Tohoku University Repository (TOUR) / 東北大学機関リポジトリ

MOCHIDA, Takanori

SHINTANI, Saya

GOTO, Seon

HAMAMOTO, Toru

DOI, Ryohei

HORI, Shinsuke

OKADA, Noriaki

MATSUDA, Akira

CHENG, Weiguo

YAMASAKI, Shinichi

MAKINO, Tomoyuki

52
Poster Session
PP-5
Dynamics and Interactions of Trace Elements in Soil Solution 
and Rice under Long-Term Application of Inorganic and Organic 
Amendments in Paddy Fields
Takanori MOCHIDA1, Saya SHINTANI1, Seon GOTO¹, Toru HAMAMOTO1, Ryohei DOI2, Shinsuke HORI2, 
Noriaki OKADA2, Akira MATSUDA2, Weiguo CHENG3, Shinichi YAMASAKI4 and Tomoyuki MAKINO1
1Graduate school of Agricultural Science, Faculty of Agriculture, Laboratory of Soil Science, Tohoku University, Miyagi, 980-8572, Japan 
2Yamagata Integrated Agricultural Research Center, Yamagata, 990-2372, Japan 
3Faculty of Agriculture, Yamagata University, Yamagata, 997-8555, Japan 
4Graduate school of Environmental studies, Tohoku University, Miyagi, 980-8572, Japan
The	necessity	of	soil	 fertility	management	 in	paddy	fields	has	been	re-evaluated	 in	recent	years	due	 to	soil	degradation	and	
increasing	fertilizer	costs.	As	a	result,	 the	utilization	of	various	 inorganic	amendments	and	organic	fertilizers,	 including	sewage	
sludge,	has	gained	attention.	However,	limited	studies	have	examined	the	dynamics	and	interactions	of	trace	elements	in	soil	solution	
and	rice	under	long-term	amendment	application.
This	study	focused	on	two	long-term	amendment	plots	at	 the	Yamagata	Agricultural	Research	Institute:	(1)	a	sewage	sludge-
applied	field,	where	elemental	fluctuations	during	midseason	drainage	were	analyzed,	and	(2)	a	field	treated	with	organic	matter	and	
silica	amendments,	where	a	controlled	pot	experiment	under	continuously	flooded	conditions	was	conducted	to	assess	the	arsenic	(As)	
reduction	effect	of	silica	amendments.	Soil	solution	and	rice	samples	were	collected	and	analyzed	using	inductively	coupled	plasma	
optical	emission	spectrometry	(ICP-OES)	and	mass	spectrometry	(ICP-MS).
In	 the	sewage	sludge-applied	field,	midseason	drainage-induced	fluctuations	 in	element	concentrations	showed	similar	 trends	
within	two	groups:	Mn,	Co,	P,	Fe,	(G-A),	and	Cu,	and	S	(G-B).	Elements	in	G-A	decreased,	while	those	in	G-B	increased	during	
drainage,	suggesting	redox-driven	processes,	including	Co	adsorption/release	on	Mn	oxides	and	P	adsorption	onto	Fe	hydroxides.
In	 the	pot	experiment,	arsenic	concentrations	 in	brown	rice	decreased	by	14.7%,	11.3%,	and	27.3%,	respectively,	across	all	
treatments	with	Si	amendments.	These	 findings	confirm	 the	effectiveness	of	 long-term	silica	amendment	 in	 reducing	arsenic	
accumulation	in	rice,	highlighting	its	potential	for	sustainable	soil	management.


English

【Poster Presentation】Dynamics and Interactions of Trace Elements in Soil Solution and Rice under Long-Term Application of Inorganic and Organic Amendments in Paddy Fields

WANG, YANGYUZI

氏  名（本籍地） 王
わん
 阳
やん
雨子
ゆ う ず
 （中国） 
 
学 位 の 種 類 博 士 （経済学） 
 
学 位 記 番 号 経博 第 218 号 
 
学位授与年月日 令和 7 年 3 月 25 日 
 
学位授与の要件 学位規則第 4 条第 1 項該当 
 
研 究 科 、 専 攻 東北大学大学院経済学研究科（博士課程後期 3 年の課程） 
 経済経営学専攻 
 
学 位 論 文 題 目 Equilibrium, Stability, and Technological Change: A Multi-Sectoral 
Analysis from the Evolutionary Perspective 
               （均衡、安定性、および技術変化: 進化論的視点による多部門分析） 
 
