東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

FindBook

Google Book

Amazon

博客來

Natural Logic and Natural Deduction for Reasoning About Natural Language.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Natural Logic and Natural Deduction for Reasoning About Natural Language./
作者:	Binhadba, Ghadah.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2021,
面頁冊數:	277 p.
附註:	Source: Dissertations Abstracts International, Volume: 83-09, Section: B.
Contained By:	Dissertations Abstracts International83-09B.
標題:	Trees. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=29002675
ISBN:	9798209781318

Natural Logic and Natural Deduction for Reasoning About Natural Language.
Binhadba, Ghadah.

Natural Logic and Natural Deduction for Reasoning About Natural Language. - Ann Arbor : ProQuest Dissertations & Theses, 2021 - 277 p.

Source: Dissertations Abstracts International, Volume: 83-09, Section: B.

Thesis (Ph.D.)--The University of Manchester (United Kingdom), 2021.

This item must not be sold to any third party vendors.

One way of measuring a natural language processing system's semantic capacity is by demonstrating that it can handle natural language inference (NLI). Automating the task of NLI requires natural language text to be converted into a meaning representation (MR) on which inference procedures can be applied. Finding a balance between the expressive power of a MR and the inferential capabilities that could be conducted using such representation is a challenge. In that respect, the aim of this thesis is to rea- son about deep semantic phenomena (such as defaults, monotonicity, quantifiers and propositional attitudes) and thus construct a MR that is able to preserve the subtle semantic distinctions people make when reasoning about such phenomena. For decades, the best way for obtaining such semantically deep MRs was by translating NL texts into some formal language such as first-order logical formulas. However, such transla- tions have proven difficult. Alternatively, and motivated by the success of natural logic systems on the pairwise entailments tasks, in which MRs are close to NL texts' surface form, we have investigated the use of a first-order logic theorem prover (to allow reasoning over multi-premises tasks) and have adapted it to work on representations that were built based on syntactical analysis (dependency trees) of NL texts. To ensure the right depth of representation, we have investigated the literature to learn about the intended semantic phenomena and hence model them in our MR accordingly. To measure such an approach's performance, we have implemented an inference system that consists of three parts: a dependency grammar to generate syntactical trees, a tree normalizer to build the desired MRs (namely inference friendly forms) and a theorem prover (CSATCHMO+) along with some necessary background information. To our knowledge, there is not much work that has been done in this line of research (com- bining theorem proving with a version of natural logic). Therefore, we have tested our system's performance on the part of a test-set that has a clear representation of semantic phenomena (The FraCas test-set) and compared it to the literature's related systems. Overall, the current findings encourage further investigation extending to other phenomena, as we obtained a comparable result when compared to the related natural logic systems and an only just better result than a natural logic theorem prover.

ISBN: 9798209781318Subjects--Topical Terms:

516384
Trees.

Natural Logic and Natural Deduction for Reasoning About Natural Language.
LDR:03561nmm a2200337 4500 001 2351202
005 20221107085322.5
008 241004s2021 ||||||||||||||||| ||eng d
020 $a 9798209781318
035 $a (MiAaPQ)AAI29002675
035 $a (MiAaPQ)Manchester_UK3b18088a-5696-426d-a50e-f99102f6efad
035 $a AAI29002675
040 $a MiAaPQ $c MiAaPQ
100 1 $a Binhadba, Ghadah. $3 3690758
245 1 0 $a Natural Logic and Natural Deduction for Reasoning About Natural Language.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2021
300 $a 277 p.
500 $a Source: Dissertations Abstracts International, Volume: 83-09, Section: B.
500 $a Advisor: Ramsay, Allan.
502 $a Thesis (Ph.D.)--The University of Manchester (United Kingdom), 2021.
506 $a This item must not be sold to any third party vendors.
520 $a One way of measuring a natural language processing system's semantic capacity is by demonstrating that it can handle natural language inference (NLI). Automating the task of NLI requires natural language text to be converted into a meaning representation (MR) on which inference procedures can be applied. Finding a balance between the expressive power of a MR and the inferential capabilities that could be conducted using such representation is a challenge. In that respect, the aim of this thesis is to rea- son about deep semantic phenomena (such as defaults, monotonicity, quantifiers and propositional attitudes) and thus construct a MR that is able to preserve the subtle semantic distinctions people make when reasoning about such phenomena. For decades, the best way for obtaining such semantically deep MRs was by translating NL texts into some formal language such as first-order logical formulas. However, such transla- tions have proven difficult. Alternatively, and motivated by the success of natural logic systems on the pairwise entailments tasks, in which MRs are close to NL texts' surface form, we have investigated the use of a first-order logic theorem prover (to allow reasoning over multi-premises tasks) and have adapted it to work on representations that were built based on syntactical analysis (dependency trees) of NL texts. To ensure the right depth of representation, we have investigated the literature to learn about the intended semantic phenomena and hence model them in our MR accordingly. To measure such an approach's performance, we have implemented an inference system that consists of three parts: a dependency grammar to generate syntactical trees, a tree normalizer to build the desired MRs (namely inference friendly forms) and a theorem prover (CSATCHMO+) along with some necessary background information. To our knowledge, there is not much work that has been done in this line of research (com- bining theorem proving with a version of natural logic). Therefore, we have tested our system's performance on the part of a test-set that has a clear representation of semantic phenomena (The FraCas test-set) and compared it to the literature's related systems. Overall, the current findings encourage further investigation extending to other phenomena, as we obtained a comparable result when compared to the related natural logic systems and an only just better result than a natural logic theorem prover.
590 $a School code: 1543.
650 4 $a Trees. $3 516384
650 4 $a Popular music. $3 555592
650 4 $a Logic. $3 529544
650 4 $a Semantics. $3 520060
650 4 $a Linguistics. $3 524476
650 4 $a Music. $3 516178
650 4 $a Wood sciences. $3 3168288
690 $a 0395
690 $a 0290
690 $a 0413
690 $a 0746
710 2 $a The University of Manchester (United Kingdom). $3 3422292
773 0 $t Dissertations Abstracts International $g 83-09B.
790 $a 1543
791 $a Ph.D.
792 $a 2021
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=29002675