格式: AZW3, DOCX, EPUB, MOBI, PDF, TXT
形态差异与变化不仅会影响语序，还是决定主语脱落的重要因素。一致词缀贫乏的语言不允许任何主语脱落，一致词缀丰富的语言则允许脱落。在一致词缀丰富的语言中，主语能否脱落与带有一致词缀的动词的指示性强度有关。动词的指示性强度可分为三类：完全指示（full-referential）、半指示（quasi-referential）和无指示（non-referential）；句子主语亦相应分为三类：完全指示主语（即主题主语thematic subject），半指示主语（虚主语，如it is raining中的it）及无指示主语（虚主语，如it seems that he has gone中的it）。有些语言，三种主语都允许脱落；有些语言只允许后两种主语脱落；而有些语言只允许最后一种主语脱落。
List of Abbreviations
The purpose of this dissertation is to study the morphological effects on word order change and subject omission through a systematic analysis of morphological evolution. Morphological change concerns two respects，morphological case on nouns and agreement inflection on verbs. We propose that both are the major factors determining word order change and subject omission.
In the analyses of word order，we propose that in agreement-rich languages affixation takes place in syntax，and that in agreement-impoverished languages affixation takes place in the phonological component via affix lowering. Agreement-rich languages can be divided into two subtypes：C-Agr languages which have complementizer agreement，and T-Agr languages which have no complementizer agreement. In C-Agr languages，verb movement for affixation is up till C，whereas in T-Agr languages，verb movement for affixation is up to T.
During morphological development，the referential power of agreeing verbs gradually gets lost. The loss of the referential power causes the disability of covering the content of a null subject and thus causes the loss of empty topics，i.e. the loss of VSO sentence pattern. Moreover，along with the gradual loss of morphological cases，the OV order is replaced by the VO order and the nominative phrase is fixed to the typical subject position，［Spec，TP］. The combination of the three syntactic changes results in SVO to become the predominant sentence pattern in a language.
In addition to the influence of word order variation，change of verbal agreement morphology also has effects on null subject permission：Licensing of null subjects is determined by the strength of the referential power of agreeing verbs. During the fusional process of agreement markers，the referential power of agreeing verbs gradually gets lost from full-referential to quasi-referential and then to non-referential. Correspondingly，subject omission gradually becomes more restricted：some languages allow three types of null subjects；some allow only null expletives；and some other languages allow only null non-referential subjects. As for languages with impoverished verbal agreement morphology or without agreement at all，there is not any null subject. This is related to the loss of empty topics，which causes the introduction of overt expletives，and the fixity of SVO as the predominant sentence pattern.
该论文的目的是通过对形态变化做历时性分析来研究形态变化对语序及主语脱落的影响。形态变化包括两个方面：形态格（morphological case）和动词的性数一致词缀（verbal agreement inflection）。这两者都是语序变化及主语脱落的决定性因素。
在语序分析中，我们提出以下假设：在有着丰富的表示性数一致的动词词缀的语言（agreement-rich languages）中，词缀通过显性动词移位附着在动词上；而在表示性数一致的词缀贫乏的语言（agreement-impoverished）中，词缀是通过在音系部分（phonological component）下移（affix lowering）附着在动词上的。有着丰富的一致词缀的语言又可根据功能语类Agr的嫁接位置分为两种：C-Agr 语言和T-Agr语言。在C-Agr 语言中，性数一致体现在C上；而在T-Agr语言中，性数一致体现在T上。换言之，在C-Agr 语言中，Agr嫁接在C上，因此动词移位至C以获取词缀；而在T-Agr语言中，Agr嫁接在T上，动词只移位至T以获取其词缀，其动词至C的移位是受［Topic］特征的驱使。
随着形态的发展变化，带有一致词缀的动词（agreeing verbs）的指示性（referential power）逐渐消失，这就导致空主题（empty topic）的消失。也就是说，主题位置必须有个句子成分填充，这意味着VSO句式的消失。另外，随着形态格的消失，主语的位置被固定在［Spec，TP］，并且只能由主格主语来填充。这就导致动词至C的移位的消失，从而使得SVO成为主要句式。
性数一致形态的变化不仅会影响语序，而且是决定主语脱落的重要因素。主语能否脱落是由带有一致词缀的动词的指示性的强度决定的。在性数标记（agreement marker）和动词词干相溶合的过程中，带有一致词缀的动词的指示性的强度逐渐削弱，从完全指示（full-referential）到半指示（quasi-referential）再到无指示（non-referential）。主语脱落也相应地受到更多限制：在有些语言中，三种主语（完全指示主语，半指示主语及无指示主语）都允许脱落；有些语言只允许后两种主语脱落；而有些语言只允许最后一种主语脱落。主语脱落的不同情况只在有着丰富的一致词缀的语言（agreement-rich languages）中加以区分，一致词缀贫乏的语言是不允许任何主语脱落的。我们认为这种限制与空主题消失及SVO成为主要句式有关。
Contemporary syntactic research can be done from two main perspectives，a formal perspective and a morphological perspective. In the formalist Chomskian model，language variation is reduced to choice of parametric values of universal principles，and parameters are restricted to formal features of functional categories（Chomsky，1995）. In contrast to the formal feature hypothesis，Rohrbacher's（1999）morphology-driven syntax is from a morphological perspective. He proposes that all syntactic parameters should be set exclusively on the basis of the phonetically perceptible content of functional categories. As Rohrbacher（1999：4）states：“… argumentation in the Principles and Parameters framework tends to fall into circularity：functional features ‘explain’ syntactic phenomena while at the same time these syntactic phenomena are often the only evidence for the relevant functional features，given that the latter can be entirely abstract.” In other words，there should be morphological evidence for the fixity of the value of any syntactic parameter.
In this dissertation，we will make a syntactic study from a morphological perspective，in order to figure out the morphological effects on syntactic variation in languages. Our syntactic study covers two aspects，word order variation and subject omission. Since human languages are in continuous change under the influence of other languages or dialects and the affection of first and/or second language acquisition，we propose that a synchronic analysis within and/or across languages is not enough to provide a profound explanation of complex language differences；rather，it is necessary to do a generalized morphological analysis from both synchronic and diachronic perspectives. In a word，we suggest that language variation is caused by morphological change.
The main questions that we will focus on are：（i）Which factors cause verb movement and word order variation？（ii）What are the significant licensing conditions of null subjects？ We propose that both morphological cases of nouns and verbal agreement inflections are the major factors that determine word order change and subject omission.
The dissertation consists of six chapters. The first chapter is a review of some relevant morphological theories on word order and subject omission. Chapter 2 is an introduction to the theoretical background and our main proposals in this dissertation. Many typologists claim that most verbal inflections are developments of free pronouns：Full lexical words may be reduced to pronouns or particles，which in turn may become clitics attached to other words，and subsequently be re-analyzed as affixes. Under the conflicts between phonological naturalness and morphological naturalness，agreement markers and verbs tend to fuse，and then inflectional affixes gradually become assimilated via the analogy operation. Our study in this dissertation mainly depends on morphological comparison；or，on classification of languages in terms of morphological richness. In the analyses of word order variation，we classify languages into two types in terms of the richness degree of verbal agreement morphology：languages with rich agreement morphology and languages with impoverished agreement morphology. Agreement-rich languages have overt verb movement for affixation；while in agreement-impoverished languages affixation takes place in the phonological component，via affix lowering. Agreement-rich languages can be divided into two subtypes：C-Agr languages which have complementizer agreement，and T-Agr languages which have no complementizer agreement. In C-Agr languages，verb movement for affixation is up till C，whereas in T-Agr languages，verb movement for affixation is up to T.
We propose that during morphological development the referential power of agreeing verbs gradually gets lost. This factor works not only in the loss of empty topics but also in the restriction of null subjects. The loss of referential power causes the disability of covering the content of a null subject and thus in turn causes the loss of empty topics，i.e. the loss of VSO sentence pattern. Moreover，along with the gradual loss of morphological cases on nouns，the nominative phrase is fixed to the typical subject position，［Spec，TP］. Therefore，SVO becomes the predominant word order.
We further argue that null subject phenomena，like verb movement，are closely related to morphology：licensing of null subjects is determined by the strength of the referential power of agreeing verbs. At the start of fusion，agreement markers have not yet lost the referential power. In a language with such distinct agreement markers，all the three types of null subjects，i.e. full-referential，quasi-referential and non-referential null subjects，are permitted to occur. During further development of agreement inflections，this referential power gradually gets lost：from full-referential to quasi-referential and then to non-referential. Correspondingly，subject omission gradually becomes more restricted：some languages allow all the three types of null subjects；some allow only null expletives，and some other languages allow only null non-referential subjects. As for languages with impoverished verbal agreement morphology or without agreement at all，there is not any null subject. We propose that this restriction is related to the loss of empty topics，which causes the introduction of overt expletives，and the fixity of SVO as the predominant sentence pattern.
In the next three chapters，we will present evidence for the morphological effects on word order and subject omission：Chapter 3 is an illustration of word order variation in the Romance languages；Chapter 4 is an analysis of word order differences in the Germanic languages；and Chapter 5 is a demonstration of morphological effects on null subject phenomena. The data analyses in the three chapters will prove that language variation in word order and subject omission can be explained by morphological changes of nominal cases and agreement inflections.
Our morphological study in this dissertation can solve the syntactic problems left by some generativists，including the syntactic differences among the Romance languages and the Germanic languages and the licensing conditions of null subjects in languages with different morphological richness，like German，Italian，Icelandic，etc. Moreover，our study provides a rejection to Y. Huang's（1994，2000）conclusion that null subject phenomena may not be explained on the syntactic level.
Chapter 1 A Study of the Contemporary Morphological Researches on Syntax
In this chapter，we will introduce some of the contemporary morphological researches on word order and subject omission. We will show that in the analyses of word order variation，most of the contemporary morphological studies are from a synchronic perspective and generally focus on one single aspect of morphology；therefore，they cannot cover all the syntactic differences among human languages. In our opinion，a diachronic comparison of morphological paradigms is quite important in the analysis of syntactic variation.
Moreover，we will show that a single or a few syntactic parameters，as proposed in many generativists' frameworks（cf. Chomsky 1981，1982；Jaeggli and Safir 1989），cannot cover all the null subject phenomena，because there are three types of null subjects and the licensing condition of each kind of null subject is different. Compared to the generativists' studies，an analysis based on morphological comparison，like Vainikka and Levy's（1999）analysis of the pronominal connection，appears to be more flexible in that different licensing conditions of subject omission are even allowed to be present in one language. Nevertheless，their null subject analysis only considers null thematic subjects，without consideration of null expletives，which is actually a hotly debated topic in syntax.
1.1 Contemporary Morphological Researches on Word Order
Rohrbacher（1999：7）claims that syntax is driven by morphological factors，i.e.，all syntactic parameters are set exclusively on the basis of the phonetically perceptible content of functional categories. He takes V-to-I movement as the issue for the explanation of his morphology-driven syntax model and proposes that movement is permitted only when a landing site for the movement is introduced in a given language，while the existence of such a landing site is determined by morphological realization of relative features in this language. The following subsection is an introduction to his strong agreement hypothesis.
