Casting a Minimalist Eye on Adjuncts [1 ed.] 9780367421939, 9780367822613, 9781032082486

Citation preview

CASTING A MINIMALIST EYE ON ADJUNCTS

Stefanie Bode

Routledge Studies in Linguistics

CASTING A MINIMALIST EYE ON ADJUNCTS Stefanie Bode

www.routledge.com

Routledge titles are available as eBook editions in a range of digital formats

9780367421939_Full cover.indd 1

11/25/2019 8:56:25 AM

Casting a Minimalist Eye on Adjuncts

This book offers a comprehensive account of adjuncts in generative grammar, seeking to reconcile the differing ways in which they have been treated in the past by proposing a method of analysis grounded in simplification based on Simplest Merge. The volume provides an up-to-date review of the existing literature on adjuncts and outlines their characteristic properties and the subsequent difficulties in adequately defining and treating them. The book compares previous attempts to account for adjuncts which have tended to use additional mechanisms or syntactic operations as a jumping-off point from which to propose a new way forward for analyzing them grounded in minimalist theory. Adopting an approach in the spirit of the strong minimalist thesis (SMT), Bode suggests an analysis of adjuncts that applies a minimalist approach based on theoretical simplicity, one which does not resort to extra mechanisms in capturing the empirical properties of adjuncts. Offering a comprehensive overview of research on adjuncts and foundational minimalist principles, this book will be of particular interest to graduate students and practicing researchers interested in syntax. Stefanie Bode is Adjunct Lecturer at the Georg-August University of Göttingen, Department of English Philology—Linguistics, Germany. Her doctoral thesis, on the verb ‘be’ in the English sentence structure, was published in 2003.

Routledge Studies in Linguistics

Time Series Analysis of Discourse Method and Case Studies Dennis Tay Heart- and Soul-Like Constructs across Languages, Cultures, and Epochs Edited by Bert Peeters Systemic Functional Political Discourse Analysis A Text-based Study Eden Sum-hung Li, Percy Luen-tim Lui and Andy Ka-chun Fung Systemic Functional Language Description: Making Meaning Matter Edited by J.R. Martin and Y.J. Doran Rarely Used Structures and Lesser-Studied Languages Insights from the Margins Emily Manetta Externalization Phonological Interpretations of Syntactic Objects Yoshihito Dobashi Approaches to the Study of Sound Structure and Speech Interdisciplinary Work in Honour of Katarzyna Dziubalska-Kołaczyk Agnieszka Kiełkiewicz-Janowiak, Magdalena Wrembel and Piotr Gąsiorowski Casting a Minimalist Eye on Adjuncts Stefanie Bode For more information about this series, please visit: www.routledge.com/ Routledge-Studies-in-Linguistics/book-series/SE0719

Casting a Minimalist Eye on Adjuncts Stefanie Bode

First published 2020 by Routledge 52 Vanderbilt Avenue, New York, NY 10017 and by Routledge 2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN Routledge is an imprint of the Taylor & Francis Group, an informa business © 2020 Taylor & Francis The right of Stefanie Bode to be identified as author of this work has been asserted by her in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Library of Congress Cataloging-in-Publication Data Names: Bode, Stefanie, author. Title: Casting a minimalist eye on adjuncts / Stefanie Bode. Description: 1. | New York : Taylor and Francis, 2020. | Series: Routledge studies in linguistics | Includes bibliographical references and index. Identifiers: LCCN 2019043417 | ISBN 9780367421939 (hardback) | ISBN 9780367822613 (ebook) Subjects: LCSH: Grammar, Comparative and general—Adjuncts. | Grammar, Comparative and general—Syntax. Classification: LCC P299.A32 B63 2020 | DDC 415/.7—dc23 LC record available at https://lccn.loc.gov/2019043417 ISBN: 978-0-367-42193-9 (hbk) ISBN: 978-0-367-82261-3 (ebk) Typeset in Sabon by Apex CoVantage, LLC

For Daniel

Contents

Acknowledgmentsviii 1 Introduction

1

2 What Are Adjuncts?

7

3 The Minimalist Framework 3.1 Basic Minimalist Assumptions  19 3.2 Merge and Structure-Building  25 3.3 Why Are Adjuncts a Challenge?  38

19

4 Approaches to Adjuncts 4.1 A Different Dimension  50 4.2 Labeling and Projection  57 4.3 A Simplification Ahead?  70

50

5 A New Proposal: Labeling Is Transfer 5.1 Adjuncts and Labeling  96 5.2 Adjuncts and Islands  120 5.3 Labeling, Head Movement, and Adjunction  130 5.4 A Final Remark  148

96

6 Conclusions and Outlook

162

Index173

Acknowledgments

I thank the following copyright holders and the authors for granting permission to reprint quotes used for argumentation: Cambridge University Press, Columbia University Press, De Gruyter, Elsevier, John Benamins, Oxford University Press, Peter Lang GmbH, Routledge, John Bowers, and Balazs Surányi. I am also grateful for the helpful open access policy of many publishers and MIT’s general copyright strategy concerning reprinting smaller quotes. I am basically indebted to every author referred to, and especially to those whose works I discuss in the book. They share the deep interest in the Minimalist Program and contributed to this book. A project like this would not be possible without them. Thanks also to my publisher, Routledge, for giving me the opportunity to present my ideas on adjuncts. On a personal note, I want to thank Elysse Preposi (Editor, Linguistics), Helena Parkinson (Editorial Assistant), Eleni Steck (Production Editor), and Ramachandran Vijayaraghavan (Project Manager) for their help. I also thank two reviewers for their helpful suggestions. I thank my family for supporting me and being there for me. Eventually, I want to express my deepest gratitude to Daniel Seely, for his invaluable comments and encouragement.

1 Introduction

Why study adjuncts today? Recent minimalist discussion elaborates on the architecture of the grammar and raises such questions as: what is part of narrow syntax or what belongs to the interfaces SEM and PHON? There is a lot of discussion in the generative approach to syntax and, broadly speaking, minimalism splits into two opposing views. Either syntax is taken to be deterministic, leading to a crash-proof situation at the interfaces, or syntax is basically free, and its output is subject to interpretation at the interfaces. The most recent debates build upon the latter view, since applying Merge freely simplifies UG in a crucial way. Adjuncts have been under debate in the previous framework of Government & Binding and, strikingly, the issue has not been settled under minimalist approaches either. Every textbook on syntax addresses adjuncts. Adjuncts actually are such a basic phenomenon of language that they need to be part of any introduction; nevertheless, we do not find a consensus among scholars, even at the textbook level. The way in which beginners of linguistics are confronted with adjuncts is particularly telling because the degree of consensus among linguists can often be judged by considering introductions into a subject. Textbooks discuss more or less the same properties of adjuncts relating to selection, theta roles, order, optionality, islandhood, and more, but the structural proposals look rather different. For instance, Koeneman and Zeijlstra (2017) use a V’-recursion so that, in terms of X-bar theory, adjuncts are placed in a X’-sister position, making it hard to distinguish them from specifiers and, more concretely, from subjects. Adger (2003) repeats the maximal level and thus resorts to segments of XP (again using X-bar terminology). He leaves room for a subject ‘agent’ to be projected in the specifier of V, and the most crucial aspect of his analysis is that he assumes an extra syntactic operation Adjoin that introduces adjuncts into the derivation (as opposed to other constituents which enter the derivation via Merge). Both textbooks use a recursion at a specific X-bar theoretic level, which is actually impossible to formulate in a purely minimalist framework. We will come back to this point later. Similarly, Radford (2006) uses the X’-level, with the

2  Introduction additional complication that, in his proposal, adjuncts and arguments basically occur in the same positions. The distinction between arguments and adjuncts gets blurred and, if it is reflected in the tree diagrams at all, only works in terms of X-bar-theoretic principles formulated in the previous Government & Binding framework. Thus, it is crucial that we observe that a frequent linguistic phenomenon as adjuncts does not receive a uniform treatment even at textbook level. Hence, it is not surprising that scientific research has not come to a general solution or consensus concerning adjuncts either. The need for a theoretical implementation of adjuncts is obvious. In the beginning of his discussion of adjuncts, Adger stresses that the mechanisms by which adjuncts are implemented into phrase structure still pose major research questions (Adger 2003: 111). After decades of research into the matter, this is a surprising fact. It should invite scholars to put adjuncts on their agenda again, because the theoretical and the empirical nature of the subject needs to be addressed in concert. Natural language makes frequent use of adjuncts. You can take any kind of linguistic data pool in any particular language and you will make this observation. In fact, it is surprising to see how little is left of a sentence if you strike out the adjuncts from an arbitrarily chosen set of sentences. This simple observation can be illustrated by looking at some everyday examples, taken from an instruction manual of a digital camera. (1) a. With the card’s label facing toward the camera back, insert it into the slot without touching the card’s contacts with your finger. b. If the battery is left in the camera for a prolonged period, a small amount of power current is released, resulting in excess discharge and shorter battery life. c. To save battery power, the camera turns off automatically after about 30 seconds of non-operation. d. If you want to zoom, do it before focusing. Apart from the observation that, let’s say in d., only two words are not part of adjuncts—namely, do it—we can make the following points. First, most of the adjuncts in (1) are internally structured as any other phrases are and some contain adjuncts themselves too. Second, adjuncts occur in various types of sentences: finite (see (1) b. and c.) or non-finite clauses, in declarative clauses, and interrogative clauses or imperatives (see (1) a. and d.), to name just a few. The reader can easily verify this him- or herself. Third, all adjuncts clearly add to the overall meaning conveyed by the sentences in (1). Fourth, various categories are used in (1) for adjuncts; that is, prepositional phrases, adverbs, finite sentences, infinitival sentences, and gerundial expressions, which all may occur in other, non-adjunct positions too. This can easily be verified by considering

Introduction 3 prepositional phrases, for instance, which occur in predicative (2), argument (3), and adjunct position (4), respectively. (2) a. She expected the man (to be) in the garden. b. There is a man in the garden. c. A man is in the garden. (3) a. They put the chairs in the garden. b. He was interested in the garden. c. He went into the garden. (4) a. She read a paper in the garden. b. She was working in the garden. c. The man in the garden is her friend. No speaker of English has any difficulties marking the adjuncts in the preceding examples. In fact, everyone has clear intuitions about what adjuncts are. Given the observation that adjuncts occur frequently in natural language and that they do not pose any particular problems for a competent speaker when it comes to distinguish them from other phrases in a structure, it is rather striking that linguists working in the tradition of generative grammar still do not agree on how to analyze these phrases. Why are adjuncts such a controversial issue? Under standard assumptions, adjuncts occur adjoined to other phrases. Chomsky (2004) describes adjuncts as being on a separate plane. What does this mean? How do adjuncts fit into a Merge-based syntax? What are the distinctive and defining properties of adjuncts? Do we need a specific operation called Pair-Merge for adjuncts (Chomsky 2004), or do proper accounts of adjuncts refer to labeling (Hornstein & Nunes 2008) and projections of features (Zeijlstra 2019)? How can we implement adjuncts while keeping to the most minimal version of a Universal Grammar (UG)? In the ideal case, there is just a single, simple structure-building device operating on linguistic expressions, generating non-adjuncts and adjuncts likewise. Basically, the title Casting a Minimalist Eye on Adjuncts describes the general aim of this book. The term minimalist actually has two meanings. First, it stresses the fact that the minimalist program investigates language from the perspective of what is needed conceptually to explain the human capacity for language. Under the minimalist view, the previous framework of Government & Binding, with its complex architecture, had to be radically minimized. The second meaning follows from the general scientific demand to keep a theory as minimal as possible. Casting a