博士論文審査委員 （主査） 
 教 授  守  健 二 教 授  黒 瀬  一 弘 
  教  授  日 置  史 郎 
  
論 文 内 容 の 要 旨 
 
In the decades following the publication of Nelson and Winter (1982)’s foundational work 
economics, an Evolutionary Theory of Economic Change in modern evolutionary, evolutionary 
economic theories have witnessed a significant transformation with the emergence and development, 
which provides a different perspective from traditional neoclassical approaches. These previous 
literature attempts to examines the theoretical foundations of evolutionary economics, tracing its 
development through multiple dimensions and exploring its implications for understanding economic 
dynamics. 
This dissertation extends Morishima (1964)’s dynamic Leontief model by incorporating 
evolutionary economic methodologies to examine the specific economic questions. Through three 
interconnected chapters, this dissertation investigates the equilibrium, ‘dual stability’ paradox, and 
endogenous technological change and market ‘selection’ mechanism. 
Chapter 2 surveys several significant concepts and approaches of evolutionary economics, which 
make significant contributions to provide a more realistic alternative to neoclassical approaches.  
Building upon these theoretical foundations, Chapter 3 explore how evolutionary concept and 
thought can be applied to resolve a longstanding issue in economic theory that traditional 
neoclassical approaches have struggled to address adequately. The stability of quantity and price 
systems has been a persistent challenge in dynamic input-output models. Unlike the static 
input-output model, this dynamic version, considering capital stocks and capital gains and losses, 
introduces significant complexities and then gives rise to three significant challenges. First, the 
quantity system exhibits instability, with output levels rapidly diverging from the dominant 
characteristic root under slight disturbances (Sargan, 1958). Second, the model suffers from ‘causal 
indeterminacy’ when initial capital stocks deviate from the balanced growth path (Uzawa, 1961; 
Dorfman et al., 1987). Third, the establishment of the price system introduces the notorious ‘dual 
stability’ issue summarized by Solow (1959) and Jorgenson (1960) as follows: in the closed dynamic 
Leontief model, if the output system is globally relatively stable, the price system is globally relatively 
unstable, and vice versa. 
Traditional neoclassical approaches attempted to address these problems through two key 
assumptions: full utilization of capacity and perfect foresight (Solow, 1953; 1959). These 
Walrasian-derived assumptions suggest that economic agents can perfectly anticipate market demand 
and future prices. However, these restrictive assumptions have been increasingly recognized as 
problematic, potentially even contributing to the dual stability issue rather than resolving it. 
Keynesian economists argue that the dual stability issue stems from unrealistic assumptions about 
capacity utilization and foresight, proposing sequential models with independent determination of 
quantity and price systems. (Jorgenson, 1960; Zaghini, 1971; Zaghini, 1991; Aoki, 1977; Filippini, 
1983). Modern classical economists, meanwhile, suggest that the problem lies in the unilateral 
determination of prices by the supply-demand system, advocating for cross-dual models where price 
and output changes interact (Flaschel and Semmler, 1986; Dumenil et a., 1987; Flaschel and Flaschel, 
2010). However, several issues warrant further discussion. The neglect of redundant capital stock and 
the absence of a comprehensive explanation for economic agents’ expectation adjustment processes 
deserve particular attention. 
This chapter aims to address these limitations by constructing a dynamic input-output model 
within an evolutionary framework. The assumption of bounded rationality and viewing the economy as 
a constantly evolving dynamic process, advocated by evolutionary economics, bring new insight and 
potential approaches to addressing these issues. Departing from neoclassical assumptions by 
introducing two key elements: incomplete utilization of capital stocks and bounded rationality in 
decision-making, this chapter makes three main contributions: First, using the bounded rationality, it 
explicitly considers redundant capital stocks in the quantity system with a ‘peculiar’ rational 
expectation function. Second, it incorporates a heterogeneous expectation function with limited 
information about future returns in the price system. Third, it demonstrates the relative stability of 
both quantity and price systems, that is, the ratio of the paths can converge to constants with 
defined limits. 
Chapter 4 proposes a theoretical framework for analysing the evolution of endogenous 
technological change through the adaptive dynamics (AD) approach within a classical input-output 
framework. It further examines the interaction between firm survival and the market ‘selection’ 
mechanism. Nelson and Winter (1982) pioneered the theory of selection in evolutionary economics, 
questioning profit-maximization justified by ‘natural selection’. However, others like Shiozawa et al. 
(2019) highlighted limitations in existing evolutionary frameworks, particularly their reliance on 
neoclassical foundations. 
Aiming to construct a more appropriate selection mechanism for evolutionary economics, this 
chapter makes several contributions. First, it introduces an ‘inner-period’ searching process to 
illustrate how homogeneous firms make their technological choice within a single period. Under 
bounded rationality assumption, the theoretical model proves that firms recursively converge to a 
temporary profit-maximizing techniques, short-term equilibrium (STE) technique, during their 
searching processes. Second, the analysis extends to examine technological innovation across 
multiple periods through the ‘inner-period’ adaptive dynamic process. It shows that while firms may 
find long-term equilibrium (LTE) technique, which are not necessarily globally ‘efficient’. Third, the 
simulation results examine the heterogeneous firms’ survival under market ‘selection’ pressure. 
Using agent-based modeling with varying market share thresholds, the results shows that high 
selection pressure (β≈ 0.1) can drive out firms with globally ‘efficient’techniques, while low 
selection pressure (β= 0) allows inefficient firms to survive. This chapter concludes that 
profit-maximizing behaviour under bounded rationality does not guarantee optimal outcomes, and 
market selection may actually hinder rather than promote efficiency under certain conditions. 
 