22.214.171.124 Strong agreement hypothesis
Rohrbacher's（1999）morphology-driven syntax is a strong agreement hypothesis which states that rich agreement and only rich agreement causes V-to-I movement：V-to-I movement occurs in exactly those languages which distinctively mark the person features［1st］and［2nd］in either the singular or the plural of at least one tense. Whether the person features［1st］and［2nd］are distinctively marked can be determined only by looking at the entire paradigm in which they appear. When a paradigm of inflectional affixes distinctively marks these features，the paradigm is referential since the affixes can unambiguously identify the subject of the clause as referring to the speaker（s），the hearer（s）or other（s）. Like other referential elements，the referential inflectional affixes are listed in the lexicon and inserted under Infl at D-structure. Since they must be affixed to a verb at S-structure，they trigger V-to-I movement，i.e.，the verb moves from V to I in order to pick up the affix（es）. When an inflectional paradigm does not distinctively mark［1st］or［2nd］，it is non-referential since its affixes cannot unambiguously determine the referential value of the subject. Such a paradigm is therefore not listed in the lexicon，and its affixes are not active in the syntax. Instead，they are only later affixed directly onto the verb by PF spell-out rules.
Since the referential inflectional affixes are listed in the lexicon，they are certainly separate lexical items which are ‘visible’ to the syntax and which are capable of projecting their own structures. Based on this view，Rohrbacher（1999）proposes that the presence of inflectional affixes，specifically distinct affixes for the 1st and 2nd person plural，determines whether or not an AgrP will be projected in the derivation of a sentence. That is to say，only languages with distinctive marking of these person features have［Spec，AgrSP］. The presence of AgrP，in turn，renders the possibility of V-to-I movement by providing a landing site for the verb to move to. If the language does not have distinct inflectional features，then no AgrP will be projected and the verb will remain in situ.
Rohrbacher's（1999）［1st］and［2nd］person features hypothesis is based on Beard's（1991）theory.In Beard's（1991）framework，in languages where a person feature is represented by both overt and non-overt morphology，overt morphology represents the marked value of the feature and zero morphology represents the unmarked value of the feature. The systematic absence of overt morphology for the third person（singular）in many languages then suggests that third person is universally the unmarked value of the person feature. This view is translated into the assumption that ‘person’ falls into two features，［1st］and［2nd］，whose positive values are marked and whose negative values are unmarked. Along these lines，Rohrbacher（1999）proposes that the person features［1st］and［2nd］are responsible for V-to-I movement.
126.96.36.199 Bobaljik's（2002）view on the strong agreement hypothesis
Rohrbacher's strong agreement hypothesis turns to be untenable under Bobaljik's（2002）rejection with empirical evidence from Faroese. Bobaljik（2002）argues that a theory like Rohrbacher's，which relies uniquely on overt morphological variation to drive syntactic variation，is insufficient. Syntactic variation can be attested in the absence of morphological variation. In some languages，such as Faroese，there are two syntactic varieties，but there is no notable morphological difference. In Faroese，both verb movement and verb-in-situ word orders are possible，as shown in（1）.（1）a. Tey nýttu fleiri orð，sum hon hevði ikki hoyrt fyrr.they used several words which he had not heard beforeb. Tey nýttu fleiri orð，sum hon ikki hevði hoyrt fyrr.they used several words which he not had heard before‘They used several words which he had not heard before.’（from Barnes 1987，cited in Bobaljik 2002：10）
Importantly，despite the variation in word order in Faroese，there are no relevant morphological differences. The basic verbal paradigm of Faroese is given in（2）.
The morphological paradigm in（2）shows that Faroese morphology is poor by Rohrbacher's criteria and hence，according to Rohrbacher（1999），the language will have no overt verb movement，which is contrary to the fact. Therefore，Faroese constitutes a counterexample to Rohrbacher's strong version of a morphology-driven syntax. Since syntactic variation does exist in Faroese in the absence of a morphological trigger，it follows that morphological variation is not the only trigger of syntactic variation. Consequently，Bobaljik（2002）concludes that Rohrbacher's strong version of morphology-driven syntax must be replaced by a weaker one：If a language has rich inflection then it has verb movement to Infl. Verbal inflection is rich if finite verbs may bear multiple distinct inflectional morphemes.（2）Faroese：kasta ‘throw’（from Lockwood 1964：76，cited in Bobaljik 2002：10）
Contrary to Rohrbacher（1999）and most other linguists，Bobaljik（2002）proposes that it is syntactic（structural）differences that underlie morphological differences，and not the other way around. Furthermore，it is differences in syntactic structure，and not morphological differences，that drive variation in verb movement. In a word，syntactic structure variation in languages causes not only morphological differences but also syntactic variation，like the permission or disallowance of multiple subject positions.
188.8.131.52 Effects of syntactic structure variation
Bobaljik's（2002）hypothesis of syntactic structure variation is based on Iatridou's（1990）view that the inventory of functional projections in the IP domain could be subject to cross-linguistic variation，i.e.，UG would allow two possibilities of IP，as shown in（3）（cited in Bobaljik 2002：17）. In other words，the presence of functional categories may vary in languages.（3）
Bobaljik（2002）suggests that the difference between（3a）and（3b）constitutes the Split Infl Parameter（SIP）：Morphologically rich languages like Icelandic would have the rich structure in（3b），while morphologically impoverished languages like Danish and English are examples of languages with the simple IP structure in（3a）. On the assumption that no more than one single overt morpheme may be inserted into a single syntactic terminal node，it follows that languages with a single inflectional head will never permit more than one overt inflectional morpheme on a single verb. That is to say，if a language has the simple syntactic structure in（3a），it then would have no more than one overt inflectional morpheme on a verb. Since Icelandic allows more than one overt inflectional morpheme，it has the rich structure in（3b）. For example，an Icelandic inflected verb which expresses both tense and agreement has its tense affix and agreement affix separately attached to the stem：in kasta-ði-r，kasta is the stem，-ði the tense affix and -r the agreement affix expressing the second person singular（cf. Bobaljik & Jonas 1996：230）. Based on such evidence，Bobaljik concludes that it is the syntax that is responsible for licensing rich agreement in some languages and for prohibiting it in others.
Moreover，he proposes that syntactic structure variation causes variation in verb movement. A morphologically rich language has rich syntactic structure with more functional categories，and the functional categories can provide landing sites for verb movement. The morphologically impoverished languages，on the other hand，have no such functional categories and hence no landing sites for verb movement. The evidence that he provides for demonstration is that Icelandic allows multiple subject positions. As the Icelandic examples in（4）（from Bobaljik and Jonas 1996：196）illustrate，subjects reflecting old information occur to the left of a TP-adverb，in［Spec，AgrP］（see（4a）），while those reflecting new information occur to the right of such an adverb but still external to the VP（see（4b））.（4）a. Í gær kláruðu（þessar mýs）sennilega *（þessar mýs）ostinn.yesterday finished these mice probably these mice the.cheese‘These mice probably finished the cheese yesterday.’b. Í gær kláruðu（？margar mýs）sennilega（margar mýs）ostinn.yesterday finished many mice probably many mice the.cheese‘Many mice probably finished the cheese yesterday.’
That the subject in（4b）stays out of VP yet not in［Spec，AgrP］illustrates that there is another functional category，i.e. T，between Agr and V.
The view of the optional presence of functional categories is similar to Rohrbacher's（1999）view that a language with rich agreement has the functional category Agr as a landing site for verb movement. However，Bobaljik's view that it is syntactic differences that underlie morphological differences is contrary to Rohrbacher's as well as most other linguists'.
184.108.40.206 Our view on Bobaljik's（2002）hypotheses
In this dissertation，we agree with Bobaljik（2000）that Rohrbacher's rich agreement hypothesis is too strong，and that a weaker version is better to solve the problems about verb movement. The fact that a language with rich agreement inflection has verb movement does not entail that languages with impoverished inflection do not have verb movement. For example，in Mainland Scandinavian（MSc.）languages which are morphologically impoverished，the verb in main clauses always move to C under the verb second（V2）constraint. Furthermore，we agree with Bobaljik（as well as Iatridou 1990）that the presence of functional categories may vary among languages. However，we do not agree that it is syntactic differences that determine morphological differences.
Actually，we find that Bobaljik's（2000）argumentation falls into circularity. On the one hand，Bobaljik states that differences in syntactic structure are determined by morphological differences：morphologically rich languages have the rich syntactic structure，whereas morphologically impoverished languages have the simple syntactic structure. On the other hand，he proposes that a language with a simple syntactic structure permits no more than one inflectional morpheme because it has no landing sites on the syntactic level for the other inflectional morphemes. In other words，the only way that we can decide whether a language has a rich syntactic structure is by observing its morphological paradigms，which are in turn determined by its syntactic structure. In this sense，we conclude that Bobaljik's hypothesis that syntactic differences determine morphological differences is theoretically untenable. Contrarily，we assume that the presence of functional categories is determined by morphological representation. For example，if a language has agreement affixes，then it has the functional category Agr.
In a word，we hold a different view from Bobaljik's hypothesis that it is syntactic structure variation that drives morphological differences. Rather，we agree with Rohrbacher（1999）and most other linguists that morphological differences cause syntactic variation. Nevertheless，from Bobaljik's counterevidence against Rohrbacher's strong agreement hypothesis，we conclude that a synchronic analysis of verbal paradigms is not enough to explain language variation，since language change is a continuum，via languages/dialects mixing or independent evolution within one language. This means that different languages may have different origins and may have reached different developmental stages at a given time，which would thus cause syntactic variation. Considering this，we will do a diachronic analysis of morphological development in our study of syntactic variation.
Different from Rohrbacher's（1999）view that the presence of the functional category Agr is responsible for V-to-I movement，Trosterud（1989）suggests that the presence of overt morphological nominative case marking is crucial for triggering verb movement.
220.127.116.11 Nominative case-marking hypothesis
Trosterud（1989）proposes that the presence of nominative case inflections in Icelandic and the absence of this marking in MSc. explain why V-to-I movement is possible in Icelandic and impossible in MSc. The following paradigms in（5）（from Trosterud 1989，cited in Rohrbacher 1999：102）are of case markers in both Icelandic and one of the MSc. languages，i.e. Swedish.（5）
According to Trosterud（1989），nominative case is assigned by the finiteness feature［+F］，which can be located either in Comp or in Infl. This feature functions as an event operator and must therefore bind an appropriate variable such as the trace of V or Infl. If the finiteness feature［+F］is located in Infl，the operator-variable relation is created by V-to-I movement. After this movement，Infl and the verb trace t are non-distinct and［+F］in Infl can hence properly ibind t，which is indicated in（6）（cited in Rohrbacher 1999：103）.i
At the same time，Agr in Infl absorbs the abstract nominative case from［+F］and this case can no longer be assigned to the subject in［Spec，IP］. Therefore，the subject must bear whatever overt nominal case is available in this position. It follows that［+F］can be located in Infl only in languages，like Icelandic，which have overt morphological nominative case.（6）
If［+F］is located in Comp in a language，V-to-I movement would result in ungrammaticality. The event operator［+F］in Comp cannot establish an operator-variable relation with the verb trace t in iVP，because t is already bound by the verb in Infl，and no other ipotential variable is at hand. However，if the verb remains in situ and Infl lowers to V，there would be no similar problems. The event operator［+F］in Comp can establish the required operator-variable relation with t，i.e. the trace of the lowered Infl. Moreover，［+F］ican assign abstract nominative case to the subject in［Spec，IP］，since［+F］and Agr are not structurally adjacent and the latter therefore does not absorb the abstract nominative case from the former. This is indicated in（7）（cited in Rohrbacher 1999：105）. It follows that［+F］located in Comp is the only option for languages like MSc. that do not have overt morphological nominative case. In short，Trosterud's（1989）theory predicts that all and only those languages that have overt morphological nominative case allow V-to-I movement.