4  Introduction minimalist eye on adjuncts thus means that we need to look at generative grammar to see how adjuncts have been handled in the past and how they are captured under recent minimal approaches. Since the advent of minimalism, scientific research essentially emphasized the need for simplification. Analyzing adjuncts by means of additional mechanisms or syntactic operations goes against the strategy of minimizing UG. This book attempts to clarify the phenomenon of adjuncts on empirical and theoretical grounds. In the spirit of the strong minimalist thesis (SMT), which reduces the computational part to simple set formation—that is to say, Simplest Merge—the ultimate goal of the book is to propose an analysis of adjuncts that does not resort to any extra mechanisms. We compare previous minimalist accounts of adjuncts in terms of simplicity and show that they complicate the grammar either by putting adjuncts in a separate dimension or by requiring featural triggers. It is noteworthy that linguists like to use adjuncts as providing evidence for the basic property of language, namely its recursive power. Adjuncts are unbounded in principle; they can occur in a stacked fashion and prepositional adjuncts frequently are used to illustrate the phenomenon of structural ambiguity (see (5 and (6)). (5) He saw the girl [with a friend] [on Monday] [in the evening] [at the party] [in the garden]. (6) He took a picture of a man with a camera. a. He took a picture with a camera (not with a cell phone). b. A man with a camera was photographed by him. Hence, adjuncts are pretty good candidates to exemplify that language is hierarchically structured. Therefore, it should come as a surprise that after decades of research in generative grammar, there is no consensus as to how adjuncts should be analyzed. How can it be that a phenomenon that illustrates the basic property of human language so well is still so poorly understood? The most recent research, which lays a focus on minimizing UG (see, for instance, Chomsky 2013, 2015, 2017; Epstein, Kitahara & Seely 2014, 2015a, 2015b, 2016, 2017a, 2017b; Chomsky, Gallego & Ott 2019), is promising and exciting. It is only natural, therefore, to finally include the phenomenon of adjuncts in those considerations and search for the simplest integration of these long-debated syntactic objects into minimalist analyses. Casting a minimalist eye on adjuncts while keeping track of theoretical simplicity forms the methodological strategy applied in this book. It is hoped that this book contributes to the most recent research in minimalism because it tries to fill an obvious gap, namely, to come to grips with adjuncts, which are a controversial though extremely frequent phenomenon of language, as stressed in the beginning of this introduction. The book has two main objectives. First, we want to show that the existing accounts of adjuncts need additional mechanisms, a strategy that

Introduction 5 is not consonant with the simplest version of minimalism. Second, we suggest a new analysis that accords with the latest minimalist reasoning. The book is structured into six chapters. After the introduction in the present chapter, Chapter Two presents a characterization of the main properties of adjuncts. Adjuncts differ from arguments in specific ways and though most readers are familiar with the basic facts, the data are included here to stress the necessity to find a minimalist analysis that captures the dichotomy. After the empirical setting we will be concerned with the minimalist framework in Chapter Three. Here we focus on basic assumptions, the structure-building operation Merge, and the question of why adjuncts pose a particular challenge to minimalist theorizing. It is worth mentioning that the book aims to address readers with different backgrounds and different degrees of expertise. Consequently, readers who are familiar with the problems posed by adjuncts and with the most recent minimalist developments might decide to skip Chapters Two and Three and move on to Chapter Four. The underlying idea of Chapters Two and Three is just to highlight central issues of the phenomenon and the framework we use before we get to specialized discussion in Chapter Four and advance a new proposal in the most current minimalist terms in Chapter Five. Chapter Four addresses specific approaches to adjuncts in minimalism. We will present the different accounts and debate the pros and cons of the respective approaches. We also attempt to clarify what an account should look like under a minimalist point of view. The different accounts are ordered and classified in sub-chapters according to their basic reasoning. First, approaches are discussed that put adjuncts into a different dimension, followed by accounts that are based on projection and labeling. Since very recent minimalist discussion elaborates on labeling and its role in a free and simple syntax, we will focus on approaches that touch upon simplification in the last sub-chapter of Chapter Four. In Chapter Five, we will finally put forth a new analysis of adjuncts referring to labeling and Transfer that is based on Simplest Merge. The gist of the analysis is that adjuncts are assembled by simple set formation just as other linguistic expressions are and the resulting structure underlies labeling conceived of as Transfer relevant to the interpretation of sets at the interfaces. We will also deal with the topic of islands here, which not only has a long tradition in generative grammar, but also relates to the phenomenon in question because adjuncts exhibit island effects crosslinguistically. The goal is to put adjunct islands into the larger context of islands and formulate a generalization in line with the most recent minimalist reasoning. In a last step, we address the issue of head movement which is related to our discussion though it is a controversial subject of its own and try to advance a potential implementation into the proposal based on labeling. A final remark on the project closes this chapter. In Chapter Six, we will summarize the most important conclusions and investigate some ideas pointing to future research.

6  Introduction

References Adger, D. 2003. Core Syntax: A Minimalist Approach. Oxford: Oxford University Press. Chomsky, N. 2004. Beyond Explanatory Adequacy. In Belletti, A. (ed.), Structures and Beyond: The Cartography of Syntactic Structures, Vol. 3, 104–131. Oxford: Oxford University Press. Chomsky, N. 2013. Problems of Projection. Lingua 130: 33–49. Chomsky, N. 2015. Problems of Projections: Extensions. In di Domenico, E., Haman, C. and Matteini, S. (eds.), Structures, Strategies and Beyond: Studies in Honor of Adriana Belletti, 3–16. Amsterdam and Philadelphia: John Benjamins. Chomsky, N. 2017. Generative Linguistics in the 21st Century: The Evidence and the Rhetoric. A Lecture at the University of Reading, 11 May 2017. Chomsky, N., Gallego, A. and Ott, D. 2019. Generative Grammar and the Faculty of Language: Insights, Questions and Challenges. To appear in Gallego, A. and Ott, D. (eds.), Generative Syntax: Questions, Crossroads, and Challenges. Special Issue of Catalan Journal of Linguistics 2019. Epstein, S. D., Kitahara, H. and Seely, T. D. 2014. Labeling by Minimal Search: Implications for Successive-Cyclic A-Movement and the Conception of the Postulate “Phase”. Linguistic Inquiry 45: 463–481. Epstein, S. D., Kitahara, H. and Seely, T. D. 2015a. What Do We Wonder Is Not Syntactic? In Epstein, S. D., Kitahara, H. and Seely, T. D. (eds.), Explorations in Maximizing Syntactic Minimizing, 222–239. New York and London: Routledge. Epstein, S. D., Kitahara, H. and Seely, T. D. 2015b. Explorations in Maximizing Syntactic Minimizing—Routledge Leading Linguists. London and New York: Routledge. Epstein, S. D., Kitahara, H. and Seely, T. D. 2016. Phase Cancellation by External Pair-Merge of Heads. The Linguistic Review 33: 87–102. Epstein, S. D., Kitahara, H. and Seely, T. D. 2017a. Is the Faculty of Language a “Perfect Solution” to the Interface Systems? In McGilvray, J. (ed.), The Cambridge Companion to Chomsky (sec. ed.), 50–68. Cambridge: Cambridge University Press. Epstein, S. D., Kitahara, H. and Seely, T. D. 2017b. Merge, Labeling and Their Interactions. In Bauke, L. and Blümel, A. (eds.), Labels and Roots, 17–45. Boston and Berlin: De Gruyter Mouton. Hornstein, N. and Nunes, J. 2008. Adjunction, Labeling, and Bare Phrase Structure. Biolinguistics 2: 57–86. Koeneman, O. and Zeijlstra, H. 2017. Introducing Syntax. Cambridge: Cambridge University Press. Radford, A. 2006. Minimalist Syntax Revisited. Arizona State University, Online Manuscript. Zeijlstra, H. 2019. Labeling. Selection and Feature Checking. To appear in Smith, P., Mursell, J. and Hartmann, K. (eds.), Agree to Agree: Agreement in the Minimalist Programme, 137–174. Berlin: Language Science Press.

2 What Are Adjuncts?

The word adjunct stems from the Latin word adiungēre. The past participle is adjunctus, which means something like joined to. In standard dictionaries, we normally find descriptions such as something added to another thing but not essential to it, or, similarly, something added or connected to a larger or more important thing. This simple definition, taken literally, actually turns a lot of things into adjuncts; for instance, a trailer adjoined to a bike, a backpack adjoined to a person, a porch adjoined to a house, a plant adjoined to a pillar, a bay window adjoined to a house, and so forth. Basically, you may come up with many wild and funny instances of adjuncts. Although this may seem trivial, all examples we might think of have something crucial in common: if one adjoins a trailer to a bike, the bike remains a bike, and the same logic applies to any of those examples. So one gets the simple equation X + Y = Y, making X somewhat invisible. What about adjuncts in language? In fact, we often find statements about adjuncts in the grammar that are quite similar. The Cambridge online dictionary, for instance, defines an adjunct as a phrase which is not necessary to the structure of the clause, but which adds some extra meaning to it. How can we interpret this statement? A phrase is clearly a syntactic object. When it is not necessary to the structure, does this mean that it is not visible in syntax proper? Since it adds meaning, it must be visible in semantics. Given the overall architecture of a grammar, one standard assumption is that semantics interprets the output of syntax. This, however, makes the statement above seem rather contradictory. How can anything be visible to semantics, then, if it was not visible to syntax before? Those adjunct phrases need to be built in the first place. Still, this statement and the simple equation also have some intuitive appeal when we look at concrete linguistic data. (1) a. The temperature was rising [AdvP very rapidly]. b. He had [AdvP extremely rarely] played cards [PP with his friends]. c. [small clause The letter written], he [AdvP happily] went to the post office.

8  What Are Adjuncts? d. A [AP large] crowd [AdvP patiently] waited [infinite clause to pay their respects]. e. He was drying his hands [PP on a dirty dish towel]. f. He greeted his mother [AdvP furiously] [PP with a rude ‘hello’]. g. [PP With a little training] he might handle it [PP in a month]. h. He [AdvP carefully] studied the envelope [finite clause before he opened it slowly]. The expressions in boldface are obvious examples of complete phrases. We can figure out their categories, as indicated in the brackets above and can assign internal structures to the expressions in question. There can be full sentences or smaller ones, finite and infinitival sentences, prepositional phrases embedding a nominal expression, and even multiple adjuncts in a single clause as in (1) b., c., d., g., and h. Many of these adjuncts occur in a final position; some appear clause internally near the verbal phrase; and one adjunct, in (1) g., occupies a position in the leftperiphery. In all examples, the adjuncts contribute to the semantic content of the overall meaning. A preliminary characterization of adjuncts therefore includes the following properties: adjuncts can occur in various positions, multiple occurrences are frequently attested, and they have an expressive impact on the semantic interpretation of the proposition even though they can be left out without yielding ungrammatical results. In a certain way, then, it makes sense to say that an adjunct is a phrase that is not necessary to the structure of the clause, but which adds some extra meaning to it. From a theoretical point of view, of course, this is no satisfying description, and we must look further into the data to get some hints about the very nature of adjuncts. Before we investigate the properties of adjuncts more carefully, a brief comment on the terminology is needed. Basically, quite a few terms are used for the syntactic objects in question. Besides adjunct, we find the terms non-argument, modifier, and adverbial, with some additional specification, such as adverbial of manner, place, and time, and so forth. First, it should be stressed that this terminology is not really helpful because it either relates to a non-property—just telling us what adjuncts cannot be—or focuses on the semantic effects of adjuncts, namely, modifying the meaning of an event or state referred to in a proposition. Again, this does not help us to capture adjuncts syntactically. The most neutral term reflecting the very problem we face when it comes to integrate those elements is adjunct because, as we observed before, it leads one to think about the mathematically weird equation of X + Y = Y. Unless X is always null, we cannot make any sense of it. Hence, the most urgent problem for a syntactic analysis of adjuncts from this perspective is coming to terms with this always being null status of the equation and investigating whether we can make sense of it considering the adjunct data. The next logical step is to examine the basic properties assigned to adjuncts.