論 文 審 査 結 果 の 要 旨 
 
本論文の主旨は、進化経済学の知見に基づき限定的合理性を仮定して、第一に、均衡の安定性を
三つの一般性レベルにおいて検証することである。すなわち、技術所与、所与の技術集合からの技術
選択、そして技術集合の進化（技術変化）、の三段階である。さらに第二に、異種エージェンントに基
づくシミュレーションにより、市場淘汰下での均衡の最適性を検証することである。 
第１章では、進化経済学の基礎概念（不確実性、限定的合理性、適応、行動学習、淘汰、経路依存、
ロック・イン）、および進化経済学における多部門分析、期間分析、不均衡分析に関して詳細な先行研
究のサーヴェイが行われる。 
第２章では、動学的レオンチェフモデル（Dynamic Leontief Model、以下 DLM）に基づいて、所与の
技術における均衡の収束・安定性が検証される。価格体系と数量体系との間で、均衡の相対的安定
性が両立しないといういわゆる「dual stability paradox」がDLMに内在する問題として 1960年代から提
起されてきたが、本章は、価格の完全予見と資本ストックの完全利用の仮定を緩和し、限定的合理性
を仮定することによってパラドックスが解決されることを証明した。 
第 3 章では、引き続き DLM に基づき、まず、所与の技術集合から技術選択が行われる場合の技術
の短期均衡の収束・安定性が考察され、逐次模索的な技術選択が部分列収束することが示される。次
に、不完全予測の下、技術集合自体が時間の経過とともに経路依存的に変化する場合の技術の長期
均衡の収束・安定性が考察され、ここでも時系列的な技術変化がやはり部分列収束することが示され
る。 
第 4 章では、異種のエージェントと市場淘汰（最低市場シェア）を導入して、シミュレーションを行い、
最適解への収束に関する反例を提示している。 
本論文は、限定的合理性の仮定の下で各種均衡（価格、数量、技術）の収束可能性について手堅
く首尾一貫した証明を与えることに成功しており、従来のシミュレーション的方法に加えて進化経済学
に一般的、理論的基礎付けを与えるという課題に着実に貢献するものと考えられる。 
以上より、審査委員一同、本論文は博士（経済学）の学位を授与するに十分な知識と能力を有して
いると判定した。 


Equilibrium, Stability, and Technological Change: A Multi-Sectoral Analysis from the Evolutionary Perspective