18.104.22.168 Counterevidence from Faroese
Rohrbacher（1999）provides counterevidence from Faroese to reject Trosterud's nominative case-marking hypothesis：Faroese has overt morphological nominative case but no V-to-I movement. The case paradigm in Faroese is shown in（8）（from Rohrbacher 1999：105）.
Rohrbacher's rejection，however，seems to be a bit deviant from the data. As shown in（1）in the preceding section，Bobaljik（2002）mentions that in Faroese both verb movement and verb-in-situ word orders are possible. Therefore，Rohrbacher's Faroese example is untenable in refuting Trosterud's view that overt morphological nominative case is responsible for V-to-I movement. However，Faroese really can be taken as a counterexample against Trosterud's theory，if the argumentation is from another perspective.
As shown in the above paradigm in（8），Faroese has distinctive nominative marking only in the singular and never in the plural. Similarly，nor does Icelandic always distinctively mark nominative case. In Icelandic neuter and strong feminine nouns，nominative case is never morphologically distinguished from accusative case，as shown in（9）（from Rohrbacher 1999：106）.
Corresponding to the morphological similarity，both Icelandic and Faroese allow two options in syntax，i.e. verb movement and verb-in-situ word orders. In section 22.214.171.124，we have introduced the two optional word orders in Faroese，and the examples in（10）and（11）are an illustration of the optional word orders in Icelandic.（10）a. að Jón hafði raunverulega keypt bókinathat John had actually bought the-bookb. að Jón keypti ekki bókina （Icelandic）that Jon bought not the-book（from H&P 1995：10，27）（11）a. þetta er tilboð［sem ekki er hægt að hafna］this is offer that not is possible to reject‘This is an offer that cannot be rejected’b. Hver sagðir þú［að sennilega hefði skrifað þessa bók］who said you that probably has written this book‘Who did you say has probably written this book？’（from Holmberg 2000：446）
In（10），the finite verb moves to a higher position and precedes sentence adverbs and the negation word，whereas in（11）it follows the negation and the adverbial，which means that it remains in situ.
We suggest that the similarities between Icelandic and Faroese in both morphology and syntax be analyzed from a diachronic perspective. As is generally known，both morphology and syntax are in continuous change. On the one hand，morphological inflections are on the loss；on the other hand，syntactic change is in progress from movement to non-movement. As Lehmann（1992：240）states：“Changes are slower in syntax than in phonology and morphology. As a result，we generally find residues of past constructions side by side with the current patterns.” In this sense，the two possible word orders in both Icelandic and Faroese finely illustrate that morphology and syntax in the two languages are in progress of development. Along these lines，we conclude that a simple synchronic observation of morphological paradigms in one language，the presence of nominal case markers in this case，cannot explain verb movement in other languages. Only when we make a diachronic comparison of the morphological paradigms in languages can we explain their syntactic differences. In Chapter 3 and Chapter 4，we will follow the diachronic approach in analyzing the syntactic differences in the Romance languages and the Germanic languages.
Like Rohrbacher（1999），Roberts（1993）suggests that verb movement is determined by verbal agreement morphology. He proposes that the relative richness of overt subject-verb agreement morphology is responsible for the presence versus absence of V-to-I movement.
126.96.36.199 The hypothesis of overt distinct agreement marking
Roberts（1993）proposes that V-to-I movement occurs in languages which have overt distinct morphological plural marking，where ‘overt’ means non-empty and ‘distinct’ means ‘distinct from the singular form（s）’（Roberts 1993：267）. This proposal seems to be compatible with the paradigms in（12）（from Rohrbacher 1999：108）.
As shown in（12），the V-to-I movement languages Yiddish and Icelandic have overt distinct morphological plural marking，whereas the V-in-situ languages English and MSc. do not have overt distinct plural marking. In English，there is no plural marker，while in MSc. the plural markers are overt but not distinct，since all the persons and numbers have the same marker. Therefore，neither English nor MSc. allow V-to-I movement.
188.8.131.52 Our view based on Rohrbacher's（1999）counterevidence
In our understanding，Roberts's（1993）proposal is similar to Rohrbacher's（1999）in that it is a one-way hypothesis，i.e.，only languages with rich agreement allow V-to-I movement. In section 184.108.40.206，it is already shown that a strong rich agreement hypothesis is not tenable in the explanation of verb movement. Moreover，Roberts's proposal does not hold water in Faroese，which has several patterns of agreement paradigm，as shown in（13）（from Rohrbacher 1999：110-111）.（13）
Rohrbacher（1999：112）states that in these paradigms the imperative singular is shorter than the infinitive and the present indicative plural and it is taken to be the stem. The plurals nevna，krevja and strúka and the plurals of the roughly 60% of all Faroese verbs which pattern similarly require the overt distinct morphological plural marker –a. For example，the present indicative plural kasta involves the stem kasta- and the regular plural suffix –a. Its derivation is shown in（14），in which a short stem-final vowel is deleted in front of a suffix-initial vowel.（14）/kasta+a/ → kast-a
This means that all Faroese verbs have overt distinct morphological plural marking in the sense of Roberts's generalization. Hence，Roberts' theory wrongly predicts obligatory V-to-I movement in Faroese.
In a word，we think that Roberts's hypothesis，like Rohrbacher's（1999），is too strong and empirically untenable and hence cannot be taken as the parameter of verb movement. As mentioned in section 220.127.116.11，in morphologically impoverished languages，like the MSc. languages，there is also verb movement. The verb movement，as we will mention in Chapter 4，is actually a relict of the V-to-C movement in Old Scandinavian. Therefore，we cannot set a parameter of verb movement by virtue of a synchronic observation of morphological paradigms in languages；rather，a diachronic study of language change is quite necessary for syntactic analysis.
1.1.5 Holmberg and Platzack（1995）
The study in Holmberg and Platzack（henceforth H&P，1995），which mainly focuses on the comparison between the insular Scandinavian languages（ISc.）and the Mainland Scandinavian languages（MSc.），also takes morphological evidence as significant in the syntactic analysis. They claim that the major syntactic differences between ISc. and MSc. can be described as the effect of the morphological differences in subject-verb agreement and nominal case.
18.104.22.168 Presence of Agr
H&P（1995）propose that the difference between ISc. and MSc. is a direct consequence of the presence of the functional category Agr in ISc. in contrast to the absence of Agr in MSc.，and such distinguishing is based on the observation of agreement paradigms in the languages. The ISc. languages are typical synthetic languages，having a fairly rich system of verbal agreement morphology as well as case morphology. In comparison，the MSc. languages are more analytic in that they have no subject-verb agreement at all and no case morphology except for the adnominal genitive. Based on the morphological evidence，they assume that the ISc. languages have the functional category Agr and the MSc. languages do not have this category. This hypothesis can account for the syntactic variation between ISc. and MSc.，which is mainly in subordinate clauses.
In ISc. subordinate clauses，the tensed verb always precedes sentence adverbs and the negation word，whereas in MSc. the tensed verb always follows sentence adverbs and the negation word，as shown in（15）and（16）.（15）a. að Jón hafð i raunverulega keypt bókina （Icelandic）that John had actually bought the-bookb. att John faktiskt hade köpt boken （Swedish）that John actually had bought the-book（from H&P 1995：10）（16）a. að Jón［ keypti］（ekki）［［ e］ bókina］ （Icelandic）IiVPVithat Jon bought not the-bookb. att Jon［ e］（inte）［［ köpte］ boken］（Swedish）IiVPVithat Jon not bought the-book（from H&P 1995：27）
In Holmberg and Platzack's earlier works（Platzack and Holmberg 1989，Holmberg and Platzack 1991，cited in H&P 1995：26），the verb is inserted in its root（or stem）form and the inflections are attached by virtue of successive head movement to functional head positions. Therefore，the inflected verb is supposed to be derived by raising V to Iin ISc.，but by lowering I to V in MSc.，respectively yielding（roughly）S-structure representations in（16）. V-to-I movement is taken to be triggered by the requirement that the verb morphology be ‘checked’ by the features of I.
Contrary to the earlier works，H&P（1995）adopt another model，in which the verb is inserted with its inflections attached，V-raising applying in order to check the verbal morphology against the features of I.Taking this model as the theoretical basis，they propose that the difference between ISc. and MSc. is a consequence of the presence/absence of the functional category Agr. Based on their assumption that the functional category Agr is an inherently maximal functional head adjoined to a projecting functional head like T，they conclude that in ISc. Agr is present and thus verb raising to T is possible，whereas in MSc. there is no Agr and thus no verb movement to T. Therefore，in MSc. subordinate clauses，the verb always follows sentence adverbs and the negation word.
Moreover，they propose another notion，the finiteness feature，in order to account for the similarity in main clauses of V2 languages and the syntactic difference between V2 languages and non-V2 languages.
22.214.171.124 Finiteness feature
Regardless of the difference in subordinate clauses of ISc. and MSc.，there is a similarity between these languages：they are all V2 languages. In other words，they all obey the V2 constraint，according to which the finite verb must occur in the second position of a sentence，as shown in the following Swedish examples：（17）a. Under sängen hittade han faktiskt pengarna.under bed-the found he actually money-theb. Faktiskt hittade han pengarna under sängen.（cited in Holmberg and Platzack 200
In the above sentences，the finite verb hittade ‘found’ always occupies the second position，and the initial element preceding the finite verb can be a prepositional phrase as in（17a），a sentence adverb as in（17b），or other elements，which will be shown in Chapter 4. Since there is no Agr in MSc.，the verb in these Swedish examples must have moved to C. In other words，V2 languages differ from non-V2 languages in that the former has the finite verb always moving to C whereas in the latter the possible verb movement is up till T. H&P（1995）assume that this syntactic difference is due to the different location of the finiteness feature［+F］.
According to H&P（1995），the finiteness feature［+F］，which can be located in either C or I，determines the target position of verb movement：the feature［+F］in C causes verb movement to C，whereas the feature［+F］in I causes verb movement to I. Verb movement to the relative target position is for the licensing of the finiteness feature［+F］，which is required by the following licensing condition：（18）Licensing Condition for the Finiteness Feature［+F］An occurrence of the feature［+F］is licit if and only if the head hosting it is lexicalized and governs a phonetically realized element bearing nominative case，or the trace of such an element.（from H&P 1995：44）
H&P（1995）propose that V2 languages differ from most other languages in having the feature［+F］in C instead of I. Along these lines，the tensed verb in main clauses of a V2 language is forced to move to C to lexicalize［+F］，allowing at most one preceding constituent（situated in［Spec，CP］）. In subordinate clauses of such a language，where C is filled by a complementizer，no verb movement to C is possible. In non-V2 languages like modern English and French，on the other hand，the tensed verb is only forced to move to I，where［+F］is situated. This accounts for the main clause-subordinate clause asymmetry，which can be found in V2 languages but not in non-V2 languages，as illustrated in（19）and（20）.（19）a. German
Karl kaufte das buch nicht. Das Buch kaufte Karl nicht.