What Are Adjuncts? 9 Readers familiar with the empirical situation may choose to go on to the next chapter. Adjuncts show many properties that are crucially not shared with arguments. This has led to the assumption that there is a quite general asymmetry between arguments on the one hand and adjuncts on the other hand. The most frequently mentioned property can be observed in the following data. (2) a. b. c. d. e. f.

John found the letter in the garden. John found the letter. *John found in the garden. *slept in the garden. John slept in the garden. John slept.

Adjuncts are generally optional; that is, they can be omitted without having a degrading effect on the sentence. In particular, as can be seen when being compared to arguments, they are not selected by the predicate. The theta-marked internal argument the letter cannot be omitted in (2) c. without making the sentence ungrammatical, and the same applies to the external argument John in (2) d. Irrespective of the type of predicate used, whether transitive or intransitive, the adjunct in the garden is not required for the felicity of the sentences. This is known to be one of the basic properties of the adjunct argument dichotomy attested in natural language. It needs to be clarified, though, what optional or non-­obligatory means here. The next set of examples needs to be considered as well. (3) a. b. c. d.

He is eating an apple. He is eating. Eat the apple! To eat apples is healthy.

It is a well-known fact that one can omit the object in (3) b. with verbs like eat. It is somehow implicit in the meaning of the verb that something is eaten. That is why those objects are called cognate objects. If you substitute eat with devour, the sentence becomes weird. Obviously, this optionality is restricted to a fixed class of lexical items. It is not a general property of internal arguments. This differs strikingly from adjuncts, which are optional regardless of the specific lexical predicate. As can be observed in the data in (3), one can also leave out the external argument. The example in (3) c. and the infinitival subject clause in (3) d. lack an overt subject. However, this is again different from the non-obligatory

10  What Are Adjuncts? status of adjuncts, because subjects can only be unpronounced under specific structural conditions. (3) c. is an imperative sentence, and clauses that are not marked for finite tense lack an overt subject. Consequently, arguments may only be omitted under specific lexical or structural conditions, whereas adjuncts are non-obligatory in a more general sense. Do obligatory adjuncts exist? Some verbs or items seem, in fact, select for adjunct phrases. (4) a. b. c. d.

He worded the letter carefully. *He worded the letter. He makes a good teacher. *He makes a teacher.

Nevertheless, one cannot assume that data like those in (4) are evidence against the general asymmetry found with arguments and adjuncts in terms of the non-obligatory status of the latter.1 If we use lexical items that are very similar in their respective meaning to the verbal expressions used in (4)—for instance, wrote in (4) a.–b. (i.e., he wrote the letter (carefully)) and became in (4) c.–d. (for example, he became a (good) teacher)—the manner adverb carefully and the attributive adjective good are optional again. A further characteristic of adjuncts setting them apart from arguments shows up in the following sentences. (5) a. b. c. d. e. f. g. h. i.

John found the letter in the garden after dinner. *John found the letter the book. John sent Mary a letter. *John sent the boy Mary a letter a book. He has a dog. He has a small dog. He has a small nice dog. He has a small nice white dog. He has a small nice white dog with a brown spot around one eye.

A predicate only allows for a restricted number of arguments, as can be seen in (5) b. and d. To save those sentences, one would have to insert coordinating conjunctions like and or or to form John found the letter and the book or John sent the boy or Mary a letter or a book. Multiple adjuncts like in (5) a., on the other hand, are unbounded in principle. Thus, we can easily add further adjuncts here. This property of adjuncts is quite obvious in (5) e.–h. too, in which the attributive adjectival expressions can occur in an iterated fashion and one can have prepositional

What Are Adjuncts? 11 adjuncts too ((5) i.). The possibility of internal structure of multiple adjuncts is shown in (6). (6) He is a [[remarkably [smart]] [American] linguist]. This is [a [certainly [cleverly stated]] [convincing] argument]. John met her [apparently [[very often] at 9 pm]]. He left the room [surprisingly [reluctantly]] [[[yesterday [morning]] after breakfast]]. e. A [very nice] [young] man [working at [the corner [shop [in the [neighborhood [where you can find many other shops too]]]]]].

a. b. c. d.

Interestingly, there can be adjuncts that are structurally related to other adjuncts. In all examples in (6), we can observe a certain degree of embedding of adjuncts as indicated by the brackets. For instance, remarkably modifies smart in (6) a. In the same way certainly may refer to cleverly stated and cleverly is related to stated in (6) b. In sum, there can be more than one adjunct in a structure. Adjuncts are unbounded in principle; they can be added in an iterative fashion and can even occur adjoined to each other. Arguments, on the other hand, have a maximum number. It is a well-known fact that three arguments at most may occur in a simple predication. In fact, linguists like to use adjuncts to provide evidence for the recursive powers attested with the structure-building property of human language. This can be illustrated with the following example too, in which the first infinitival purpose-clause functions as an adjunct to the matrix predicate, and the second acts as an adjunct to the predicate of the first adjunct. (7) The professor sent an assignment to his colleague [1 to make a copy for his student tutor [2 to print it out for the students]]. A further property concerns the phrasal behavior of structures with adjuncts, as demonstrated by the sentences in (8). (8) a. b. c. d. e. f. g. h. i.

He could go to the party with a friend. . . . and go to the party with a friend he did. . . . and go to the party he did with a friend. . . . and Mary did so too. . . . and Mary did so with her brother. * . . . and go he did to the party. * . . . and Mary did so to the concert. What John will do is repair the car perfectly. What John will do perfectly is repair the car.

12  What Are Adjuncts? Adjuncts may undergo fronting with the phrase to which they are adjoined and they may be elided with the phrase or they can be stranded. Hence, adjuncts behave as if they were inside or outside the phrase in question. VP-preposing and VP-ellipsis show that, in contrast to adjuncts, arguments ((8) f.–g.)) are contained in the verbal phrase. VP-ellipsis; VPpreposing; and, finally, Pseudo-Cleft sentences as in (8) h.–i. can target VPs with or without the adjuncts, namely, “as if these were not there” though they still belong to the V-projections.2 Consequently, this presents another reason for assuming that there is a principled asymmetry between adjuncts and arguments. A similar observation contributing to the general adjunct–argument dichotomy can be made in (9). (9) a. b. c. d.

He will (definitely) go home. He has (definitely) gone home. He is (definitely) going home. *He has (definitely) going home.

The phrase behaves the same with or without the adjunct. This means that the selectional properties are not altered by the presence of adjuncts. Hence, adjuncts do not intervene with the selectional dependencies attested in the data above. The modal will requires a bare infinitive, perfective have goes with a past participle, and the progressive needs an —ing form in its complement. The phrases in question behave as if the adjunct was not there. The sentences in (10) exemplify a frequently mentioned property concerning the positions available for adjuncts. (10) a. b. c. d. e.

After midnight John doesn’t eat anything. John doesn’t eat anything after midnight. John thoroughly considered his options. John considered his options thoroughly. *John considered thoroughly his options.

The adjuncts in (10) either precede the VP/TP or they follow those phrases in (10) a.–d., while the internal argument does not show this kind of flexibility but has to occur in a sister position of the verb, as (10) d. illustrates. Some prepositional adjuncts can switch positions easily, as (11) shows. (11) a. He met her on Monday in the park. b. He met her in the park on Monday.

What Are Adjuncts? 13 The data in (12) illustrate how adjuncts interact and go hand in hand with a specific reading that relates to the functional layering of clauses. (12) a. b. c. d. e. f.

John probably has intentionally reviewed the article carefully. Fortunately they usually do their homework thoroughly. He might perhaps again be traveling to this country. He perhaps might again be traveling to this country. She cleverly solved the problem. She solved the problem cleverly.

In (12) a.–d., we observe that probably attaches to the highest layer and can be paraphrased by it is probably that . . ., intentionally is speaker-­ oriented, usually applies to the aspectual domain of the sentence, and thoroughly is the lowest adjunct and, as a manner adverb, is connected to the verbal action. The examples in (12) e.–f. demonstrate that specific positions of adjuncts correlate with the semantic reading we get. Although cleverly functions as a modal adverb (it was clever that . . .) in (12) e., the same adverb plays the role of a manner adjunct in sentence final position in (12) f. To sum up, roughly speaking, adjuncts either occur at the left or at the right periphery of the phrase they modify. In particular prepositional adjuncts are not strictly ordered with respect to each other. Though there is some flexibility, one can also observe a (semantically relevant) order of adjuncts, which tend to go hand in hand with the layers of sentence structure. The order of manner—subject-oriented/temporal/aspectual—speaker-oriented/modal/pragmatic adjuncts are broadly linked to the vP-TP-CP clausal spine, respectively. In this connection, it makes sense to compare with adjuncts in nominal expressions, as in (13). (13) a. b. c. d. e. f. g.

We firmly believe it./our firm belief They frequently visit him./their frequent visits This is absolutely strange./an absolute stranger He smokes a cigarette after dinner./his after dinner cigarette He probably quickly left./his probable quick departure *He quickly probably left./*his quick probable departure He went home after he had finished his tasks. /a man who had finished his tasks

There is a striking parallel between adjuncts occurring in a sentence and adjuncts occurring in a nominal expression, which suggests that a similar layering might be possible. Alexiadou (2013) also stresses the parallels between adverbial and adjectival modification not only in terms of their distribution, but also in relation to their respective form. In some

14  What Are Adjuncts? languages, for instance, we find the same form for both (fast in English), or modifiers are transparently derived from each other (Alexiadou 2013: 460). Note also that the categories for adjuncts involved in a sentence— AdvP, PP, CP—correspond to AP, PP, CP in a nominal expression.3 It is also noteworthy that neither category is restricted to adjuncts. One can easily check this in (14), for instance, for CPs occurring in object and subject positions too. The same applies to the other phrases. (14) a. John knows [that he learned a lot]. b. [That he learned a lot] is obvious. Another often cited characteristic property involves the reading of nominal expressions occurring inside adjuncts. The relevant data is given in (15). (15) a. b. c. d.

[Which plan [that Maryk had]] did shek discuss? [Which plan [that Maryk would go abroad]] did she*k discuss? [Which pictures [that Johnk took]] did hek sell? [Which pictures [of Johnk]] did he*k sell?