Tohoku University博士（農学）博士学位論文 (Thesis(doctor))thesi

殷,子鈞

Japanese

自然資源管理に対する意識と評価に関する研究　−国立公園と野生生物を対象に–

河原,輝宙

哺乳動物における加齢にともなう精子幹細胞機能の維持と低下

川端, 望

刊行のことば
　増田聡教授は，2024年 3月 31日をもって本学を退職されました。増田教授の長年にわたる本学
での研究・教育などにおける業績を称えるために，本誌研究年報『経済学』の第 82巻第 1号を，
増田教授のご退職を記念する特別号として刊行いたします。
　増田教授は，1982年 3月に東京大学工学部都市工学科を卒業され，1987年 3月に東京大学大学
院工学系研究科都市工学専攻博士課程を修了，工学博士の学位を授与されました。そして，株式会
社三菱総合研究所を経て，1990年 10月に東北大学教養部に講師として着任され，教員としての第
一歩を踏み出されました。その後 1993年 4月に東北大学情報科学研究科助教授，1997年 4月に東
北大学経済学部助教授（地域計画担当），1999年 4月に大学院経済学研究科助教授を経て，2000年
4月に同教授に昇任し，以来 2024年 3月末に退職するまで，長きにわたり研究，教育，管理運営
に携わられ，本研究科を支え続けてこられました。
　増田教授のご研究は多岐にわたりますが，大きくは防災型土地利用をめぐるまちづくりや自治体
政策に関する研究，および東日本大震災後の地域産業・地域社会の復興に関する研究の二つに分け
られます。第一の分野の研究は，地域の防災計画と自治体の都市計画を連携させるために必要なプ
ロセスと現状を分析した試みであり，従来の都市計画論にとっては異分野であった地形学や防災研
究等の成果を取り入れて，新たな分析視角を提示しました。第二の分野の研究は，東日本大震災後
の被災自治体における復興計画や復興事業に関する一連の著作です。これらは，広域的災害から東
北の経済・社会をいかに復興させるかという課題に対して，コミュニティ計画やまちづくりの観点
から，調査，分析，および政策提言を行ったものです。このように，増田教授の研究は，都市計画
や防災に関する工学的な分析手法を，土地利用や復興事業等，現実の政策課題に関する社会科学的
研究に応用したものと言えます。根拠に基づく客観的で精密な議論を数多くの著書，論文，調査報
告等で展開されたことで，わが国の地域経済・地域経営論の発展に大きく寄与されました。
　また，増田教授は，東北都市学会会長，日本都市学会理事等の要職を歴任し，都市工学や都市計
画学に関する学会の運営や発展に大いに貢献されました。学内では，2010年 4月から 2011年 3月
まで大学院経済学研究科経済経営学専攻長，2015年 4月から 2017年 3月まで大学院経済学研究科
副研究科長を歴任し，管理運営面でも多大な貢献をされました。特に，2011年の東日本大震災後
から 2024年 3月まで本研究科に設置された「震災復興研究センター」では，センター長として震
災後の地域の復興モデルや都市計画の作成に尽力されました。
　最後になりましたが，増田教授が今後ともますますご壮健で活躍されることを祈念すると共に，
私たち後進に対し，引き続き温かいご指導・ご鞭撻を賜りますよう心よりお願い申し上げる次第で
す。
　2024年 3月 31日
 東北大学大学院経済学研究科長　　川　端　　　望　


刊行のことば（増田聡教授退職記念号）

庄子, 真岐

震災伝承施設の保存と活用に関する市民評価の分析 ―震災遺構仙台市立荒浜小学校を事例として―

中嶋, 紀世生

中村, 哲也

西川, 芳昭

農山村の活性化に向けた地域支援型農業の可能性　―日仏比較分析からの考察―

高千穂, 安長

被災度が大きい小規模自治体での災害教訓伝承の現状と課題　―宮城県女川町の事例―

Tohoku University博士（生命科学）thesi

Cui, Ying

Analysis of the molecular mechanism in host-mediated early senescence of nodules triggered by Bradyrhizobial effector proteins

李, 云蒙

 
 1 
 
氏 名  李
LI
 云
YUN
蒙
MENG
 
学 位 の 種 類  博 士（情報科学） 
学 位 記 番 号  情 博 第  ８５６  号 
学位授与年月日  令和７年 ３月２５日 
学位授与の要件  学位規則第４条第１項該当 
研究科、専攻  東北大学大学院情報科学研究科（博士課程）システム情報科学専攻 
学位論文題目  Chat Translation Evaluation: From Datasets to Metrics 
         （チャット翻訳評価： データセットから評価指標まで） 
論文審査委員  （主査）東北大学教 授 鈴木 潤 
             東北大学教 授 周 暁   東北大学教 授 長谷川 剛 
             東北大学教 授 乾 健太郎 東北大学准教授 坂口 慶祐 
 
論 文 内 容 の 要 旨 
 
Chapter 1 Introduction 
 
This section introduces the main research goal and the contributions of the thesis. 
 
