モࢆグࡋ࡚ࡋࡲ࠺ࢱࣉࡢㄗࡾࡢ⏘ฟࡣ v_lxc 㸦ືモࡢㄒᙡ㛵ࡍࡿ࢚࣮ࣛ㸧ከࡃ㸪ࡇ࠺ࡋ ࡓㄗࡾࡣసᩥࡢホ౯ᐤࡋࡸࡍ࠸࠸࠺ࡇ ࢆᩍᖌࡣᩍᐊ࡛ᩍ࠼ࡿᚲせࡀ࠶ࡿ⪃࠼ࡽ ࢀࡿ㸬
㸳㸬⪃ᐹ
ᮏ◊✲࡛ࡣ㸪BNS㸪ࣟࢪࢫࢸࢵࢡᅇᖐศᯒ㸪 Ỵᐃᮌศᯒࢆ⏝࠸࡚㸪ⱥㄒᏛ⩦⪅ࡢ࢚ࢵࢭᑐ ࡍࡿホ౯࠼ࡿᩥἲⓗㄗࡾࡢᙳ㡪ࢆㄪᰝࡋࡓ㸬 ࡑࢀࡒࢀࡢศᯒ࠾࠸࡚␗࡞ࡿ⤖ᯝࡀᚓࡽࢀࡓ㸬 ඹ㏻ࡍࡿ⤖ᯝࡋ࡚ࡣ v_lxc㸦ືモࡢㄒᙡ㛵ࡍ ࡿ࢚࣮ࣛ㸧ࡀඃࢀࡓ࢚ࢵࢭࡑ࠺࡛࡞࠸ࡶࡢࢆ ศࡅࡿୖ࡛㔜せ࡞ᩥἲⓗㄗࡾ࡛࠶ࡿࡇࡀ♧ࡉ ࢀࡓ㸬 Ỵᐃᮌศᯒ࡛ࡣ㸪カ⦎ࢹ࣮ࢱࡢࡳࡢ⤖ᯝ࡛࠶ࡿ ࡀ㸪4 ✀㢮ࡢᩥἲⓗㄗࡾࡢࡳࢆ⏝࠸࡚࡞ࡾ㧗࠸ ⢭ᗘ࡛ホ౯ࢆண ࡛ࡁࡿࡇࡀ♧ࡉࢀࡓ㸬ᡂ⏣ 㸦2013㸧ࡼࡿホ౯ࡣᩥἲⓗㄗࡾ↔Ⅼࢆᙜ࡚ ࡓホ౯࡛ࡣ࡞ࡃ㸪࢚ࢵࢭయࡢホ౯࡛࠶ࡿ㸬ࡇ ࡢホ౯ࢆᩥἲⓗㄗࡾࡢࡳࢆ⏝࠸࡚ண ࡀ࡛ࡁࡿ ࠸࠺ࡇࡣ㸪࢚ࢵࢭホ౯࠾ࡅࡿᩥἲⓗㄗࡾ ࡢ㔜せᗘࡀ㧗࠸ゎ㔘࡛ࡁࡿ㸬 ᮏ◊✲⤖ᯝࡣ㸪ⱥㄒᏛ⩦⪅ࡢࣛࢸࣥࢢᣦᑟ ࡸ⮬ືホ౯࠾࠸࡚㔜せ࡞♧၀ࢆྵࡴ㸬ࣛࢸ ࣥࢢᣦᑟ࠾࠸࡚ࡣ㸪ືモ㛵ࡍࡿ▱㆑ࡀࣛࢸ ࣥࢢࡢホ౯ࡁࡃᐤࡍࡿࡇࡽ㸪ᩍᖌࡣ㸪 ࡑ࠺࠸ࡗࡓ▱㆑ࡘ࠸࡚ࡣ㸪㔜Ⅼⓗᣦᑟࢆࡍࡿ ࡇࡀồࡵࡽࢀࡿ㸬ࡲࡓ⮬ື᥇Ⅼ㛵ࡍࡿ◊✲㛤 Ⓨ࠾࠸࡚ࡣ㸪ྃᵓ㐀࡞ࡢ⤫ㄒゎᯒࡸn-gram ࢆ⏝࠸࡚ㄒ㡰㛵ࡍࡿ࢚࣮ࣛࢆ᳨ฟࡋ㸪ࡢㄗࡾ ẚ࡚㔜ࡳ࡙ࡅࡍࡿᚲせࡀ࠶ࡿࡇࡀ♧၀ࡉ ࢀࡓ㸬ཧ⪃ᩥ⊩
1) Burstein, J., Chodorow, M., & Leacock, C. Automated essay evaluation: The Criterion online writing service. AI Magazine, 25(3), pp.27–36. (2004)
2) Forman, G. An extensive empirical study of feature selection metrics for text classification. The Journal of machine learning research, 3, pp.1289-1305. (2003)
3) Jones, E. ACCUPLACER's essay-scoring technology: When reliability does not equal validity. In P. F. Ericsson & R. Haswell (Eds.), Machine scoring of student essays. (pp. 93-113). Logan, UT: Utah State University Press. (2006)
4) Kitamura, M. Influence of Japanese EFL Learner Errors on Essay Evaluation. Annual Review of English Language Education in Japan, Vol.22, pp.169-184. (2011)
5) Nagata, R. Whittaker, E., & Sheinman, V. Creating a manually error-tagged and shallow-parsed performance learner corpus. Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, pp.1210-1219. (2011)
6) ᡂ⏣┿. Konan-JIEM Ꮫ⩦⪅ࢥ࣮ࣃࢫ࠾ ࡅࡿྡモᚋ⨨ಟ㣭せ⣲ࡢศᯒ. ᮾிᅜ㝿Ꮫㄽ ྀɆゝㄒࢥ࣑ࣗࢽࢣ࣮ࢩࣙࣥᏛ㒊⦅.Vol.9,pp. 1-12. (2013)
7) Tang, L. & Liu, H. Bias analysis in text classification for highly skewed data. ICDM ’05: Proceedings of the Fifth IEEE International Conference on Data Mining, pp. 781–784. (2005)
㈨ᩱ-1 Ꮫ⩦⪅ࡢ࢚ࢵࢭࡢ㸦University Life㸧
My university life is very interesting. Because I <v_lxc crr="do">act</v_lxc> many things <prp crr="">since</prp> now. First I <uk crr="am a member of">join</uk> <at crr="a"></at> cercle. I feel <ord crr="very good about this"><prp crr="about"></prp> this very good</ord>. <uk crr="I kill time by">My killing time is</uk> writing <n_num crr="novels">novel</n_num> or drawing <n_num crr="pictures">picture</n_num>. <uk>This has many people like me</uk>.So I concentrate <prp crr="on"></prp> this. Second is summer vacation. I <v_tns crr="did"><v_lxc crr="do">act</v_lxc></v_tns> many <n_num crr="things">thing</n_num> in <pn crr="my"></pn> summer vacation. My best memory is <at crr="the"></at> seminar