The R-expression Mary occurring within a relative clause in (15) a. and c. can be co-referential with the pronoun located in the matrix clause, which suggests that R-expressions within adjuncts do not interfere with binding condition C, while R-expressions that are located inside a complement ((15) b. and d.) cannot be construed with pronouns in the matrix clause. This observation is known as the (anti-)reconstruction effect (in the case of adjuncts): (16) a. [Which plan [that Maryk would go abroad]] did she*k discuss < which plan that Maryk would go abroad>? b. [Which pictures [that Johnk took]] did hek sell ? The R-expression in (16) a. is c-commanded by a pronoun with matching features and, if co-referential, would violate binding condition C, whereas this is not the case with b. because of the assumption that adjuncts like the relative clause presented earlier do not reconstruct. However, this needs to be investigated further, as the following examples prove. (17) a. [Go home [after Johnk arrived]] he*k did. b. [These pictures [of Johnk]] seem to himk to be on sale.

What Are Adjuncts? 15 The adjunct in (17) a. should not reconstruct; still, predicate-preposing of the VP containing an adjunct does not bleed binding condition C in (17) b. Contrary to what would be expected if complements would reconstruct in general, an R-expression contained in a DP undergoing A-movement (as opposed to A-bar-movement) does not violate condition C in the reconstructed position. Consequently, the data is trickier than a simple (non)-reconstruction property correlated with a (non)-argument property would predict. Another well-known fact about adjuncts again has to do with movement. The data in (18) exemplify displacement facts about adjuncts. (18) a. b. c. d. e. f. g. h.

Where did you meet him? How did you solve the problem? Why did you leave the room? When did you tell her about it? How did you say that you fixed the sink? How could you say that John fixed the sink?4 When will you say that he told her about it? When do you think that he will tell her about it?

The adjuncts where, how, why, and when can be displaced, as other phrases can, by A-bar-movement. The movement can cross clausal boundaries just as with arguments. As (18) e. and f. and (18) g. and h., respectively, demonstrate, both downstairs and upstairs readings are attested. Hence, adjuncts have the property of undergoing (long) whmovement, which clearly does not distinguish them from arguments that can also be A-bar-moved. Note, however, that Delfitto (2006) mentions a class of adverbs including items such as too, even, and almost, which cannot undergo A-bar-movement. Hence, those particle-like items would need a closer inspection but will be ignored here. Only arguments take part in A-movement. Since raising and passive are restricted to DPs, it is not surprising that adjuncts displaying other categories cannot participate in A-movement. Yet DP adjuncts cannot undergo A-movement, unlike arguments, as (19) c. shows. (19) a. *Yesterday was worked.5 b. *The garden was worked in. c. The man seems () to be liked . The example in (19) b. may also follow from the next property. A further observation has been made on the basis of data such as the following.

16  What Are Adjuncts? (20) a. b. c. d.

*Which concert did she meet him [after ]? Which concert did they send him [to ]? *Who did he go home [before arrived]? Who did he say [ arrived]?

Moving a phrase out of an adjunct yields an ungrammatical result, as evidenced by the sentences in (20) a. and c. This is not the case with internal arguments, as you can see in (20) b., and with external arguments as well, as (20) d. illustrates. In general, then, adjuncts form islands to extraction. Still, some data require deeper investigation. (21) a. Who did John leave the party [with ]? b. Which song did he arrive [humming ]?6 Though the wh-phrase occurs within an adjunct, extraction is possible. Nevertheless, most of the data are uncontroversial and point in the direction of islandhood with respect to adjuncts. Whether it is correct to consider this property as belonging to the argument-adjunct dichotomy, though, is to be examined more thoroughly, because subjects also form islands to extraction, in contrast to objects, as (22) illustrates. (22) a. *Who did [a picture of ] please her?7 b. Who did you take [a picture of ]? Taking stock, we observe that, in general, adjuncts differ strikingly from arguments in various ways and that they can behave syntactically as if they were invisible. They also exhibit some flexibility in linear order and are unbounded in principle. The following overview visualizes the results so far. The empirical data show that it makes sense to speak of an adjunct–argument asymmetry or a clear dichotomy between those two functions, although they also share some of the properties listed in Table 2.1. The basic question is how to implement adjuncts in a minimalist setting. How do they enter the derivation? How can we capture their invisibility to some syntactic operations on the one hand and their respective existence at the interfaces on the other hand? Of course, adjuncts contribute to the meaning semantically (at SEM) and have to be linearized and pronounced (at PHON). Consequently, one has to think about adjuncts on empirical and theoretical grounds in connection with the overall architecture of the grammar. Before we investigate how recent

What Are Adjuncts? 17 Table 2.1 Overview: Adjuncts and Arguments Adjuncts

Arguments

not selected (optional) unbounded in principle, multiple instances recursively or iteratively added “as if not there”- behavior - in VP-preposing/VP-ellipsis and Pseudo-cleft-sentences - no selectional intervention effects various positions - relatively flexible - readings: structural layers - parallel between adjuncts in clauses and those in DPs behavior under wh-movement - no reconstruction - no extraction from adjuncts - (long) wh-movement no A-movement

selected (not optional) limited in number (depending on the predicate), general maximum is three arguments no “as if not there” behavior

fixed positions

behavior under wh-movement - reconstruction - extraction from objects - (long) wh-movement A-movement

approaches deal with adjuncts, we will consider the basic assumptions of the minimalist framework, which are essential to the investigation of the question of how to implement adjuncts in minimalist terms.

Notes 1. A reviewer rightly stresses that the manner adverb actually is not needed in this case either and provides a context to the example in (4) b, such as I know what I want to say in the headline, but I haven’t worded it (yet). Hence, this basically underlines the general optionality of adjuncts. 2. Using a projectional system, one could also assume that only V-projections but not V-heads can be moved (Daniel Seely p.c), but a solution like this won’t work in a projection-free syntax. 3. In German, predicative adjectives and adjuncts in an adverbial function are formed with the same items, but attributive modifiers agree in number and case with the nominal expression to which they refer. Despite this difference in inflection, it is striking that adjectival expressions are used as adjuncts in DPs and clauses in German as well. 4. In the examples (18) f. and g., an upstairs reading of the respective adjuncts is strongly favored because of the affinity between how-could and when-will. 5. In German impersonal passives, like Gestern wurde gearbeitet (‘yesterday was worked’), the adjunct does not occupy an argument position but is internally merged to C, hence in a non-A position. 6. Adding phrases, as in i–ii, makes data like (21) less felicitous, again stressing the rather robust island property of adjuncts: i. *Who did John leave the party yesterday night with? ii. *Which song did he arrive at the station yesterday humming?

18  What Are Adjuncts? 7. Also, one has to further check the cross-linguistic validity of (22) a. and b. In German, it is possible to extract from external and internal arguments, but not—and this is crucial—from adjuncts. i. ii. iii.

Von wem gefiel ihr ein Foto? (Of whom did a picture please her?) Von wem machtest du ein Foto? (Of whom did you take a picture?) *Welchen Kuchen kam er an bevor sie hereinbrachte? (*Which cake did he arrive before she brought in?)

References Alexiadou, A. 2013. Adverbial and Adjectival Modification. In den Dikken, M. (ed.), The Cambridge Handbook of Generative Syntax, 458–484. Cambridge: Cambridge University Press. Delfitto, D. 2006. Adverb Classes and Adverb Placement. In Everaert, M. and van Riemsdijk, H. (eds.), The Blackwell Companion to Syntax, Vol. 1, 83–120. Malden, MA, Oxford, and Carlton: Blackwell Publishing.

3 The Minimalist Framework

Up to this point, we focused on the empirical side of the phenomenon of adjuncts. Before we can elaborate on the question of how best to deal with adjuncts in theoretical terms and enter the scientific debate, it is necessary to sketch out the theoretical framework of minimalism; that is, what are the major assumptions and goals, how are structures assembled in natural language, and in what way do adjuncts pose a challenge to a minimalist setting? Minimalism is both a theoretical framework with a tradition in generative grammar and a scientific method aiming at simplicity and principled explanation. This duality is important to keep in mind as we go on.

3.1 Basic Minimalist Assumptions The five questions to begin with are coined by Chomsky, after the scholars who raised the respective issue. We might paraphrase the issues (without a particular ordering) along the following lines: i. Humboldt, or making infinite use of finite means; ii. Plato, or going beyond the input; iii. Darwin, or finding the distinctively species specifics; iv. Gallistel, or linking mind and brain; and v. Descartes, or creating mindful uses. Those five questions, taken seriously, have a strong impact on how we should conceive of what is called the human faculty of language. Take a look at Figure 3.1. Crucially, we first have to find an answer to the question of what language is before we can address the other questions. If you don’t know what it is, you cannot give reasonable answers to the other questions. UG plays a decisive role in answering those questions properly. It basically defines what is part of human language in a way that makes it possible to explain how it is acquired by the individual and how it evolved in the human species. A basic minimalist setting, thus, embodies the following view on UG.

20  The Minimalist Framework

What is it?

How is it related to the mind?

How is it acquired? Human Language

How did it evolve?

How is it linked to the brain?

Figure 3.1 The Five Questions

(1) The Role of Universal Grammar

i. UG must capture the basic properties of language, the domainand species-specific part of the phenomenon language. ii. UG must have evolved in a relatively short evolutionary window. Hence, it must be designed as evolvable. iii. UG must explain the fast language acquisition. In fact, any child acquires any language of the environment in similar stages and has mastered the specific language in only a couple of years.

Complex things cannot emerge rapidly. Consequently, it is a logical implication of i–iii that the language-specific human part, UG, may not be complex but must be simple. Minimalism thus is also a consequence of the scientific method applied to linguistics seeking to clarify in the simplest way what is minimally needed to explain the human capacity for language; what is part of UG, the language-specific capacity all humans share, and what can be factored out as following from language-independent principles. This view on language also implies that we are clear

The Minimalist Framework 21 about what does not belong to language proper. We can externalize language by means of speech and we can use language for communication, but both are outside the narrow domain of the faculty of language (see Hauser, Chomsky & Fitch 2010). This means that we need to keep in mind the very basic distinctions between the UG component of language on the one hand and the related non-UG part on the other hand (see Figure 3.2)), and how those parts interact with the three distinct units of body, mind, and society, which no one would ever claim to be the same. Take a look at the illustrating Table 3.1 too. UG is simple and uniform. The capacity for language is shared among, and unique to, humans. Note that we also have to distinguish between UG, the human capacity for language, and particular languages individuals grow in the course of language acquisition. Particular languages such as English, German, and French are just specific instantiations of UG. This necessarily yields a tension between uniformity on the one hand and diversity on the other hand (Figure 3.3). Although the latter can be directly observed, this is not possible with the underlying, invisible, universal part that ultimately serves as a principled explanation for the visible variation we observe among natural languages. UG is innate and need not be learned, whereas a child exposed to data has to learn the specific properties of the language of his/her environment. Let us therefore assume with Chomsky (2017: 200) that the human language capacity can be described as in (2)

UG

Non-UG

Language is a mind-internal organ- a system interacting with other systems.

Language is not speech, but may be externalized as spoken language (or sign-language)

Language makes complex thought possible.

Language is not communication, but may be used for it.