Karl bought the book not the book bought Karl not
daß Karl das Buch nicht kaufte
that Karl the book not bought
Ulf köpte inte boken. Boken köpte Ulf inte.
Ulf bought not the-book the-book bought Ulf not
att Ulf inte köpte boken
that U. not bought the-book（20）a. EnglishJohn did not buy the book. The book John did not buy.that John did not buy the book
b. FrenchJean n'a pas achèté le livre. Le livre Jean n'a pas achèté.John has not bought the book the book John has not boughtque Jean n'a pas achèté le livrethat John has not bought the book（both（19）and（20）from H&P 1995：45）
126.96.36.199 Our relative proposals
In this dissertation，H&P's（1995）hypothesis of the presence of Agr is one of the main theoretical bases of our proposals. Based on this，as well as Iatridou's（1990）optional presence of the functional categories that we have mentioned in section 188.8.131.52，we assume that the presence of functional categories is driven by the requirement of morphological representation；the details of the proposal will be presented in Chapter 2.
Nevertheless，H&P's（1995）morphological parameter does not give a definition of morphological richness. In other words，it does not provide an elaborate classification of agreement paradigms in terms of richness degree. For example，the ISc. languages mentioned in their framework have fairly rich morphology and the MSc. languages have much impoverished morphology. Actually，MSc. has no agreement at all. It follows that their morphological distinction is applicable only to languages with Agr and languages without Agr. However，there are also languages with degraded morphological richness，8like German，and languages which have agreement inflection（s）but are generally regarded as morphologically impoverished，like English. The former has overt verb movement but the latter does not have verb movement. It seems that their morphological parameter cannot account for these differences. Considering this，in this dissertation we will provide a more elaborate classification of agreement paradigms，in order to cover more language differences and thus provide more evidence for the hypothesis that syntactic variation is the effect of morphological differences.
In addition，we think that H&P's（1995）view of the two alternative locations of the finiteness feature［+F］in non-V2 languages and V2 languages is somewhat arbitrary：they provide no morphological evidence for the determination of the locations of the feature［+F］. Since the finiteness feature is related to agreement and tense，and since the MSc. languages have no subject-verb agreement at all，the finiteness feature in MSc. is unlikely to be in C. In order to solve this problem，we will replace the finiteness feature with the agreement feature and propose that the position of the agreement feature，which can be in C or T，is determined by the presence/absence of complementizer agreement. The details of our proposals will be given in Chapter 2.
1.2 Researches on Subject Omission
Subject omission is generally regarded as determined by verbal morphology. For example，Chomsky（1981，1982）claims that only languages with a rich inflectional system，which can recover the contents of the null subjects，are null subject languages. However，there is no uniform conclusion on the degree of morphological richness that permits null subjects. Some other linguists try to explain the complex null subject phenomena from a morphological perspective. For instance，Vainikka and Levy（1999）suggest that there is a correspondence between morphological paradigms and subject omission and that the distribution of third person subjects is crucial in determining the status of subject omission. In the following sections，we will introduce these theories on subject omission.
Chomsky（1981，1982）proposes that the distribution of referential null subjects is determined by the recoverability of the content of the subject：a pronominal may be left out only if its content can be recovered morphosyntactically. The null subject is inherently unspecified for Ф-feature values；it becomes specified by virtue of being in a Spec-head relation，enabling coindexation with finite I（nfl）which contains Agr typically realized as an affix on the finite verb or auxiliary. Through coindexation with I，the null subject inherits the Ф-feature values of Agr. In other words，Chomsky（1981，1982）proposes a Null Subject Parameter，which claims that in languages with widespread null subjects（and null objects）there should be a rich inflectional morphology，especially an elaborate system of verbal agreement. For instance，Italian has null referential subjects，null semi-argument subjects，and null expletive subjects，because the verbal agreement system of Italian is very rich. In contrast，Swedish，which has no agreement，does not permit subject omission. The relevant examples of the two languages are respectively shown in（21）and（22）.（21）a. Pavarotti dice che pro mangia gli spaghetti. （Italian）Pavarotti says that eat.3sg the spaghetti‘Pavarotti says that（he）eats spaghetti.’b. pro piove.rain.3sg‘（It）is raining.’c. pro sembra che Pavarotti mangi gli spaghetti.seem.3sg that Pavarotti eat the spaghetti‘（It）seems that Pavarotti eats spaghetti.’（from Y. Huang 2000：50）（22）（Swedish）Igår köpte Johan en ny cykel. Den ställde *（han）i köket.yesterday bought John a new bike. it put he in kitchen-the‘Yesterday John bought a new bike. He put it in the kitchen.’（from Holmberg and Platzack 2005：4）
1.2.2 Jaeggli and Safir（1989）
Jaeggli and Safir（henceforth J&S，1989）try to provide an answer，with their Null Subject Parameter and Morphological Uniformity respectively presented in（23）and（24），to the distribution of null subjects.（23）The Null Subject ParameterNull subjects are permitted in all and only languages with morphologically uniform inflectional paradigms.（24）Morphological UniformityAn inflectional paradigm P in a language L is morphologically uniform iff P has either only underived inflectional forms or only derived inflectional forms.（from J&S 1989：29-30）
According to（24），if in the paradigm some forms are morphologically divisible into stem+affix while other forms are bare stems，then it is not uniform. The paradigms for Spanish，German，Irish，Chinese，and Japanese are all morphologically uniform whereas those for English and French are not. In languages with distinct verbal agreement inflections for all persons and numbers，the null subject can be identified by the agreement affixes with the relevant person/number features，which are considered a part of the tense/Infl node. In languages with no agreement，on the other hand，identification can be achieved by the inheritance of agreement features by a lower Infl from a higher Infl or C-commanding NP.
However，J&S's（1989）Null Subject Parameter cannot explain the null subject phenomena in languages like German and Icelandic which have a uniform morphology，i.e.，only derived inflectional forms，to show person and number agreement but which allow no null thematic subjects. In order to solve this problem，they add another principle，the identification principle. They argue that rich agreement is not a sufficient condition to achieve identification；government is also crucial. Moreover，they suggest that the tense and Agr inflections should be represented in the same node for identification of null subjects. Since tense is the source of case assignment，it is crucial in these identification configurations. Therefore，according to the identification rule in（25），the crucial factor appears to be whether the category containing the identifying features is also the locus of case assignment，i.e.，whether the tense and Agr inflections are inserted in the same node.（25）Identification by Agreement
Agr can identify an empty category as thematic pro iff the category containing Agr case-governs the empty category.（from J&S 1989：35）
Since both German and Icelandic are V2 languages，they suggest that in these languages Tense is located in Comp whereas agreement is located in Infl. It follows that these languages do not permit null thematic subjects.
1.2.3 Shortcomings of the two null subject hypotheses
Chomsky's（1981，1982）Null Subject Parameter can account for the distinctions between morphologically rich languages like Italian and morphologically impoverished languages like English，but it cannot explain languages like German and Icelandic which show person and number agreement with their subjects but which do not permit null thematic subjects.Besides，it cannot answer why languages like French with similar rich verbal agreement morphology as Italian do not permit any subject omission.J&S's（1989）Null Subject Parameter，which is based on their Morphological Uniformity，has similar problems. Moreover，their Morphological Uniformity fails to cover some other morphologically impoverished languages like the MSc. languages. According to the Morphological Uniformity，morphological paradigms can be classified into two types：uniform and mixed. The MSc. languages evidently have uniform paradigms in that there is no agreement affix but only one uniform affix for all persons and numbers in each tense，but they do not allow any null subject.
Furthermore，neither of the parameters in Chomsky（1981，1982）and J&S（1989）considers null expletives. As is generally known，there are three types of null subjects. In terms of the reference of the subject，Hofherr（2003）classifies null subjects into three types：full-referential，quasi-referential and non-referential. The first type is also called null thematic subjects and the latter two types are often uniformly called null expletives. In specific cases where a distinction between them is necessary，they are respectively named null quasi-arguments and pure expletive pros in some linguists' frameworks. Obviously，in both Chomsky's（1981，1982）framework and J&S's（1989）research，only the first type of null subjects，i.e. null full-referential subjects，are considered. The same problem is found in Vainikka and Levy's（1999）research，which will be introduced in the following section.
In addition to the failure of the two parameters mentioned above to cover all null subject phenomena，J&S's（1989）explanation of German based on their identification rule also has problems. They suggest that in V2 languages，like German，Tense is located in Comp whereas agreement is located in Infl. According to the identification rule in（25），only when the tense and Agr inflections are inserted in the same node can an empty category be identified and thus be licensed. It follows that German cannot identify null subjects and hence it does not permit null thematic subjects. However，the German structure is not as they describe. According to den Besten（1983），the V2 inversion in German always involves movement of the inflected verb into C. But the verb that moves to C carries both tense and Ф-features；therefore，the German structure is not that Tense is located in Comp and agreement is located in Infl. In conclusion，J&S's（1989）identification rule cannot provide a correct explanation for languages like German which is morphologically uniform and rich but permits no null thematic subjects.
1.2.4 Vainikka and Levy（1999）
In Vainikka and Levy's（henceforth V&L，1999）framework，subject omission is determined by the pronominal connection between verbal agreement affixes and personal pronouns. Like Rohrbacher（1999）and Beard（1991），V&L（1999）also take into consideration the distinction between first/second persons and third person. However，there is a difference in the markedness of these person features. In Rohrbacher's and Beard's theories，［1st］and［2nd］person features are regarded as marked whereas the third person feature is unmarked. Since overt morphology represents the marked value of the feature and zero morphology represents the unmarked value of the feature，the systematic absence of overt morphology for the third person（singular）in many languages then suggests that third person is universally the unmarked value of the person feature. Along these lines，Rohrbacher（1999）proposes that distinct markers of first/second persons in a morphological paradigm are a crucial factor in determining the referentiality of the paradigm and the presence of the functional category Agr which provides a landing site for verb movement.
Contrary to Rohrbacher and Beard，V&L（1999）hold the view that the third person is of more significant status in a morphological paradigm，because the third person referent is much less easily determinable ― in the absence of further contextual information the third person referent could be anyone that is not the speaker or the hearer at the moment.Their evidence is that in some languages，like Hebrew and Standard Finnish，there is a pronominal connection in both first and second persons but not in third person. Based on this，they suggest that the distribution of third person subjects is crucial in determining the status of subject omission. The first and second person agreement features may be base-generated in the subject position，due to their ability to restrict the set of potential referents in ways that are relevant for the conversational situation. In languages which take this option，the first/second person subject pronouns can be omitted. In contrast，due to their more remote connection to the conversational situation，the third person features cannot be the only features in the whole paradigm which are base-generated in the subject position. In other words，if there is no pronominal connection in third person in a morphological paradigm，the third person subject in such a language should be base-generated in the Agr position instead of the subject position and thus cannot be omitted.