Recent developments in neural machine translation (NMT) have demonstrated notable 
improvements in the performance of machine translation systems, especially in tasks involving 
the translation of formal documents such as news articles and academic papers. 
However, chat data often include culturally specific language, including idioms, proverbs, and 
local expressions, necessitating contextual understanding and sometimes external knowledge 
beyond direct translation. Multiple speakers in chat scenarios present additional complexities in 
translation tasks. These characteristics hinder machine translation models in preserving 
accuracy and fluency. 
 
A crucial step in improving the performance of translation models when translating chat is to 
investigate the need for evaluating chat translations and critically assess the quality and 
limitations of these models through evaluation tasks. Effective evaluation metrics for chat 
translation should include nuance, stylized content related to the source fragment, dialog 
consistency, colloquialisms, and cultural references. 
 
This thesis introduces a series of studies from building an assessment dataset, developing a 
user-centered assessment methodology, and finally proposing robust assessment metrics tailored 
for chat translation.  
 
 
 2 
Chapter 2 Backgrounds 
 
This section introduces the related works of the thesis. 
 
Machine translation (MT) has progressed through various phases. The advent of neural machine 
translation (NMT), particularly with the development of the transformer model, marked a 
significant leap forward, enhancing machine translation capabilities through the use of attention 
mechanisms that yield more accurate translations. 
 
The inaugural workshop concentrating on chat translation was the 5th Conference on Machine 
Translation (WMT20), establishing foundational work for the field. As research progresses in this 
field, WMT and its derivative studies increasingly acknowledge the significance of source content 
preservation and the retention of speaker style in chat translation. 
 
Numerous studies have demonstrated that large language models (LLMs), such as GPT, perform 
well in translation tasks, particularly in literary translation scenarios. These studies suggest that 
LLM translations may be preferable to traditional NMT models. 
 
Traditional evaluation metrics, such as BLEU, which utilizes single character alignments, both 
derive evaluation scores based on the textual similarity between model output and reference text. 
The Multidimensional Quality Metrics (MQM) framework offers a comprehensive and flexible 
method for refining translation quality assessment tasks.  
 
Chapter 3  
 
This section introduces the first study of the thesis. 
 
This study proposed a mistranslation warning message during cross-language chat scenarios and 
conducted a survey to explore how these warnings assist communication. Participants were 
presented with a simulated cross-language chat situation and tasked with selecting the most 
appropriate response from three options. Whenever a translation error occurred, the system 
displayed a warning message. 
 
As a result, the survey answers gathered through crowdsourcing showed that the warning 
messages help to reduce the confusion in cross-lingual chats. Additionally, the survey identified 
features desired by users for these warning messages. The outcomes of this study will be 
instrumental in developing assistive features that detect and inform users about translation 
errors in chat interactions. 
 
 
 
 3 
Chapter 4 
 
This section introduces the second study of the thesis. 
 
The availability of parallel datasets suitable for chat translation is quite limited. 
To address this gap, we created a Japanese-English parallel chat dataset named the Bilingual 
Persona Chat (BPersona-chat) by filtering, translating, and evaluating existing monolingual data 
from Persona-chat and JPersona-chat. To ensure quality, we focused on selecting clear and 
coherent chat from the Persona-chat dataset through a crowdsourcing approach. Professional 
translators were asked to give accurate translations according to the personalities provided in 
each piece of data. Two machine translation models were used to generate translations as well. 
Human annotators assigned the binary label for each chat translation to evaluate the quality. 
 
Chapter 5 
 
This section introduces the third study of the thesis. 
 
For situations where participants cannot communicate in the same language, a refined error 
detection system is expected to alert them to potential errors, guiding them to adjust their texts, 
thus minimizing translation issues in multilingual chats. We developed classifiers to evaluate the 
performance of generated translations. The primary task is to predict whether a given utterance 
includes translation errors. 
 
Assuming neither speaker knows the other's language, in this research, we define chat as a 
two-sentence dialogue between two individuals communicating in different languages, thereby 
simplifying the task. 
 
Upon training our models and applying the classifiers to the translations, we found that the 
classifiers exhibited different advantages based on the type of information they accessed. 
Although some errors persisted, we could primarily employ these classifiers to identify chat 
translations characterized by coherence and accuracy together. 
 
Chapter 6 
 
This section introduces the fourth study of the thesis. 
 