Figure 3.2 UG and Non-UG

Table 3.1 Body ≠ Mind ≠ Society Body

Mind

Society

speech articulation and perception

language brain activities

communication external behavior of groups

22  The Minimalist Framework

Universal Grammar simple uniform invisible

Specific Languages complex diverse visible

Figure 3.3 The Basic Tension

(2) The Basic Traits of the Human Language Capacity

i. It is uniquely human. ii. It is stable among the species. iii. It is a creative system of its own: a generative procedure yields infinite hierarchical expressions obtaining semantic interpretation (a language of thought) which can be externalized as spoken language.

The very idea of simplicity leads us to SMT. According to Chomsky, language is only optimal concerning SEM, whereas externalization is secondary (Chomsky 2014: 7). UG should be as simple as possible and optimal in the sense that it can interact with the interfaces in the best way. Independent principles of computational efficiency may constrain the generative procedure and the interaction as well, so that it is possible to reduce UG to a minimum. The principle of uniformity captures this idea. Specific languages differ at externalization, that is, on the surface, but share the basic principles of structure that are necessary for the thought side. SEM incorporates a dual semantics, argument structure on the one hand and discourse-related semantics on the other hand. This duality can be captured nicely by External and Internal Merge. PHON deals with the particular properties of specific languages, comparing, for instance, cross-linguistic variation concerning word order and agreement. Roughly speaking, language is uniform in terms of thought, but diverse in terms of sound. Chomsky (2005) summarizes what is needed in a minimal and optimal grammar by developing a three-factor system:

The Minimalist Framework 23 (3) The Three Factors

i. UG, genetic endowment, the domain-specific part accounting for the basic property of human language ii. External data related to the interfaces iii. General properties and laws, the factor related to nature in general

Considerations of computational efficiency also fall under the third factor. We will come back to this point later in our discussion. If UG contains the means to capture the basic property of human language, one needs to tell what the basic property is. Consider the following passage taken from Chomsky (2016). (4) Why does language invariably use the complex computational property of minimal structural distance in the relevant cases while always disregarding the far simpler option of minimal linear distance? [. . .] A broader thesis is that linear order is never available for computation in the core parts of language involving syntax-semantics. Linear order then, is a peripheral part of language, a reflex of properties of the sensorimotor system, which requires it: we cannot speak in parallel. (Chomsky 2016: 12) Berwick, Bolhuis, Chomsky, Huybregts and Everaert (2015) provide massive evidence for structures not strings to be relevant in human language at different linguistic levels, showing that “the syntactic operations that map linguistic objects to the semantic interface do not use the simple properties of sequential string order, that is linear precedence. Instead, they rely exclusively on [. . .] hierarchical structural relations” (Berwick, Bolhuis, Chomsky, Huybregts & Everaert 2015: 732). We can, in fact, observe that it is hierarchical structure that matters and not linear order. This is easily exemplified in the data below. (5) a. [The people who are working with Mary] *is/are nice. b. [The husband of John’s sister] shaved *herself/himself. c. [Who do [you expect the man in the garden talking [to (*Mary)]]]? d. When did you say [that John will arrive]? e. The man entered the room [extremely tired]. Though Mary in (5) a. is linearly adjacent to the copula, the agreement features are determined by the people. In (5) b. the reflexive cannot be bound by the linearly closer noun sister but looks for the far more distant

24  The Minimalist Framework noun husband. In (5) c. a long-distance dependency between who and the preposition to in the most embedded position can easily be formed, and this applies to the adjunct when and the embedded clause in (5) d. as well. In (5) e. the adjectival adjunct extremely tired can be construed by speakers of English with the non-local subject the man without any difficulties. In all cases the relations that need to be established hold over a structural distance, and despite the fact that other elements are linearly closer, the decision is always based on the hierarchical structure. This strong reliance on hierarchy instead of linear adjacency applies to any language and can be regarded as the basic property of language.1 It can hardly be derived from communicational or speech-related aspects but has to be taken as an immediate consequence of UG. The syntactic operation Merge, a recursive operation that forms larger objects or sets from smaller ones, captures this basic property in an optimal way. It makes sense then to understand language as an optimal interaction between the computational system (Merge) and the interfaces (to SEM and PHON) as suggested by Chomsky (2007) in the simple equation recursion + ­interfaces = language. The simplest possible operation, Merge, should thus be defined as follows (see Chomsky 2015a: 97). (6) Merge (X, Y) → {X, Y} =

i. A computational operation creating new elements from elements already constructed ii. As simple as possible: lacking order, being binary and unbounded in principle iii. Output of syntax and input to conceptual–intentional interface (SEM) and sensory–motor interface (PHON)

To explain the basic property of human language, one only needs to resort to Merge in terms of a simple, binary set formation procedure. If this is the ideal case, however, there should also be a place for adjuncts in this simple system. From the perspective of SMT, what is minimally needed then is an inventory of linguistic items that an efficient structurebuilding procedure can operate on to produce larger hierarchical objects that are legible at the interfaces. In short, the overall architecture of grammar that emerges from this picture is summarized in Figure 3.4. Now, where and how do adjuncts fit into this picture? Adjuncts are clearly composed of atoms—items from the lexicon—and they form larger syntactic objects. Consider the complex example in (7). (7) He pressed the button [DP only halfway] [PP with his right index finger] [Inf-clause to remove the memory card].

The Minimalist Framework 25

An inventory of atoms • Lexicon A computational procedure • MERGE Interpretation • SEM & PHON Interfaces Figure 3.4 The Minimal Architecture of the Faculty of Language

Each adjunct, the DP, the PP, and the infinitival clause indicated by the brackets, is a complex syntactic object. Consequently, there must be a computational procedure building adjuncts. Still, as we observed in the previous chapter, adjuncts differ from arguments and they show characteristic properties that set them apart from other syntactic objects. How can these properties be captured? Do they follow from operational or featural distinctions, or can they be derived from interpretational factors? What is the most minimal solution? In order to investigate those questions, we have to take a closer look at the way structures are built in minimalist theorizing.

3.2 Merge and Structure-Building One of the basic questions is how to integrate adjuncts into a structure. Hence, it is necessary to think about the principles of structure-building first. If an object X is to be integrated into a larger one, Z, it is obvious that X can contain objects too. To get started, take the non-linguistic example of creating a painting and think about integration as it applies in this case. The workspace is the canvas and we clearly deal with the arrangement of objects in this workspace. The whole painting is composed of parts. You can also find paintings containing other paintings, and these are arranged in the very same way. So, we have objects with particular features, such as color, size, and shape, which are assembled in a given workspace. Larger objects contain smaller ones, and though we put the objects together to construe a single painting, each object in the

26  The Minimalist Framework picture retains its specific properties. Even though in language we deal with linguistic expressions, we actually need a similar procedure deriving larger objects from smaller objects (with linguistic features). Yet what makes the procedure applied to language unique is that it yields hierarchical objects by means of a recursive procedure. This is in clear contrast to the painting example and supports the uniqueness of structure-­building in human language. It is useful to regard how the structure-building devices in generative grammar developed up to the current state with a “minimal device called Simplest Merge meeting the interface conditions in accordance with relevant third factor conditions” (Epstein, Kitahara & Seely 2017a: 64). So let us have a brief look at the history of structure-building (see also Fukui & Narita 2014; Epstein, Kitahara & Seely 2015, 2017b) before we consider how this affects the issue of adjuncts. In the pre-minimalist era, in the framework of Government & Binding (GB), the module of X-bar theory, which is part of a rather complex conception of UG, is responsible for the building of structures. (8) Projection of lexical features of X to levels X’, XP = labeling by projection which follows from the Principles of X-bar theory:

i. Every phrase has a unique head (endocentricity). ii. Heads may have complements and specifier. iii. Branching is binary.

Following May (1985), Chomsky (1986) assumes that there can also be a segmented category XP-XP in cases of adjunction. This yields the general format in (9) used in GB, which is applicable to intransitive and transitive projections equally, as the simplified example with an adjunct in (10) shows. (9) X-bar theory with XP segments: [XP adjunct [XP specifier [X’ X0 complement]] adjunct] (10) Adjuncts in X-Bar theory: The man . . . [VP [VP worked/read the book] [PP in the garden]] The multi-segmental category cannot be used in a minimalist setting but let us stress two points dating back to the GB era. First, the distinction made between ‘adjuncts’ and ‘adjunction’ is worth mentioning. Adjunctions create new nodes by the rule Move-Alpha, while adjuncts are part of (base-generated) X-bar structures. In a derivational approach such as minimalism, structure-building is always “creative,” and movement is structure-building too. Second, the main problem with the XP-XP addition to the X-bar format is that GB theory builds on lexical heads and the

The Minimalist Framework 27 projection of their respective features, which include argument-­structures but not adjuncts (non-arguments). Hence, adjuncts do not really participate in endocentric structuring because they are not required by a head. On the contrary, they are optional and can even be iterated. In order to be implemented at all into a format that is based on projection and licensing, the segment “trick” is used. The category does not project but is expanded from XP to another segment XP. GB-licensing mechanisms such as theta, case, government, and the projection principle fail to apply to adjuncts in general. Consequently, a non-projectional solution is needed in this case. XP-segments exempt adjuncts from the projection in a projectional theory. Observations about domains and typology lead Rizzi (1990) to try to reconcile adjuncts with projection and licensing by assuming that some higher functional head selects an adjunct. Considering specific instantiations of functional heads and adverbs (compare may–possibly, will–probably, perfective have–recently), one might argue for a connection like this, but the relation is rather semantic in nature and cannot be used in a free minimalist syntax either. In terms of GB, Rizzi (1990: 50) provides the more articulated structure (11) with possible sites of attachment for different types of adjuncts. (11) [AgrP NP [Agr’ [TP [TP [T’ T [VP [VP [V’ [V’ V selected phrases] non-selected small clauses] manner Adv]] reason Adv]]] Different functional layers of the clausal spine serve as adjunction sites here. Note that there are non-branching nodes and vacuous intermediate projections. In general, VP-ellipsis, VP-preposing, and pseudo-cleft sentences as discussed in the previous chapter may provide an argument for the segmented VP-VP structure because by definition the target can be either segment of the non-projecting category. Speas (1990, 1991)2 defines projection as an uninterrupted sequence of projections of X. A maximal projection is thus immediately dominated by some other category and a minimal projection is immediately dominated by a word. It is important for the understanding of adjuncts now that they can be defined neither as sisters to X’, because X’-levels are undefined under this view, nor in terms of Chomsky-adjunction with XP-XP segments, since there is only one maximal node, namely, the topmost. From this perspective, adjuncts cannot be part of the projection and require some additional mechanism adding them after D-structure has been assembled. The upshot is that adjuncts are not a part of the projectional system but need further assumptions in order to be captured in the grammar. What should be kept in mind is the following observation: the problems adjuncts raise for X-bar theory are rooted in the idea of projection in