184.108.40.206 Pronominal correspondence
In terms of the licensing of null third person subjects，V&L（1999）divide languages into two main groups：those which allow omission of thematic，referential third person subjects in matrix clauses（Group A）and those which do not（Group B）. Group A actually contains languages in which null subjects can occur in all persons. Spanish，Italian，Tamil，and Chinese are examples. Group B can be divided into two subtypes（V&L 1999：661）：（26）Group B1：languages in which the agreement paradigm reveals a pronominal connection in both first and second persons，but not in third person（Hebrew and Standard Finnish）；Group B2：languages in which no such pronominal connection can be identified（English，Swedish，German，French，Russian，and Colloquial Finnish）
Since languages in Group A are generally regarded as null subject languages，V&L（1999）have less discussion of them. As for Group B，they propose that there is a correspondence between morphological paradigms and subject omission：if there is a connection between agreement inflections and corresponding pronouns，then the subject is allowed to omit；otherwise，there is no subject omission. The following data from Finnish and Hebrew provide evidence for their hypothesis. In Standard Finnish，first and second person subject pronouns are optional，as shown in（27b），whereas in the third person an overt subject NP is normally required，as shown in（27a）.（27）a. *Nousi junaan.step.past.3sg train-into‘（He/she）boarded the train.’b. Nousin junaan.step.past.1sg train-into‘I boarded the train.’（from V&L 1999：614）
Correspondingly，there is a clear phonological relationship in Finnish between the agreement suffix and the corresponding pronoun in the first/second person plural，while in the third person plural no such relationship exists. The correspondence of verbal agreement affixes and personal pronouns is shown in the following table（28）（from V&L1999：639）.
As table（28）shows，the first and second person agreement suffixes in Finnish are phonologically related to the corresponding person pronouns，and the relationship is more obvious in the plural（-mme vs. me；-tte vs. te）. Although it is less obvious in the singular，Hakulinen（1979：87，cited in V&L 1999：639）states that it also exists，at least historically. In Finnish diachronic and synchronic phonology，there is a general /ti/→/si/ rule；therefore the 2sg pronoun is reconstructed as *tinä，and the 1sg suffix can be reconstructed as *-m. In the third person，on the other hand，there is neither a synchronic nor a diachronic relationship between the pronouns and the affixes. It follows that the third person subject is not allowed to omit，as illustrated in（27a）.
A similar pattern is found in the Hebrew past and future tenses. As V&L（1999：644）state，Hebrew thematic subjects may be omitted in the first/second person in the past and future tenses，and a discernable phonological similarity exists between first and second person pronouns and inflectional endings. Hebrew subject omission in the first/second person is exemplified in（29）and the agreement paradigm of Hebrew is shown in（30）（both（29）and（30）from V&L 1999：644）.（29）a. Halaxti itxa ki racita.go.past.1sg with-you because want.past.2sg.m‘I went with you because you wanted.’b. Elex itxa im tirce.go.fut.1sg with-you if want.fut.2sg.m‘I will go with you if you want.’
As the Hebrew agreement paradigm in（30）shows，in the past tense，the second person agreement affixes are strongly related to the pronoun forms：ata vs. –ta，at vs. –t，and atem vs. –tem. A similar correspondence is found in the first person plural（anaxnu vs. –nu），and perhaps in the singular：ani vs. –ti. In the future tense，there may be a similar relationship，in that the consonant t is found in both the second person pronouns and affixes，while n occurs in the first person pronouns and in the 1pl affix. In contrast，there is no such pronominal relationship between the third person pronouns and affixes. Therefore，the third person subject in Hebrew cannot be omitted，as shown in（31）.（31）a. Hu halax itxa ki racitahe go.past. 3sg.m with-you because want.past.2sg.mb. *Halax itxa ki racitago.past.3sg.m with-you because want.past.2sg.m‘He went with you because you wanted.’（from V&L 1999：643）
Unlike languages in Group B1，such as Standard Finnish，which have mixed null subject patterns，languages in Group B2 generally do not allow subject omission because there is no pronominal connection in any person. For example，in Colloquial Finnish，a relationship between the pronouns and the affixes is no longer identifiable in the synchronic grammar，not even in the plural. The following table（32）（from V&L 1999：663）is a comparison of the morphological paradigms in Standard Finnish（SF）and Colloquial Finnish（CF）.
Correspondingly，Colloquial Finnish has become a non-null-subject language. Its non-null-subject sentence pattern is shown in（33）.（33）？*（Mä） näin sut eilen.I.nom see.past.1sg you.acc yesterdayI saw you yesterday.（from V&L 1999：662）
220.127.116.11 Our view on Vainikka and Levy's（1999）hypothesis
V&L's（1999）morphological hypothesis about the relationship between verbal agreement affixes and pronouns is compatible with typologists' conclusion that most verbal inflections are developments of free pronouns，which will be introduced in Chapter 2. This hypothesis very well accounts for subject omission for different persons. However，there is a shortcoming in their theory，which is the same as that in Chomsky（1981，1982）and J&S（1989）：it only considers null thematic subjects，without consideration of null expletives. For instance，in Icelandic，both null quasi-arguments and null pure expletives are allowed，whereas in German，only the latter type is allowed，but neither of them permits null thematic subjects.V&L（1999）categorize German into Group B2 as a non-null-subject language but do not mention Icelandic. It is therefore very difficult to make a definition of null subject languages in terms of their classification.
In this dissertation，we suggest that the three types of null subject phenomena，i.e. null full-referential subjects，null quasi-referential subjects and null non-referential subjects，be analyzed respectively. We propose that the permission of different kinds of null subjects is related to the referential power of agreeing verbs；and the loss of the referential power is gradual from full-referential to quasi-referential and then to non-referential. In other words，the gradual loss of the referential power of agreeing verbs causes the gradual disallowance of null subjects. For example，in Italian where the verbal agreement affixes are quite distinct for each person and number，which means that the referential power of the agreeing verbs is quite strong，all the three types of null subjects are allowed. In Icelandic，the verbal agreement affixes which express the first person singular and the second person singular have fallen together，which entails a slight loss of the referential power. Correspondingly，full-referential subjects are not allowed to omit. In German where assimilation happens to more verbal agreement affixes，only null non-referential subjects are allowed. Along these lines，we conclude that a classification of agreement paradigms in terms of morphological richness and a diachronic study of morphological change are necessary in the analysis of subject omission. In Chapter 5，we will do a detailed analysis of subject omission by virtue of such a diachronic study.
In this chapter，we have introduced some of the contemporary morphological studies on word order and subject omission. Since most of the contemporary morphological studies are from a synchronic perspective and generally focus on one single aspect of morphology，they cannot cover all the syntactic differences among human languages，including difference in word order and subject omission. For example，we can find counterevidence from Faroese to Rohrbacher's（1999）and Trosterud's（1989）strong agreement hypotheses and Roberts's hypothesis of overt distinct agreement marking. In our opinion，a diachronic comparison of morphological paradigms is quite important in the analysis of word order variation.
As for subject omission，the generativists' parameters cannot cover all the complex null subject phenomena in human languages. Since there are three types of null subjects and the licensing condition of each kind is different，we may not hope to cover all the null subject phenomena with a single or a few syntactic parameters. Comparatively，Vainikka and Levy's（1999）analysis of the pronominal connection appears to be more flexible in that different licensing conditions of subject omission are even allowed to be present in one language. Nevertheless，it does not consider the permission of null expletives，which is actually a hotly debated topic in syntax.
In this dissertation，we will try to give an account of the syntactic differences in languages，including differences in word order and subject omission，through a diachronic study of morphological change. Before that，in the next chapter we will introduce our main proposals in this dissertation and some theoretical foundations on which our hypotheses are based.
Chapter 2 Theoretical Background and Main Proposals of Morphological Effects
In the last chapter，we reviewed some contemporary morphological studies on word order and subject omission and analyzed their problems. We proposed that a single parameter and a synchronic observation of morphological paradigms in languages cannot account for the complex language differences，including word order variation and different null subject phenomena. In this dissertation，we will follow a diachronic approach in syntactic analyses. In other words，we will study morphological change and syntactic change from a diachronic perspective.
In this chapter，we will present our main proposals of morphological effects，together with an introduction to the theoretical background for the present study. Our purpose is to provide a generalization of morphological effects on word order variation and subject omission through a diachronic study of morphological change，which can thus avoid the problems of the synchronic study of morphological comparison mentioned in Chapter 1.
2.1 Morphological Evolution
Morphological evolution is regarded in this dissertation as the major cause of language variation. Since the evolutional process is gradual so that different languages might have reached various evolutional stages at a given time，we will first discuss morphological development. Since verbal agreement morphology is more significant in explanation of word order variation and null subject permission，our discussion will focus on the developmental process of verbal agreement markers. Section 2.1.1 is an introduction to the theoretical background about the evolution of agreement markers，and the following subsections are a presentation of relevant morphological evidence that illustrates the evolutional process.
2.1.1 Theoretical background
18.104.22.168 Original morphological inflections
As cited in Ariel（2000：197），many typologists have made a lot of empirical studies and have concluded that most verbal inflections are developments of free pronouns. For instance，Givón（1971）have discussed the development of full pronouns into agreement markers in Bantu languges；Comrie（1978，1981）have analyzed the Mongolian languages；Moscati et al.（1969）the Semitic languages；Steele（1977）a few Uto-Aztecan languages；Mithun（1991）the North American languages；Haiman（1991）Northern Italian dialects；and Dixon（1980）the Australian languages.
Anderson（2001）makes a detailed statement about the development of verbal inflections. He claims that much of inflectional morphology arises over time as the result of a well-known cycle，by which full lexical words may be reduced（semantically and phonologically）to pronouns or particles；these in turn may become attached to other words as clitics，and subsequently re-analyzed as affixes or even incorporated into the shapes of inflected words. In a word，modern verbal agreement affixes are developed from clitics，which are in turn developed from full pronouns.
The following data from Basque serves as an example of the development of agreement markers from full pronouns. The agreement markers in Basque finite verbs are mostly very similar to the corresponding free pronouns，as shown in the following possible forms of the verb joan ‘go’ in（1），so that they must have been derived from incorporation of free pronouns into the finite verb.（from Trask 1996：117）
22.214.171.124 Tendency of morphological evolution
According to the typical nineteenth century evolutionary theory of language，there are three stages of evolutionary advance in language morphology：isolating，agglutinative and fusional；and the line of evolutionary advance is from isolating languages through the agglutinative to the fusional stage. Greenberg（1957：60）further proposes that fusional languages may take a retrograde movement to isolating languages（also see Dixon 1997）. For example，Indo-European languages seem to be changing from a synthetic to a more analytic or isolating type.
The brief description of the three types of languages given in Dixon（1997：41）is as follows. An isolating language is one in which each meaning element makes up a distinct word（e.g. Vietnamese and Chinese）. In the agglutinative stage，a word is likely to contain several meaning elements（or morphemes）but these are clearly separable（e.g. Turkish and Swahili）. In a fusional language，a word may contain several meaning elements，some of them being fused together so that a single vowel may simultaneously mark，say，tense，voice，and person and number of subject（e.g. Latin and Sanskrit）. Lehmann（1992）provides data to show the distinction between agglutinative languages and inflectional languages. As illustrated in the following comparison of the Turkish inflections of yol ‘way’ and kuş ‘bird’ with those of Latin via and avis（from Lehmann 1992：97），agglutinative languages are different from inflectional languages because they maintain bases distinct from endings and they show little morphophonemic change at morpheme boundaries. In Turkish，the endings can be neatly separated from the stem and are the same for both examples，though with modifications by phonological harmony；by contrast，in Latin the endings have merged with the vowel of the base and can only be determined by historical methods.