on the sea. I went to Ho-chi-min and Singapore. I got many friends <prp crr="from">around</prp> Hyogo university. And I <av crr="sometimes">sometime</av> meet <pn crr="them">friends</pn>. Last I have many friends <prp crr="from">in</prp> high school, junior high school and <aj crr="other">etc</aj> <n_num crr="groups">group</n_num>. We always talk about each <n_o crr="other's">other</n_o> <n_num crr="lives">life</n_num> <prp crr="by">in</prp> e-mail or internet. And We play <prp crr="">in</prp> inside or outside home. We play funny <n_num crr="games">game</n_num>. For example, <n_lxc crr="one of us">a friend</n_lxc> <v_agr crr="calls">call</v_agr> <prp crr="">in</prp> Macdonald <con crr="and"></con> <v_lxc crr="says"></v_lxc> "Please give me <at crr="a"></at> hundred <n_num crr="hamburgers">hunbergar</n_num>.“ And others look <prp crr="at"></prp> him and laugh. I have many friends, so my university life is very interesting.
㈨ᩱ-2 ࢚࣮ࣛࢱࢢࡢ
ࢱࢢ ᩥ
n_num This is the only one <n_num crr="thing"> things</n_num> you have to do. n_lxc She listened to his <n_lxc crr="speech">speak</n_lxc>.
n_o I went to <n_o crr="Nihonbashi in Osaka">Osaka Nihonbashi</n_o>. pn I took Martin and a frien of<pn crr=”his”>him</pn> to the park.
v_agr The number of students who work part-time after school <v_agr crr="has been increasing"> have been increasing</v_agr>.
v_tns I'll make researvations for the ferry as soon as I <v_tns crr="find">will find</v_tns> out the schedule.
v_lxc He wanted to <v_lxc crr="conceal">cancel</v_lxc> his guilt.
v_o If it <v_o crr="is forgotten"><v_agr crr="forgets">forget</v_agr></v_o>, plants are going to die.
mo “The phone is ringing.” “I <mo crr=”will”>’m going to </mo>answer it.” aj It was a <aj crr=”genuine”>genius</aj>diamond.
av He worked <av crr=”hard”>hardly</av>today. prp He took full advantage<prp crr=”of”>with</prp>his position.
at She is active in <at crr=”the”>a</at> development of low cost water pumps. con Clint hit a home run, <con crr=”but”>and </con> I didn’t.
rel I phoned all his friends, none of <rel crr=”whom”>who</rel> could tell me where he was. itr <itr crr=”Which”>What</itr> would you like to eat, Japanese or Chinese food? o_lxc He <o_lxc crr=”made an attempt”>had an attempt</o_lxc> at the conquest of the peak.
ord When did you buy that <ord crr=”large old brown wooden”>old brown large wooden</ord>table?
uk <uk crr=”X”>…</uk>
In case of the UK tag, correction 㸦crr=”X”㸧 is not annotated in the tag. f <f>The last day,</f>