28  The Minimalist Framework combination with the projection principle. Since projection does not help to syntactically integrate adjuncts, one should be open to the possibility of a better solution to adjuncts in a projection-free syntax, which we will be concerned with in this book. However, the relative status of a projection plus the idea that there can be no vacuous projection is relevant and also taken up later in the minimalist version of bare phrase structure. Structure-building in terms of X-bar theory is one component of GB; a further one is given by the general rule Move-Alpha, which captures the property of displacement in the framework of GB. Consider the overview given in Figure 3.5 illustrating the levels of representation and the rules or principles having a filtering effect on the output. Specifiers represent targets of phrasal movements, and heads of functional categories enforce spec-head-agreement. A wh-phrase agrees with a C node carrying a question feature and a nominal expression agrees with the head of T (or Infl or Agr). Movement is structure-preserving in terms of X-bar positions, which means that phrases can only substitute specifiers, and heads only raise to other heads in the X-bar theoretic sense. In this way Move-Alpha, understood as Move anything anywhere, is structurally constrained. Filters such as the case-filter or the theta-­ criterion operate on the output levels of representation to avoid further over-­generation. As one can observe in Figure 3.5, the architecture of GB is pretty complex, which makes it impossible to explain the rapid and sudden emergence of language in the human species (Darwin’s question). As for Plato’s question, GB models the task of language acquisition as a process of parameter setting. The principles of UG are enriched by options that account for cross-linguistic variation and the growth of

Lexicon items listed with properties Deep-Structure

X-bar, EPP, control, theta

Move-Alpha Surface-Structure

case filter, theta, binding, subjacency, ECP

Spell-out point (overt syntax) LF logical form: Move-α, subjacency, PF phonetic form binding, ECP (non-overt syntax) Figure 3.5 An Overview of Government & Binding (GB) Theory

The Minimalist Framework 29 language by an individual. Parametric choices within UG, of course, further complicate the conception of the human language capacity and are incompatible with a minimalist conception of UG. Minimalism raises basic questions about simplicity such as what is conceptually required, what can be minimized in line with the scientific demand for simplicity. Language relates sound and meaning (thus, neither GB’s D- nor ­S-structure is needed). The principle of Inclusiveness restricts the syntactically accessible elements to lexical items and their features; hence, no new features or objects can be added in the course of the derivation beyond those already present on the lexical items (see Chomsky 1995a: 228). If syntax can only operate on lexical items, arrange and assemble them, and establish relations between them, there can neither be intermediate levels, bar-levels as in GB, nor non-branching nodes, which are only motivated in terms of theory-­internal considerations. The basic operation Merge (12) is a binary operation that recursively applies yielding hierarchical arrangements of linguistic expressions. Theory-internal objects such as GB’s bar-levels or traces of movement violate Inclusiveness. (12) Merge (α, β)

i. ii. iii. iv.

Merge is a binary operation. Merge is set formation. Merge is unordered. If there is a label, it is either (a copy of) α or β.

Lexical items enter the computation, and the derivation is bottom-up with the most deeply embedded unit created first. Ignoring the internal derivation of the phrases a book and in the garden, we would get the derivations in (13) by means of Merge (including labels) for expressions containing adjuncts. (13) a. b. c. d.

{read {read {read, a book}}, in the garden}} {V {V {V, D}}, P}} {work {work, in the garden}} {V {V, P}}

Neither the bare phrase structure (BPS) versions in (13) a. and c. nor the minimal trees using categories in (13) b. and d. can distinguish between the internal argument, the DP, and the adjunct, the PP, because both can be directly merged to the lexical verb; that is, an adjunct occurring in a intransitive projection looks like an argument in a transitive projection. In view of the dichotomy we observed in the previous chapter, this obviously does not capture the empirical behavior of adjuncts properly. Returning to the complex GB framework is of course not an option.

30  The Minimalist Framework If we keep to the assumption that minimalist structure-building reduces to Merge, an innate property of UG, the initial state, we get (14). (14) {γ {α, β}}, where γ = α or β and the order of α and β is irrelevant The Extension Condition on Merge guarantees a bottom-up cyclic derivation. Merge applies at the root (= the highest) node only. In early minimalism, a second syntactic operation, Move, is considered responsible for displacement. Crucially, movement is conceived as an operation driven by features; that is, movement must happen for some reason called feature-checking. This immediately explains the examples under (15). (15) a. b. c. d.

*It seems Mary to win. Mary seems to win the race. It seems that Mary will win the race. *Mary seems that will win the race.

The case-features of Mary remain unchecked in (15) a. as opposed to those in (15) b. In (15) c., Mary is predicted to be frozen in place because the case-feature has been checked in the embedded sentence. There is no reason for moving the DP any further, as it illicitly happens in (15) d. Another fact that is noteworthy in the context of projection is the relative status of the projection line and its levels, which has also been brought up by Speas, as mentioned earlier, but the minimal motivation for this shift is a conceptual one. (16) The Relational Definition of Projection A category that does not project any further is a maximal projection Xmax (XP), one that is not a projection at all is a minimal projection Xmin (X0), and any other is an X’ invisible for computation (compare with Chomsky 1995b: 396). Consequently, maximal and minimal only make sense relationally. There is no upper limit as in GB with XP-XP segments, but any further projection of a second X would yield another possible specifier (using the old term). There is no way to define segments; consequently, it is impossible to refer to adjuncts in terms of segments. Adjuncts cannot be defined in reference to projection. There are heads and non-heads and no multisegmented syntactic objects. A huge advancement in minimalism consists in the ultimate unification of structure-building and movement. In the first era of minimalist reasoning, there is Merge, on the one hand, and a second syntactic operation

The Minimalist Framework 31 driven by feature checking, Move, on the other hand. In a second step, Move is basically reduced to Merge, namely to Internal Merge. External and Internal Merge both apply freely; Internal Merge yields copies since the copy-theory of movement has finally been argued to follow from the No Tampering Condition (NTC) (17) and Inclusiveness as third factor considerations on efficient computation. (17) The NTC Merge does not change syntactic objects already constructed. Merge builds structures, but it does not destroy structures. (18) a. The book John will read . b. The book was read . Although External Merge takes two separate objects, Internal Merge takes a syntactic object that is part of a syntactic object already constructed (hence internal to it) as its input. In (18) a. and b. the syntactic object, the lower copy of the book, occurs within a larger object already constructed and the higher copy is merged to a higher position. No additional operation of movement is necessary to derive the second relation; that is, there is only one syntactic operation Merge. Consequently, the theories of the phrase structure component and the transformational component are unified for the first time in the history of generative grammar. This simplification is desirable in minimalist theorizing, and in the context of the SMT in particular, since it requires “that the generative process is optimal: the principles of language are determined by efficient computation and language keeps to the simplest recursive operation designed to satisfy interface conditions in accord with independent principles of computational efficiency” (Berwick & Chomsky 2016: 71). The SMT also implies that linearization is part of the post-syntactic process of externalization. Linearization of unordered sets (for instance, V precedes the object) takes place at the “sound”-side (PHON). Linear order plays no role at the “meaning-side.” Semantic properties, predicateargument, theta roles, binding, scope, and discourse-related properties are determined hierarchically by c-command in the course of syntaxSEM mapping. Thus, semantically and syntactically significant structures are hierarchically organized, which corresponds to the basic property of human language. None of these relations involve linear order, which is manifested at the PHON interface only. Bare output conditions of SEM require that structures must be interpretable and legible at SEM. Compare the semantic ban on vacuous quantification, ruling out (19) a.

32  The Minimalist Framework (19) a. *Who does John like Mary? b. Who does John like ? The universally restricted number of arguments provides a further example. Four arguments related to one predicate, or two external arguments, say two agents, associated with a single predicate are never attested in any language and seem to be conceptually impossible (unless mediated by coordinative conjunctions). If Move is just Merge and Merge applies freely, then “Move” in terms of Internal Merge also applies freely, without featural triggers. This view gives us Simplest Merge only. (20)  Simplest Merge = Unordered, binary set formation applying freely without labeling To fully understand the way to Simplest Merge, let us study the development of the labeling algorithms from projection to projection-free syntax. Note first that the “wrong choice” of a label necessarily yields a semantically weird syntactic object, for example, the set {read {the book}} labeled as a nominal object cannot be assigned a proper interpretation. Let us briefly consider three different labeling algorithms (LA) Chomsky proposed over the years. (21) The first LA (Chomsky 1995a) The output of Merge (α, β) is labeled by α, if a. α selects β as its semantic argument, or b. α agrees with β: that is, β is attracted by α for the purposes of spec-head-agreement (feature-checking) Recourse to such external relations as (semantic) selection and (formal) agreement create a complication of the LA. To simplify the identification of the label, one can resort to the syntactic objects (SOs) and the operation applied. This causes a reformulation of the LA to (22). (22) The second LA (Chomsky 2008) The output of Merge (α, β) is labeled by α, if a. α is an LI, or b. β is internally merged to α

The Minimalist Framework 33 The LA in (22) raises immediate questions too. Why should Internal Merge in b. require a labeling mechanism that differs from External Merge in a.? What happens if a head, a LI, moves and how is External Merge of two non-heads captured? Chomsky (2013) suggests eliminating (22) b. from the LA reducing it to Minimal Search, a third factor principle, so that Merge can be minimized in consonance with SMT to simple, elementary set formation. The label of a syntactic object is determined independently by the LA, conceived of as a search mechanism for head detection. (23) The third LA (Chomsky 2013) The label/head of an SO Σ is the most prominent LI within Σ. In {H, XP} Minimal Search finds H first, which provides the label. In {XP YP} Minimal Search is ambiguous. It finds X and Y. There are two strategies to resolve the conflict:

i. Modify XP, YP  YP . . . {XP, } ⇒ label = XP ii. Identify XP, YP  shared prominent feature ⇒ label = feature

Crucially, strategy i. yields movement/Internal Merge and strategy ii. yields agreement/Agree. Chomsky (2013) and Epstein, Kitahara & Seely (2014) show that the third LA nicely derives displacement and successive cyclic movement. (24) a. Who C did you say that she likes b. XP-Q YPQ YPdecl c. He seems to be very nice. d. XP-φ YP-φ YP YP In (24) the copy who cannot remain in the intermediate position without causing a labeling conflict (cf. (24) b. XP[Q], YP[decl] do not share the same features). After Internal Merge of a higher copy to the matrix C, the intermediate copy is no longer visible to the LA. Hence, YP can be labeled YP. In the highest position, XP and YP share the feature [Q], which can label the set properly. The same reasoning applies to A-movement in (24) c.–d. Only in the highest position, the DP he agrees with the finite TP, resulting in labeling by the prominent feature φ, whereas the copy in the embedded position is invisible to Minimal Search so that the set can be labeled by non-finite T. UG is thus further simplified by containing exactly the part that captures the basic property of human language. Consequently, (25) and (26) represent a minimal background against which one has to test how adjuncts fit in.