As Dixon（1997：42）states：“Modern-day isolating languages may have had a fusional ancestor，whose portmanteau morphemes have been lost through phonological and morphological change；… Present-day agglutinative languages may have had an ancestor of mere isolating profile，with what were distinct words having developed into grammatical affixes（e.g. postpositions into cases）. If the proto-language was agglutinative，then many or most of its modern descendants are likely to have merged grammatical morphemes into a more fusional structure.” This statement entails that the tendency of morphological evolution is towards more fusional and then more simplified：the more natural morphology in agglutinative languages becomes more unnatural and fusional，whereas the morphology in fusional languages becomes more simplified，which causes a fusional language gradually to become isolating.
2.1.2 Process of morphological evolution
In this subsection，we combine some linguists' views（cf. Mayerthaler 1981，Wurzel 1989，Lehmann 1992，Trask 1996，Dixon 1997）and their empirical evidence and generalize the process of morphological evolution as follows. Typically，morphological change proceeds toward less markedness/more naturalness. This is illustrated in the change of German suppletive-comparative paradigms，e.g.，Middle High German übel ‘bad’ ― wirser ‘worse’―wirsest ‘worst’ >；New High German übel―übler―am übelsten（Mayerthaler 1981，cited in Wurzel 1989：13）. The developmental tendency towards naturalness implies that morphological paradigms would tend to become uniform：with one uniform inflectional affix for all cases（including agreement affixes and case affixes）. In other words，there would be only two alternative morphological paradigms：either with quite distinct inflectional affixes（at the initial development）or with one uniform affix，in the latter case inflected words being formed by virtue of regular rules.
However，morphological change does not uniformly proceed in this direction. According to Wurzel（1989），morphological evolution is under the conflicts of phonological naturalness and morphological naturalness. Regular phonological changes very often disrupt regular inflectional paradigms，but at the same time the pressure of morphological analogy tends to maintain or restore the regular paradigms. Hence，an increase in phonological naturalness involves a decrease in morphological naturalness，which thus results in the occurrence of ‘mixed’ morphological paradigms，i.e. the gradual loss（i.e. the assimilation）of inflectional affixes.
Wurzel（1989：18-19）uses an example to illustrate the conflict between phonological and morphological naturalness. In Proto-Germanic masculine，n-stems like drupe-n ‘drop’ had the plural form drupan-iR with the clear categorial coding -iR. Because of the strict fixation of word accent on the first syllable in Germanic，the effect of phonological reduction processes making for greater articulatory ease at first reduced -iR，so that the relevant form in Proto-Norse was drup-an. Finally，by a further reduction，final /n/ was also deleted. The plural form of Old Swedish drup-i was drup-a. Due to its smaller number of syllables，the form drupa was easier to pronounce than the initial form drupaniR，and thus was phonologically more natural. But this increase in phonological naturalness resulted in the loss of a functionally important morphological distinction：in Old Swedish the form drupa occurred not only in the noun plural but also in the genitive/dative/accusative singular and in the genitive/accusative plural. Thus，the development of phonological naturalness has resulted in a decrease of morphological naturalness.
In response，as early as in Old Swedish，a morphological change began. The flexive –a，ambiguous in the paradigm，was replaced in the noun plural by the unique plural coding –ar of masculine a-stems. The noun plural form of drup-i thus became drup-ar. Morphological naturalness increased and markedness decreased. But a postvocalic［r］is relatively difficult to pronounce. Hence，the increase of morphological naturalness induced a reduction of phonological naturalness. The morphological change in Old Swedish thus shows that the conflict between phonological naturalness and morphological naturalness is a driving force of morphological simplification.
Another factor that causes morphological simplification is the analogy process. As Lehmann（1992：232-3）states：“…sound changes may contribute to the possibility of new morphological contrastive devices，which then may be extended by analogy…Many morphological modifications in languages are spread by analogy，though the innovations may be a result of phonological change，or also of some other process such as borrowing.” For example，analogical leveling has applied to plural forms of German borrowed words. As Trask（1996：28）states，German has a wide variety of patterns for plurals：Weg ‘way，’Wege；Mann ‘man’，Männer；Mensch ‘person’，Menschen；Uhu ‘eagle-owl’，Uhus；Lehrer ‘teacher，’Lehrer；Bruder ‘brother，’Brüder；Hand ‘hand，’Hände；Blume ‘flower，’Blumen；Buch ‘book，’Bücher；Mineral ‘mineral’，Mineralien. However，loan words either take no case endings or take the -s plural：Test-s，Bungalow-s，Teenager-s，Schock-s ‘shock，’Kamera-s ‘camera’，etc. These may be the first signs that German may be going the same way English went many centuries ago：generalizing the once obscure -s plural at the expense of a dozen other patterns. The above evidence thus indicates that the evolutional tendency of nominal morphology is towards simplification.
Actually，not only noun plural forms but also morphological cases on nouns become more simplified. During the evolutional process，morphological cases gradually get lost，due to phonological loss，which in turn causes the introduction of grammatical categories for replacement. For example，in earlier Latin，the elaborate case systems depended crucially upon distinctions in the final syllables of inflected nouns. As phonological changes began to reduce and obliterate the final syllables，prepositions came to be used more frequently to reinforce the case distinctions（Trask 1996：128）. Similar cases are found in English. As Lehmann（1992：15）states，Early English still contained examples of an instrumental case form；in late Old English，the uses of the instrumental were taken over by the dative. Through changes in final vowels，the instrumental was no longer distinct，and was lost as a separate case form. Later，as final endings were lost in English，the dative case itself was no longer distinct；most of its former functions are now expressed by prepositional phrases.
As for nominal morphology，simplification always means loss of inflectional affixes，like plural markers and morphological cases. The loss is certainly gradual，because the phonological change which causes the loss is by its nature gradual. As Dixon（1997：55）states：“While the evolution of categories happens rather quickly，their loss may be quite gradual. The loss of a category can be due to phonological attrition，which is by its nature a gradual process.” In contrast，the case in development of verbal agreement morphology is more complex. Since agreement markers are developed from full pronouns and are then fused with verbal stems，there should be several stages during the fusional process. In some languages，the agreement inflections retain their independent status，like in agglutinative languages，whereas in some other languages they have already fused with verb stems and have become agreement affixes. Afterwards，under the conflicts of phonological naturalness and morphological naturalness，agreement affixes tend to get lost；or more specifically，assimilation tends to happen among the agreement affixes.
In the following subsection，we will show some verbal agreement paradigms of different richness degrees，in order to illustrate the gradual evolutional tendency of verbal agreement morphology；and in section 2.3 we will show data for illustration of the development of nominal cases. In Chapter 3 and Chapter 4，more data from two main language families will be shown for further illustration.
2.1.3 Verbal agreement paradigms of different richness degrees
In this subsection，we will show some verbal agreement paradigms to illustrate the evolutional tendency of morphology，which is towards more fusional and then more simplified. As Ariel（2000：198）states：“…pronouns and inflections form a continuum from independent words to complete fusion：there are various intermediate stages.” In other words，except for the three main evolutional stages that the nineteenth century evolutionary theory classifies，i.e. isolating，agglutinative and fusional，the fusing process of agreement markers and verb stems can also be classified into several stages.
During the evolutional process of verbal agreement morphology，agreement markers tend to fuse with verb stems and finally become an inseparable part of verbs，i.e. inflectional affixes. The fusion in various languages might have reached different developmental stages at a specific time：in some languages the agreement inflections retain their independent status，whereas in some other languages they have already fused with verb stems. After the complete fusion，inflectional affixes tend to get lost under the conflicts of phonological naturalness and morphological naturalness，which thus causes the gradual simplification of morphological paradigms.
Considering that developments of verbal paradigms are regularly from complex to simple，it is natural that the original pattern of a verbal paradigm is rich in the extreme，with agreement affixes distinctively marking each person feature in the singular and plural，because of the distinctiveness of original personal pronouns. For example，Greek has a very rich verbal paradigm，which contains three numbers（singular，dual and plural），three voices（active，middle and passive），four moods（indicative，subjunctive，optative and imperative），and seven tenses（present，imperfect，future，aorist，perfect，pluperfect and future perfect）. The present，perfect，future，and future perfect indicative are called ‘primary’ tenses；while the imperfect，pluperfect，and aorist indicative are called ‘secondary’ tenses. The personal endings vary according to not only person and number，but also tense，voice and mood. Its verbal paradigm is shown in（3）（from Goodwin 1892：131）.
During the evolutional process，agreement markers tend to fuse with verb stems and gradually get lost under the conflicts of phonological and morphological naturalness. Or more specifically，affix assimilation tends to occur in a morphological paradigm. German is such an example；its verbal paradigm is given in（4）（from Hammond 1981：44）.
As the paradigm in（4）shows，the German verbal agreement inflections do not specify all persons and number distinctively：the second plural form and the third singular form are the same（–t），while the first plural and the third plural have the same form –en.
Under the influence of phonological change and the analogy process，assimilation among agreement inflections goes further and then agreement inflections gradually get lost. There are two alternatives with respect to the loss of agreement inflections：either some of the agreement inflections get lost and only a small number of them are retained，as in English；or agreement as a functional category is lost. For example，in the Modern Scandinavian languages，there is no agreement at all；a uniform inflectional affix is used for all persons and numbers in each tense. The loss of verbal agreement inflections can be demonstrated by the striking change of the Danish ththmorphology between the 11 century and the 14 century，as shown in the verbal paradigms in（5）（from Vikner 1997：206）.
2.1.4 Our hypothesis about ways of language evolution
We propose that there are two possibly alternative ways of language evolution（including morphological and syntactic change）：natural evolution and evolution under standardization. If language evolution follows a natural way，it should be gradual，since the phonological change which causes morphological change is by its nature gradual. As for language evolution under standardization，the developing speed is unpredictable. For example，Maiden（1995：9）attributes the expansion of Italian to political unification，so that Italian has gone through strikingly little change till now. The MSc. languages are also regarded as developed under standardization（Henriksen and der Auwera 1994：3），so that there is little difference in these languages while they differ in many aspects from the ISc. languages. As Henriksen and der Auwera（1994：3）states：“Without this concept（i.e. standardization）one would still not know why Swedish，Danish and Norwegian are considered different languages，even though mutual intelligibility is very high，whereas some northern and southern dialects of German，which are hardly mutually intelligible，are not considered separate languages.”
As mentioned above，morphological development is gradual，because the phonological change that causes morphological change is by its nature gradual. The gradual natural evolution of morphology implies that the changes of nominal morphology and verbal morphology may not be at the same stage at a given time：some languages may have rich verbal morphology but impoverished，or relatively impoverished，nominal morphology. For instance，Modern French has lost the nominal case distinctions，but it still has a rich verbal morphological paradigm. On the other hand，some languages might have undergone the simplification of verbal morphology，i.e.，there may be assimilation among the agreement inflections for some persons and/or numbers，but they may retain a rich morphological case system. Modern German is such an example.Moreover，in languages which have undergone gradual change，there are several intermediate stages of morphological development：some languages have fairly rich morphology；some have assimilation happening among the inflectional affixes；some others which have gone further in the developmental process have lost most of the inflectional affixes.