When evaluating chat translations, it is essential to consider grammatical and lexical correctness, 
the conversational nuances and stylistic elements specific to the source material, and the 
consistency within the dialogue. In this study, we introduce the Multidimensional Quality Metrics 
for Chat Translation (MQM-Chat) to address the unique challenges of evaluating chat 
 
 4 
translations.  
Building on the established Multidimensional Quality Metrics (MQM) framework, MQM-Chat 
includes seven specific error types: Mistranslation, Omission or Addition, Terminology or Proper 
Noun Errors, Unnatural Style, Ambiguity and Disambiguation, Buzzword or Loanword Issues, 
and Dialogue Inconsistency.  
 
We applied MQM-Chat to evaluate the translation capabilities of five advanced models: the large 
language models (LLMs) GPT-4 and LLaMA3, the commercial translation model DeepL, the 
multilingual model by Facebook used in WMT21, and the bilingual model developed by team 
SKIM for WMT23. 
 
To validate the effectiveness of MQM-Chat, we compared it with the standard MQM framework. 
We collected human annotations for sampled data using standard MQM and assessed the 
differences between the two evaluative approaches.  
Our findings indicate that MQM-Chat offers fine-grained classification, identifying errors 
categorized under standard MQM into more specific labels, with around 30\% reflecting 
chat-specific challenges such as Ambiguity and Disambiguation, Buzzword or Loanword Issues, 
and Dialogue Inconsistency. 
 
We evaluated five state-of-the-art translation models using MQM-Chat to analyze their handling 
of chat content. This research facilitates the creation of annotated Chinese-to-English and 
Japanese-to-English chat translation data. Ultimately, the contributions of this study aim to 
deepen the understanding of chat translation and provide valuable resources for future 
advancements in the field. 
 
Chapter 7 Conclusion 
 
This section concludes the thesis. 
 
This thesis makes significant contributions to improving the evaluation of chat translation by 
developing specialized datasets, creating error detection mechanisms, and introducing 
multi-dimensional metrics.  
 
By establishing metrics and resources, this thesis not only enhances future developments in chat 
translation evaluation. Additionally, the implications of this research extend beyond chat 
translation, potentially serving as a model for evaluating other informal content types, such as 
user-generated reviews, social media posts, and community discussions.  
This thesis supports the overarching objective of developing machine translation systems that are 
responsive to real-world communication needs. 
 
論 文 審 査 結 果 の 要 旨
本論文は、対話翻訳（チャット翻訳）の評価タスクに関して、データセットの構築から評価基準の
設計まで、幅広く検討するものである。本論文は一連の研究成果をまとめたもので、全 8 章から構成
されている。
第１章は序論として、対話翻訳の重要性ならびに現状を概観し、現行のシステムが直面している課
題について説明している。
第２章では、対話翻訳分野のこれまでの研究の歩みと現在の研究動向を整理し、本研究が取りくむ
べき課題を位置づけている。
第３章では、アンケート調査を通じて対話翻訳研究の必要性とユーザーの需要を統計的に検証する
ともに、対話翻訳およびその評価が持つ可能性について分析している。
第４章では、本研究により新たに構築された対話翻訳評価データセット「BPersona-chat」につい
て、その設計目標やデータ収集方法を詳細に説明している。本データセットは、当該分野のさらなる
進展を促進する貴重なリソースとして高く評価できる。
第５章では、正確性と一貫性の両方の観点から対話翻訳を評価する二値分類モデルを新たに提案
し、対話翻訳評価の新たな枠組みを詳述している。本モデルは対話翻訳評価のさらなる発展に有用な
基盤を提供している。
第６章では、対話翻訳タスクの新しい評価指標である「Multidimensional Quality Metrics for Chat 
Translation (MQM-Chat)」を提案し、その設計意図や導出過程を説明している。さらに、この章では、
提案指標を用いた実際の評価事例と得られたアノテーション結果の分析を通じ、現在の対話翻訳が内
包する困難と課題を詳らかに明らかにしている。
第７章では結論として、全章を総括し、研究の到達点を示すとともに結論を述べている。
第８章では研究対象者の個人情報、プライバシー保護などの倫理的配慮を述べている。
以上、本論文は対話翻訳を対象にデータセットの構築から新たな評価基準の策定に至るまで、多面
的に分析し、具体的な洞察を提示するものであり、情報科学の発展に寄与するところが少なくない。
よって、本論文は博士（情報科学）の学位論文として合格と認める。


チャット翻訳評価： データセットから評価指標まで

Not a member yet