34  The Minimalist Framework (25) UG = Simplest Merge (no specification of a label by Merge) LA = third-factor-driven determination of the label by Minimal Search independent of Merge (26) Syntax = Merge = set formation {α, β} (linear order irrelevant = post-syntactic) Let us highlight the main idea of the current development. We get the simplest possible computational mechanism that applies freely and without limits in a recursive fashion. Merge is binary and unordered (with linear order being part of PHON only), but it generates an infinite array of hierarchically ordered expressions that serve as input to the interfaces. Yet in recent minimalism we find two general views (Table 3.2), namely, the one presented above and a second one that still assumes labeling and projection to be the driving force in syntax. Merge should actually be blind to content; thus, it should not have any access to content either, since syntax is not an interpretive component. Merge builds structures to be interpreted at SEM and PHON, respectively. This again would favor a label-less, un-triggered syntax, as articulated on the right side of Table 3.2. If Merge would need a trigger—say, Agree—we actually arrive at an evolutionary scenario that would foremost require Agree and Merge as dependent on it. Given that Merge captures the basic property of hierarchy that we observe in human language, the role of a trigger of Merge should raise doubts. Nevertheless, we have to examine what happens under both views when it comes to analyzing adjuncts. A further minimalist concept refers to the Transfer to the Interfaces, namely in terms of multiple Spell-Out (MSO) and phases. MSO means that, at the end of a cycle (phase), the syntactic structure is sent off to the interfaces in chunks with no levels of representation at all. Evidently, this reduces the computational load because Transfer makes those chunks of structure inaccessible to further syntactic operation. Minimizing the resources of the computation is basically a third-factor principle not part of UG. One standard way to characterize phases is to define them as designated heads marked by uninterpretable features (C, v as phase heads). The Phase Impenetrability Condition (PIC) enforces Table 3.2 Opposing Views in Recent Minimalism Syntax: Label Creation

Interpretation: Label Identification

Labels drive the computation

Labels provide instructions for interpretation Syntax is free, not crash-proof

Syntax is triggered, crash-proof

The Minimalist Framework 35 successive cyclic “movement” to the edge of a phase. As we discussed before, labeling can also derive “movement,” but the respective “intermediate landing site” would follow from the PIC. We mentioned the third factor quite a few times. Therefore, it is useful to address this issue explicitly and discuss what exactly might follow from third-factor considerations. Note first that simplicity is a condition of systems in general. Hence, a free version of Merge is actually also in line with the third factor. (27) Free = Simplest Merge  No triggering (not driven by features, no labels)  No ordering  Binary operation It is worth bearing in mind how Chomsky defines the third factor, which, strikingly, is not part of UG, hence not specific to language: (28) The third factor falls into several subtypes: (a) principles of data analysis that might be used in language acquisition and other domains; (b) principles of structural architecture and developmental constraints that enter into canalization, organic form, and action over a wide range, including principles of efficient computation, which would be expected to be of particular significance for computational systems such as language. It is the second of these subcategories that should be of particular significance in determining the nature of attainable languages. (Chomsky 2005: 6) What are the possible constraints in terms of third-factor considerations such as computational efficiency? Let us take a look at the possible candidates contributing to efficiency and their respective impacts on the linguistic system. We have already mentioned most of the following principles in the course of the previous discussion, so it should suffice here to simply list them accordingly and point to the effects they have on the computation (see the arrows in (29)). (29) Binarity {x, y} not {x, y, z}  Constrains Merge  Yields c-command Every Merge operation establishes exactly one new relation in the computation if it is binary. Note that the system would add a serious

36  The Minimalist Framework complication if Merge were to apply to three syntactic objects at once, because the operation would have to create three relations—that is to say, between X and Y, Y and Z, and X and Z—all contained in one set, and structurally indistinguishable. (30) Inclusiveness Condition/NTC: the operation leaves the syntactic objects intact.  No elements can be added inside SOs.  No already-constructed relations can be destroyed.  Strict cyclicity (Extension Condition).  Copies in case of Internal Merge. In general, cyclicity governs structure-building from the bottom to the top. Merge leaves the two objects that it operates on unchanged. It is said to apply at the root node, which means that it adds to the entire set that has already been constructed, but it must not add anything to the inside of a set. In fact, the NTC itself, as formulated by Chomsky in On Phases (2008: 138), does not suffice to derive strict cyclicity and the stability of relations already constructed. What basically is needed is a ­no-tampering with relations, such as c-command and sisterhood enforcing Merge to the edge of SOs. This observation leads Epstein, Kitahara and Seely (2012) to the Law of Conservation of Relations (LCR),3 a generalization capturing the Extension Condition and subsuming the NTC. In general, a ­structure-preserving principle that enforces bottom-up cyclicity and prohibits intersecting sets (Parallel Merge) is computationally efficient. Note that such a principle applying in the computation does not hold for PHON in which elements are added and deleted too. (31) Minimize Computation: Phases/PIC  Locality effects: successive-cyclic “movement” It is worth pointing out that this third-factor principle has an impact on the overall architecture of the faculty of language, because it enforces multiple, cyclic Transfer of the syntactic products to the interfaces for interpretation. (32) Architecture of the Faculty of Language Syntactic computation by means of Merge  Transfer making SOs inaccessible to further Merge operations  Interpretation (33) Minimal Search  Head or prominent feature-detection in the case of labeling  Next or closest matching candidate: binding, control, agreementeffects

The Minimalist Framework 37 For the relevant intervention phenomena, consider the data in (34). (34) a. Johnk said that Billm likes himselfm/*k. b. Johnk persuaded Marym PROm/*k to fix the car. c. The man was/*were reading books. The examples in (34) illustrate binding (in a.), control (in b.), and agreement (in c.). No matter how the examples in (34) a.–c. are analyzed in detail, what is relevant here is that they all evoke local interpretation. The relations that must be established have been treated differently. Control, for instance, has been analyzed by means of movement (Internal Merge) by some scholars (Boeckx, Hornstein & Nunes 2010). Agreement may reduce to labeling as shown above, with specific valuations becoming visible at PHON only. Binding concerns the semantic interpretation of particular nominal expressions at the SEM interface. What those relations have in common is the local character that actually might fall out from Minimal Search. (35) Full Interpretation  The computational output must not contain superfluous symbols. Any symbol occurring at the respective interface must be interpretable.  The effects of the Theta-criterion, OP-Var-interpretation (at SEM), copy-pronouncing and case-related effects (at PHON) may follow. The first effect of Full Interpretation abandons, for instance, indices. The second can be illustrated with the sentences in (36). (36) a. b. c. d.

*John met Mary a book. *Who did you meet Mary? *Who do you like who? *He likes he.

Hence, there are principles of efficiency acting on systems in general, and they also have an impact on the language system. Some apply to the interfaces such as Full Interpretation or to the mapping to the interfaces being output related in nature. Note that labeling is viewed in terms of object identification, which is necessary for the proper interpretation at the interfaces. Consequently, Minimal Search can be regarded as a further third-factor constraint having an effect on the linguistic interfaces. It is minimal in conception and reduces options by means of search,

38  The Minimalist Framework which contributes to computational efficiency. Syntax lacks asymmetry and symmetry breaks down at the interfaces (order being lack of symmetry). Asymmetry is interpreted at the interfaces. With binarity, the NTC, phases and the PIC, Full Interpretation, and Minimal Search as third factor effects on the syntactic computation, UG can be kept minimal and Merge can be freed from any interpretive mechanism. The simplest conclusion is that both forms of Merge apply freely, with the output examined (at phase level) for Transfer and ultimate interpretation (see Chomsky 2015b: 14). With Simplest Merge as a given, there are only symmetric sets. Hence, the question remains of how to deal with adjuncts in syntax. We observed a basic asymmetry: the adjunct plus its host always behaves like the host does. How are adjuncts integrated into structures? How do they enter the derivation? How can we account for their distinctive properties? What happens at Transfer? It has often been pointed out that adjuncts behave as islands. Transfer to the interfaces makes chunks of structure inaccessible to syntactic operations, so there might be a connection. What happens at the interfaces? Binding facts involving (anti‑)reconstruction are interpretive in nature, and interpretation starts, so to speak, when ­structure-building comes to an end. To sum up, we can say that a free syntax adopting Simplest Merge is clearly a progress on the way to a simplified UG. The point at issue is that an approach to adjuncts that is formulated in terms of Simplest Merge and SMT is to be favored over accounts postulating additional devices to handle adjuncts. To implement adjuncts in a minimalist setting is, however, no easy task, and we will see in the following sections why adjuncts pose such a challenge.

3.3 Why Are Adjuncts a Challenge? In a Merge-based syntax, X-bar theoretical terms such as specifier and adjunct cannot play a crucial role in minimalist reasoning because they refer to intermediate bar-levels and segments of XP. Consider (37). (37) a. [X W [X Z [X X Y]]] b. [X W [X Z X]] c. [W W [X Z X]] Given inclusiveness, there are no intermediate levels, X’, as in X-bar theory. So what counts as a specifier, what is an adjunct, and what is a complement? Take W and Z in (37) a., for instance. Are they multiple specifiers or adjuncts? Why is Y in a. a complement? Could it be an adjunct to an intransitive verb along with Z and W as further adjuncts? Also, W and Z in b. might be two adjuncts, or Z may be a specifier and W may be an adjunct, or the other way around. In the same vein, it is not clear whether Z in c. is

The Minimalist Framework 39 an adjunct in an intransitive projection, or a specifier, or a complement in a transitive projection. If there is projection (labeling), one can distinguish between first Merge to a label and second or further Merge operations to the same label. If there is no projection (no labeling), the only distinction available is head versus non-head. Yet all pre-minimalist approaches to adjuncts resort to the GB-terminology of the X-bar format; that is, to the terms specifier, head, complement, and adjunct. Consequently, the first challenge is to approach adjuncts without referring to those terms. The second challenge follows from the diversity of existing accounts. Basically, all options for maximal positions in X-bar theory have been suggested for adjuncts. We examine each in turn to illustrate this point. The first account presented here is Larson’s (1988) analysis of adjuncts as complements in the VP-shell. Larson focuses on the principle of binary branching and offers a way to implement double objects on the one hand and adjuncts as a byproduct of the X-bar trees resulting from c-command relations on the other hand. In a sentence like (38), the adjunct in the afternoon is posited inside of a VP with the structure (39). (38) John met Mary in the afternoon. (39) [VP met [V’ Mary [V’ [PP in the afternoon]]]] Notice that ‘Mary’ c-commands the PP. Consider next what binding theory in (40) and the licensing of polarity items in the examples in (41) say about adjuncts. Also see Larson (2004) for a more detailed discussion about the c-command issue. (40) a. John met her on Mary’s birthday. b. John met Mary on her birthday. c. John met the girls on each other’s birthday. (41) John worked rarely before any of the exams. The data, in fact, seem to favor the hierarchy Larson proposes, because her/Mary/the girls c-command the PP adjunct and can bind the respective nominal expressions inside according to principle B and A of the Binding theory applied to (40) b. and c., respectively. If there were co-reference between the pronoun and the R-expression in the adjunct in (40) a., this would violate principle C. In (41), the negative adverb is placed in the specifier position of V according to Larson and can license the Negative Polarity Item (NPI) occurring in the phrase in the complement position of V. Interestingly, then, adjuncts such as rarely would be projectable in specifier positions too. Those relations raise problems under an adjunction structure as in (42), in which the adjunct necessarily c-commands the NP object and not the