The case is different in languages which have changed under standardization. As Henriksen and der Auwera（1994：3）state，standardization is the process whereby a community imposes uniformity on its language，in response to a growing desire of political，religious or cultural authorities for improved communication across dialects. In this sense，the evolutional process of such languages does not follow a natural way. We assume that in such languages morphological development tends to be uniform，since its morphological change does not follow the routine，i.e.，it is not an immediate result of the conflicts of sound change and analogy.In other words，these languages have either fairly rich morphology or most impoverished morphology. For example，Italian has fairly rich verbal agreement morphology in that there are distinct agreement markers for all persons and numbers without any assimilation，whereas the MSc. languages have completely lost agreement and retain only tense affixes uniformly for all persons and numbers. The morphological development in the MSc. languages is evidently different from that in Modern English，which is also considered a most impoverished language. Modern English has lost the case distinctions and most of the agreement inflections in Old English and Middle English，but it retains one agreement affix –s for marking of the third person singular. Therefore，Modern English shows one of the intermediate stages in the gradually evolutional process，whereas the uniformity in Italian and MSc. shows that there is no intermediate stage in the development under standardization.
Corresponding to morphological development，syntactic change can also be categorized into two types. In languages developing in a natural way，the syntactic change is also gradual so that in these languages we often find residues of past constructions side by side with the current patterns. However，there is an obvious relationship between morphological change and syntactic change in these languages，so that we can explain their syntactic variation by virtue of an analysis of the morphological differences. In languages developing under standardization，on the other hand，the syntactic change may not be at the same speed as the morphological development. In other words，there may be no direct relationship between morphological development and syntactic change：some languages might have lost the rich morphological system but have retained the past sentence patterns of their ancestors. Therefore，the syntactic differences between the languages developing under standardization and the languages developing in a natural way cannot be explained by virtue of an observation of the morphological development. The analyses of the syntactic variation in the two types of languages will be given in Chapter 3 and Chapter 4.
2.2 Language Classification with Respect to Verbal Agreement Morphology
In this section，we will give a classification of languages in terms of verbal agreement morphology. We propose that languages can be roughly divided into two types in terms of the presence of the functional category Agr：languages with Agr and languages without Agr. As for the former type，there are two ways of subclassification. In terms of the adjunction position of Agr，it can be subdivided into C-Agr languages and T-Agr languages，whereas in terms of the agreement richness degree，it can be subdivided into the agreement-rich type and the agreement-impoverished type.
2.2.1 Presence of the functional category Agr
Since the classification of languages is in terms of verbal agreement morphology，we need first to discuss the presence of the functional category Agr. We assume that the presence of functional categories in a specific language，except that of the core functional categories（C，T and v），depends on its morphological representation. In other words，only when there are morphological features in the paradigm，which require relevant functional categories as landing sites，can other functional categories than the core functional categories exist in a language. Our assumption is based on Iatridou's（1990）view，which has been mentioned in Chapter 1，that languages may employ a different set of functional projections（also in Bobaljik and Thráinsson 1998；Platzack 1983；Fuss and Trips 2002）. We have introduced Holmberg and Platzack's（H&P，1995）view on the presence of the functional category Agr，in their argumentation of the distinction between ISc. and MSc.：the functional category Agr is present in ISc. and absent in MSc. Combining Iatridou's and H&P's views，we assume that Agr is present only if a language has verbal inflections expressing person/number agreement.
In Chomsky（1998），Agr is regarded as a category lacking interpretable features and is hence excluded from the list of functional categories. Chomsky's argumentation（Chomsky 1998：55）is as follows. Suppose α is a lexical item that consists of uninterpretable features only and selects β，yielding the syntactic object K=｛α，β｝ with label α，as shown in（6）. In the course of a convergent derivation，α will disappear，for it does not have interpretable features at LF. This will leave K and higher projections of α without a label. But terms without labels are not well-formed syntactic objects. Hence，Chomsky（1998）concludes that elements such as Agr cannot exist，since it is such a selector，which selects TP or VP，without interpretable features.（6）a. K=｛α，β｝ b.
Differently，in Holmberg and Platzack's（H&P，1995）framework，Agr is regarded as an inherently maximal functional head，which is adjoined to a projecting functional head. The reasons that Agr is not assumed as a sentential head are as follows（H&P 1995：19）. Firstly，if Agr is a sentential head，it should be a nominal category，since it encodes nominal features such as person，number and case. If this is true，we would find sentences which are bare，complementizerless，but finite（hence putatively Agr-headed）in positions where nominal categories occur. However，sentences which occur as arguments crucially require a ‘nominal’ complementizer（in English that）at the head of the sentence. This fact indicates that the bare sentence is not a nominal category. Secondly，if Agr is the head of a sentence，we would expect heads taking sentential complements to select properties of Agr，similar to cases in which sentences are selected according to finiteness，mood，or tense. However，the case does not happen in any language in which a verb would select a sentential complement with respect to features of Agr，such as second person，or plural，or feminine. Therefore，H&P conclude that Agr is not a sentential functional category，but an inherently maximal functional head，which must be adjoined to a projecting functional head.
In our understanding，H&P's（1995）hypothesis can solve the problem about the existence of Agr discussed in the Minimalist Program.Since Agr is adjoined to a projecting functional head，the projection including Agr would not be left without a label；its label is still that of the projecting functional head. It follows that the whole projection turns to be a well-formed syntactic object with interpretation at LF. In this sense，H&P's（1995）hypothesis avoids the problem of the interpretation of Agr at LF.
The existence of the functional category Agr is a major theoretical basis for our argumentation in this dissertation，in terms of the presence of which human languages are roughly classified into two types. We assume that the functional category Agr is present only when there is morphological requirement，i.e.，when a language has agreement affixes which requires the presence of Agr as their landing site. Therefore，if a language has agreement affixes，even if it has only one affix like Modern English，the functional category Agr is present in this language；otherwise，Agr is absent. For example，Chinese，Thai，Japanese and the MSc. languages have no agreement at all and hence the functional category Agr is absent. In a word，in terms of the presence of the functional category Agr，languages can be divided into two types：languages with Agr and languages without Agr.
Our view about the presence of Agr is different from Rohrbacher's（1999）. As mentioned in Chapter 1，Rohrbacher（1999）claims that the presence or absence of overt distinctive marking of the subject-verb agreement features［1st］and［2nd］person determines whether or not an AgrP will be projected in the derivation of a sentence. The presence of AgrP renders the possibility of V-to-I movement by providing a landing site for the verb to move to. If a language does not have distinct inflectional features，then no AgrP will be projected and the verb will remain in situ. In this dissertation，however，we assume that languages with agreement affix（es），regardless of the number of the affixes，all have the functional category Agr.
2.2.2 C-Agr vs. T-Agr
In terms of the adjunction position of Agr，languages with Agr can be classified into two subtypes：C-Agr languages and T-Agr languages. Based on H&P's（1995）suggestion that Agr is an inherently maximal functional head which is adjoined to a projecting functional head，we assume that Agr can be adjoined to C or T，and that languages where Agr is adjoined to C are named C-Agr languages，while languages where Agr is adjoined to T are named T-Agr languages.
The reason that we suppose that the functional category Agr can be adjoined to C is that in some languages，like Old Scandinavian，the verb agrees in person and number with the topic rather than the subject，as shown in（7）.（7）mér þótti vit vera í hellinumme.dat seemed.3sg we.nom be in the cave‘I thought we were in the cave.’（from Faarlund 1994：62）
In the Old Scandinavian example in（7），the verb þótti agrees in person and number with the initialized pronoun mér in dative case，which is in the topic position［Spec，CP］，rather than with the nominative pronouns vit. This fact thus indicates that the agreement in（7）is C-Agr rather than T-Agr.
In addition to the verb agreement with the initialized element，there is another kind of C-Agr：in some languages，agreement is manifested on the complementizer，as shown in the West Flemish examples in（8）.（8）0a. K weten dan-k（ik）goan weggoan. C+1sg（West Flemish）I know that-I I go leave‘I know that I am going to leave.’0b. K weten da-j（gie）goat weggoan. C+2sg0c. K weten da-se（zie）goat weggoan. C+3fsg0d. K weten da-tje（jij）goat weggoan. C+3msg0e. K weten da-me（wunder）goat weggoan. C+1pl0f. K weten da-j（gunder）goat weggoan. C+2pl0g. K weten dan-ze（zunder）goat weggoan. C+3pl（from Law 1991：254，cited in Hofherr 2003：48）
In the above West Flemish examples in（8），there is an agreement affix attached to the complementizer dan ‘that’，and the agreement affixes are similar to their corresponding personal pronouns. What immediately follows the agreement marker is a subordinate clause with its subject omitted，which is licensed by the complementizer agreement，according to Law（1991）. We assume that the agreement markers are inserted into C and are later fused with the complementizer. The change of the complementizer form（dan―da）provides evidence for this fusion. Furthermore，we assume that the agreement markers on the complementizer in（8）are developed from the corresponding personal pronouns which serve as the subordinate subject in these examples，because they have similar forms. Taking typologists' view that verbal inflections are developed from full pronouns and Ariel's（2000）gradual fusion hypothesis，which have been introduced in section 2.1，we conclude that the complementizer agreement in West Flemish have not completely fused with verbs and thus appear to adjoin to the complementizer. In this dissertation，we only consider agreement manifested on verbs；and we regard those languages in which the verb agrees with the topic rather than the subject，as shown in the above Old Scandinavian example in（7），as C-Agr languages.Moreover，since agreement in C-Agr languages are manifested on verbs which move to C，rather than on the complementizer agreement，we cannot use the term ‘complementizer agreement’ for C-Agr languages. Instead，we simply use ‘C-Agr’ to refer to the agreement feature in cases where the verb agrees with the topic.
2.2.3 Classification of agreement paradigms
We have proposed that languages can be roughly divided into two types in terms of the presence or absence of the functional category Agr：languages with Agr and languges without Agr. In this section，the classification of agreement paradigms is only for languages with Agr. Since there are several intermediate cases between languages with a fairly rich agreement system and languages without Agr at all，in particular among those languages having been developing in a natural way，we need to make an elaborate classification of agreement paradigms in terms of their richness degree.
Considering the complexity of agreement paradigms and the different syntactic effects that the simplification of agreement morphology causes，in this dissertation we will adopt two different kinds of classification of agreement paradigms，one with respect to the richness degree of agreement paradigms and the other with respect to the referential power of agreeing verbs. This is because the loss of the referential power of agreeing verbs in some languages has no explicit morphological manifestation. For example，in French，the agreement inflections have retained morphological distinctions but lost phonological distinctions.In addition，word order variation and subject omission are related to agreement morphology in different ways. Subject omission is determined by the strength of the referential power of agreeing verbs，while word order variation is related to verb movement to the functional category Agr. Considering this，we will adopt different kinds of classification in our analyses of word order variation and subject omission.
126.96.36.199 Classification in our analysis of word order variation
In our analysis of word order variation，we suggest that languages with Agr be classified into two subtypes in terms of the agreement richness degree：the agreement-rich type and the agreement-impoverished type. We consider languages with distinct agreement markers for all persons and numbers as agreement-rich languages，including those in which there is some assimilation among the agreement inflections. For example，Italian has six distinct agreement markers for three persons and two numbers，whereas in German some agreement inflections have fallen together. Nevertheless，in this dissertation，both are categorized into the agreement-rich type，since they have agreement inflections for all persons and numbers and the agreement inflections have distinctions. This means that the agreement paradigms of these languages are not quite simplified. As for languages which have lost most agreement inflections，like English，they are certainly agreement impoverished. In section 2.4.1，we will propose that languages with different richness degrees of agreement morphology have different affixation levels：in languages with rich agreement morphology affixation takes place in syntax via overt verb movement，whereas in languages with impoverished agreement morphology affixation takes place in the phonological component via affix lowering to the verb stem. Different affixation levels cause variation in verb movement and hence word order variation. Evidently，in C-Agr languages verb movement for affixation is up till C，whereas in T-Agr languages the verb moves to T to pick up affixes.
188.8.131.52 Classification in our analysis of subject omission
In our analysis of subject omission，both of the two kinds of classification of agreement paradigms will be adopted：one is with respect to the agreement richness degree，and the other is with respect to the referential power of agreeing verbs. The former is for analysis of licensing of subject omission，whereas the latter is for analysis of different null subject phenomena. This is because language classifications in terms of the licensing of subject omission and in terms of different null subject phenomena are different. In the former case，we classify languages into full-null-subject languages which allow all the three types of null subjects，semi-null-subject languages which allow only null expletives，and non-null-subject languages which do not allow any null subject. In the latter case，i.e.，in terms of different null subject phenomena，language classification is in correspondence to the three types of null subjects classified in Hofherr（2003），which has been mentioned in Chapter 1. In that case，we classify languages into full-referential languages which allow all the three types of null subjects，quasi-referential languages which allow null quasi-referential and null non-referential subjects，and non-referential languages which allow only null non-referential subjects. In other words，language classification in terms of null subject phenomena covers only languages which allow null subjects（no matter what kind of null subjects），whereas language classification in terms of the licensing of null subjects is applicable to all human languages.
As mentioned in section 2.1.3，Ariel（2000）claims that there are various intermediate stages during the developmental process of agreement inflections：pronouns and inflections form a continuum from independent words to complete fusion. Based on this，we suggest that the evolution of agreement morphology be divided into three stages in terms of the fusion degree of agreement markers and verbs. At the first stage，the fusion is not very obvious so that there are distinct agreement markers for all persons and numbers（some languages have the dual number）. Romanian and Italian are examples. At the second stage，the fusion degree becomes higher and partial assimilation appears among the agreement markers，as in French and German. During constant language development，agreement markers gradually lose their independent status and fuse with verb stems. Moreover，under continuous conflicts of phonological naturalness and morphological naturalness，morphological paradigms gradually become more simplified. Therefore，the third stage of morphological development is of the highest fusion degree，with the loss of agreement markers for some person（s）and/or number（s），as in English. Corresponding to the three fusional stages，we classify agreement paradigms into three types in terms of the agreement richness degree：fairly rich，degraded rich and impoverished. In our classification，languages with fairly rich agreement morphology allow all the three types of null subjects，like Romanian and Italian；languages with degraded rich agreement morphology allow only null expletives，like Icelandic and German；and languages with impoverished agreement morphology do not allow any subject omission，like English. Nevertheless，language classification is not so simple and definite，because of the complexity of agreement paradigms. For instance，in our analysis of word order variation，French is to be categorized into the agreement-rich type，because it has relatively distinct agreement markers for all persons and numbers. On the other hand，since there is some assimilation among its agreement affixes，it is evidently a language with degraded rich agreement morphology. However，French does not allow any kind of null subject，which means that it must be categorized into the agreement-impoverished type，which is evidently a contradiction. Considering this，we suppose that French be analyzed in terms of the referential power of its agreeing verbs. As will be shown in Chapter 3，the French agreement affixes have morphological distinctions but lost phonological distinctions，which thus entails the loss of their referential power. It means that French is at the intermediate stage between degraded rich and impoverished. In section 2.4.6 and Chapter 3，we will further discuss this issue.
In addition to the complex phenomena in French，languages with degraded rich agreement morphology also behave differently in licensing different kinds of null subjects. As we mentioned，there are two types of null expletives，null quasi-referential subjects and null non-referential subjects. Icelandic allows both of the two types of null expletives，whereas German allows only null non-referential subjects. In order to make a clear explanation of the different null subject phenomena，we will give a subdivision of degraded rich agreement morphology in the following subsection.
184.108.40.206 Subdivision of degraded rich agreement morphology
As mentioned above，the evolutional process of agreement morphology can be roughly divided into three stages with respect to the fusion degree of agreement markers and verb stems. Corresponding to the three fusional stages，agreement paradigms can be classified into three types with respect to the richness degree：fairly rich，degraded rich and impoverished. Nevertheless，this classification of agreement morphology is also not very definite，because the morphological evolution is gradual so that at each stage there may be some slight distinction.
For example，at the second stage，there are some languages which have different richness degrees of verbal agreement paradigms. As mentioned above，the French agreement inflections have distinct morphological representations but have lost phonological distinctions. Since morphological change is firstly caused by phonological naturalness，the loss of phonological distinctions indicates that the morphological paradigm in French is in process of change. In addition，at this stage，some languages like Icelandic have five out of six distinct agreement markers，whereas in some other languages，like German，the agreement inflections for some persons and numbers have fallen together. Considering this，we will give a subdivision of degraded rich agreement morphology；we suggest that the agreement richness degree is determined by the third person agreement inflection.
As mentioned in Chapter 1，Beard（1991，cited in Rohrbacher 1999：114）states that overt morphology for the third person（singular）is absent in many languages，which means that third person is universally the unmarked value of the person feature. Based on this，Rohrbacher（1999）proposes that the person features［1st］and［2nd］are responsible for V-to-I movement. In our understanding，however，the absence of overt morphology for the third person actually implies the slower development of the third person agreement markers. As Ariel（2000：198）states，the development of pronouns into inflections can take place at different times for different persons/numbers. And according to her Accessibility Theory（Ariel 2000：199），only first and second person referents are consistently highly accessible，and hence only they merit a grammaticalization process of reduction which ultimately results in the obligatory coding of first/second persons by highly accessible markers（the agreement morphemes）. Third person referents are not consistently highly accessible，and hence lower accessibility markers（pronouns，full NPs）are more suitable for coding them. Although the Accessibility Theory only talks about the distinction of the original development of agreement markers in first/second persons and third person，it actually indicates the slower development of the third person agreement markers.
The distinction between first/second persons and third person has evidence from the separate morphological systems of Old Scandinavian，which reflect the gradual process of re-grammaticalization of pronouns after the fusion of agreement markers and verbs. As Faarlund（1994：48-9）states，Old Scandinavian had three separate morphological systems. First，there was a system of personal pronouns proper，which included the pronouns used for the first and second persons and reflexives for the third person. Second，there was a system of third-person personal pronouns，which was used only in the singular and only for the masculine and the feminine genders. The third system contained third-person singular neuter and third-person plural in all genders. The three separate morphological systems show that morphological developments in first/second persons and third person are at different speed. In a word，the morphological evidence from Old Scandinavian and Ariel's（2000）Accessibility Theory indicate that distinct third person agreement markers are significant in reflecting morphological richness. In Chapter 4，we will provide more evidence by virtue of a morphological comparison of the Germanic languages. For example，the agreement morphology of Old Scandinavian was a little more simplified than that of Ancient Scandinavian，and the simplification is reflected on the assimilation of the third person singular with the second person singular. In addition，in Modern Icelandic only the third person singular form is identical to the second person singular，whereas in Modern German both the third person singular and the third person plural have fallen together with some other person agreement markers. Correspondingly，Modern German has more restriction than Modern Icelandic in subject omission：unlike Modern Icelandic which allow both of the two types of null expletives，it only has null non-referential subjects. In a word，we propose that languages with degraded rich agreement morphology be subdivided in terms of the assimilation of the third person agreement inflections，and such assimilation certainly has its syntactic effects，which will also be shown in Chapter 4. In Chapter 5，we will take the third person agreement marker as a significant factor in determining the strength of the referential power of agreeing verbs and in explanation of the three types of null subject phenomena.
2.3 Theoretical Background about Syntactic Effects of Morphological Change
It is generally accepted that morphological change has direct effects on syntactic change. For example，the loss of inflectional endings will cause the evolution of new categories，which in turn will cause the change of word order. Such morphological effects on syntax can be found even in morphologically rich languages like Latin. As mentioned in section 2.1.2，in earlier Latin the elaborate case systems depended crucially upon inflectional distinctions，which were later lost due to phonological changes. Then prepositions came to be used more frequently to reinforce the case distinctions（Trask 1996：128）. Correspondingly，the preferred SOV word order of earlier Latin was replaced by SVO word order（Trask 1996：177）.In addition to the change of word order from OV to VO，other syntactic effects of morphological development，e.g. the loss of V-to-C movement，the loss of empty topics，the loss of subject omission，etc.，can also be found in languages.
In this section，we will introduce the theoretical background on which our main proposals of the morphological effects on syntax are based. Since morphological development is by itself gradual，its influence on syntactic change are correspondingly gradual. Therefore，the study of the syntactic effects should be from a diachronic perspective. The following subsection is an introduction to Haeberli's（2000）historical study of the connection between morphological and syntactic changes，in which a diachronic comparison of agreement paradigms is made in the syntactic analysis. In this dissertation，we will follow the diachronic approach in our syntactic analyses.
2.3.1 Haeberli's（2000）diachronic study
Haeberli（2000）claims that there is a historical continuity among morphology，word order and subject omission：the loss of V2 sentence pattern in English is a consequence of the loss of null expletives and ultimately of a change in the verbal morphology. As illustrated in the examples in（9），Old English（OE）exhibited word order patterns which were reminiscent of the V2 phenomenon：fronting of some constituent often leads to subject-verb inversion and hence to a word order in which the finite verb occurs in second position.（9）a. ［Ðas gifu］sealde seo ceasterwaru on Tharsum Apollonio þamthis gift gave the citizens in Tharsus Apollonius thetiriscanTyrian‘The citizens of Tharsus gave this gift to Apollonius the Tyrian.’b. ［Him］geaf ða se cyngc twa hund gildenra pænegahim gave then the king two hundred golden pennies‘Then，the king gave him two hundred pence in gold.’c. ［On his dagum］sende Gregorius us fulluhtin his days sent Gregory us baptism‘In his time，Gregory sent us Christianity.’（cited in Haeberli 2000：88）
During the Middle English（ME）period，however，the frequency of V2 decreases considerably and，as observed by van Kemenade（1987），V2 starts being lost by around 1400.
Together with the existence of V2 structures，null expletives were licensed in Old English（OE（10））and Early Middle English（EME（11））（cited in Haeberli 2000：99-100）：（10）a. … þæt pro for mannum gesewen wære，þæt …that before people seen was that …‘…that it could be seen by people that …’b. And eft pro is awriten，þæt se bið awyrged，þe …