40  The Minimalist Framework other way around. NPI licensing and binding data as presented in (40) and (41) would wrongly be predicted as ungrammatical. The literature refers to the contradictory evidence we observe here as Pesetsky’s paradox (Pesetsky 1995). Recall that VP-preposing such as Meet Mary John did on her birthday, in which the adjunct is not fronted along with the VP, provides evidence for the adjunction structure as illustrated in (42): (42) [VP [VP V NP] PP] The binding data based on c-command suggest a right-branching structure like (39) as proposed by Larson, but VP-preposing point to a leftbranching VP shown in (42). Data exhibiting properties of both structures simultaneously pose a challenge to any syntactic theory. Pesetsky (1995) assumes that the single string can have two distinct structures: cascade (39) and layered (42) phrase-markers in his terms.4 If both structures are attested, this would not represent a conclusive argument for Larson’s analysis to begin with. One should object to a dual structure approach in that it leaves open why binding principles exclusively apply to (39) and not also to (42), and the other way round, namely, why VP-preposing is based on (42) and not on (39) too. Another observation is that ­VP-preposing as an instance of Internal Merge must be accounted for in terms of structure-building in narrow syntax, whereas binding might well be related to interpretational rules at SEM. To postulate two distinct structures for a single string would imply that the sentences are structurally ambiguous, but they actually do not have two different readings, as is the case in examples like He helped the girl with the car. Bruening (2014) makes an interesting proposal to resolve Pesetsky’s paradox in terms of phase-command instead of c-command, which allows him to keep a single structure for the data at hand. Crucially, he takes (42) to be correct. Consider his line of reasoning in the quote (43). (43) This particular conflict between constituency tests and command has been extensively discussed in the literature [. . .]. This literature has not taken the obvious step of rejecting c-command; instead, various mechanisms have been proposed to rescue it: dual structures (Pesetsky 1995), temporary constituents that are destroyed due to later processing [. . .], remnant movement [. . .], or ambiguous structures [. . .]. All of these accounts suffer from problems, but there is a more general and fundamental issue: command and constituency are not sensitive to the same nodes. (Bruening 2014: 348–349) We will not go into the details of his analysis. What is relevant to our discussion is that he argues that not c-command but phase-command is

The Minimalist Framework 41 the adequate restriction accounting for the binding facts which, under this view, can be reconciled with the structural hypothesis in (42) and does not favor the VP-shell proposal. The implications of the VP-shell analysis turn out to be additional costs of the analysis. If left-to-right is simply translated into higher-to-lower, movement operations must be stipulated and the positional options multiply by means of shell-projection. The points listed in (44) can be regarded as additional costs in this sense. (44)

i. Head movement of V is required. ii. Multiple adjuncts require more shell-structures and multiply the positional options for adjuncts (specifiers). iii. Arguments and adjuncts both occur either in specifier or complement positions rendering a distinction impossible.

Cinque (1999) proposes a detailed analysis of adverbs as being restricted to occur in specifier positions. His work is part of the so-called cartographic project. He assumes a fairly fine-grained 30-level hierarchy of functional projections supposed to be universal. Each functional head licenses a specific adverb in its specifier. His main argument is based on the observation that the order of adverbs is similar across particular languages. Furthermore, he argues that sign-languages further motivate his analysis because adverbs are expressed manually and non-manually in the same way as aspect, agreement, and negation categories are expressed. Another piece of evidence he sees is in the order children acquire functional heads. Aspectual adverbs are acquired earlier than temporal (and still higher) ones, which would reflect the acquisition of functional heads (Cinque 2004) accordingly.5 The sentences in (45) with adverbs licensed in the specifier of a functional head could hence be represented as in (46). (45) a. They often have talked about it. b. They have often talked about it. (46) [FinP They [Fin’ have [X-functP often [X-funct’ Xfunct [PerfP [VP talked about it]]]]]] What must happen in the sentences in (47)? (47) a. They (probably) will probably again be quickly walking down the road in the afternoon. b. He will (slowly) open the door slowly.

42  The Minimalist Framework If any adverb or prepositional adjunct can only be licensed in the specifier of a functional head, we have to stipulate that there must be a lot of unexpressed, empty functional heads along the clausal spine for the five different adjuncts occurring in (47) a. Also, the analysis requires that, besides head movement, phrasal movement operations take place in a sentence with adjuncts which would not take place without those adjuncts. To derive the final position of slowly and down the road and in the afternoon in (47) a. and b., full VPs have to move around those phrases stipulated to be in specifier positions and still higher phrases too. This is what Ernst (2014) calls roll-up movements. It follows that the costs of this analysis appear rather high. Cinque primarily talks about adverbs. PPs pose a problem under his account. To see why, consider the data in (48). (48) a. *John in his office read the book.6 b. John read the book in his office on Monday. c. John read the book on Monday in his office. The sentence in (48) a. is predicted to be the normal case contrary to the fact. To derive (48) b. the VP read the book needs to be moved around the final prepositional phrases. This does not reflect the quite general observation that, first, PPs tend to occupy a final position, and, second, that the order of prepositional adjuncts is not fixed as (48) a.–c. illustrate. With respect to movement operations required in Cinque’s approach, Bobaljik (1999) presents an interesting argument against fixed adverb positions, which he calls the hierarchy paradox. He shows that the hierarchical order of adverbs Cinque suggests leads to locality violations, violation of the Head Movement Constraint (HMC) in particular.7 In sum, the analysis requires a lot of movement operations: head movement and also roll-up movements. Full VPs, for instance, need to be raised into a higher specifier position for adjuncts to ever occur in a final position. The functional heads that license the adverbs are in general not morphologically expressed but are silent heads. Another issue raised is the status of linear order in Cinque’s approach. We mentioned before that linear order is primarily an aspect of externalization; that is, of the phonological surface, and should by no means be equated with the hierarchical structure of the syntactic component. Yet this seems to be the case with Cinque’s reasoning for adverbs/adjuncts as specifiers. Moreover, the observed relation between an adjunct and a functional domain may also be attributed to the respective semantic interpretation of an adjunct at the semantic interface and is no absolute argument for anchoring an adjunct in the specifier of a fixed functional head. How about adjuncts as adjuncts again? In the framework of Barriers, in the tradition of Government & Binding, Chomsky (1986) analyzes

The Minimalist Framework 43 adjuncts as left or right sisters to XP. This yields the X-bar format, repeated here for convenience in (49). (49) [XP adjunct [XP specifier [X’ X complement]] adjunct] In the Barriers framework, adjuncts are not fully included in the projection. They are not dominated by XP as are specifiers and complement, which means that they are not fully integrated into the projection of their host—an effect of which is their positional flexibility to occur left- or right-adjoined to the phrase they modify. Ernst (2004) argues again for adjunction of adjuncts. He assumes that adjuncts can be freely adjoined to XPs in the syntax and are later mapped to the interfaces where rules apply that are not specific to adjuncts. He stresses that the cartographic approach is counter-evidenced by examples as (50): (50) a. She cleverly solved the problem. b. She solved the problem cleverly. To ensure the two different readings, the cartographic approach needs to posit two different homophonous adverbs cleverly. Moreover, the parallels between clausal adjuncts and adjuncts occurring in nominal expressions suggest that ordering effects are driven by scope and compositional semantics. Ernst suggests that adjuncts take a Fact-Event object (proposition/event) capturing different domains or zones (_VP, _TP, _CP). Consequently, the following (51) a. is ruled out semantically, but syntactically possible like (51) b. Merge, the syntax, does not distinguish between types of adjuncts under this view (Ernst 2004: 35). (51) a. #Karen wisely was probably going home. b. Karen probably was wisely going home. On page 448, Ernst summarizes the different modification types and their respective scope domains: event internal and event external are both associated with the predication, while event external modification would also be linked to the broader functional domain by his account. The reader is referred to Ernst’s overview and the respective discussion of his own examples. He suggests that, since the order of adjuncts is not always rigid, the orders attested in the data are lexically determined (by entries of the adverbs in question) and driven by the semantic interface. Another factor is to be located at the phonological interface. Heaviness and length on the one hand and directionality parameters concerning cross-linguistic variation on the other hand determine the linear order of sequences. This factor of heaviness at PHON can be illustrated by (52)

44  The Minimalist Framework where the extra-large direct object follows the adjunct. The latter fact is obvious if we compare, for instance with the order in German in (53) which is much more flexible than in English. In addition, German exemplifies head finality in the VP/TP as opposed to the head initial behavior of English phrases. (52) Ernst presents [PP-adjunct in this article] [DP-pbject a preliminary overview of some major properties of adjuncts that need to be accounted for in a proper analysis]. (53) a. b. c.

. . . dass der Student das Buch gestern in der Bibliothek gelesen hat . . . that the student the book yesterday in the library read has . . . dass der Student in der Bibliothek gestern das Buch gelesen hat . . . that the student in the library yesterday the book read has . . . dass in der Bibliothek der Student gestern das Buch gelesen hat . . . that in the library the student yesterday the book read has ‘. . . that the student has read the book yesterday in the library’

Note that Ernst’s approach is minimalist in nature. Syntax is free and the interfaces SEM and PHON interpret the Merge products, respectively. Still, one needs to clarify what adjunction means in a Merge-based syntax. With Kayne (1994), Ernst assumes that there is no distinction between specifier and adjunct—there are no segments in the sense of May (1985) or Chomsky (1986), but symmetric phrase-structures with only two basic non-head positions (complement and adjunction). The Spec is defined as a type of adjunction. This yields structures as the following (See Ernst 2004: 109). (54) [XP Adv-1 [XP X [PredP Adv-2 [PredP Adv-3 [PredP Pred VP]]]]] Consequently, adjuncts are XPs merged with non-heads just as specifiers are. In essence, the structure then is identical to the following one, which is in line with the minimalist Inclusiveness Condition. (55) [X Adv-1 [X X [Pred Adv-2 [Pred Adv-3 [Pred Pred [V V . . .]]]]] So why not stop here and take Ernst’s account as the minimalist solution? As (54) and (55) show, there is still a label and lexical entries of the adjuncts in question determine the distribution in Ernst’s approach. In Chapter Two, we focused on the asymmetry between adjuncts and arguments. How is the dichotomy captured under the scopal view on adjuncts? Is it a pure matter of the semantic interface? What is the most minimal approach to adjuncts?

The Minimalist Framework 45 In the forgoing discussion, we concentrated on the various ways adjuncts have been treated in the previous generative tradition. Approaches in the pre-minimalist era use the X-bar format and propose to analyze adjuncts in terms of complements, specifiers, or adjuncts. Ernst (2004) is the first minimalist account. It is striking that the proposals either put adjuncts syntactically in the same position as subject arguments or object arguments. Recall the dichotomy of arguments versus adjuncts that we discussed in the second chapter. Thinking about extraction, we should keep in mind that adjuncts behave as islands, but that subjects (contrary to objects) are islands in English too, as the data in (56) exemplify (also compare with Nunes & Uriagereka 2000; Johnson 2003). (56) a. *Which party was John bored during? b. *Which book did a review about annoy John? c. Which book did John read a review about? Yet we observed that arguments are obligatory, whereas adjuncts are not. We can have an unbounded number of adjuncts, whereas the maximum number of arguments is restricted to three (or less depending on the specific predicate chosen). Two basic questions follow from the approaches we discussed up to this point. Do adjuncts require a syntactic treatment other than arguments, or is the distinction to be captured in terms of semantics? We learned that both routes are taken in the literature. Either the distribution of adjuncts is regulated by the syntactic projection, or semantic principles determine the appearance of certain types of modifiers. A general overview of both directions can also be found in Alexiadou (2004) and Ernst (2014). Let us just summarize some of the pros and cons discussed by Ernst and also addressed here in Table 3.3, indicated by pluses and minuses. Table 3.3 Two Basic Approaches Cartographic Approach

Scopal Approach

UG-derived, rigid word order; licensed by functional heads

free adjunction in syntax; constrained by semantic zones and morphological weight + constituency (VP-ellipsis) - c-command effects (Barss & Lasnik) - zones not worked out in detail + scope and semantics predict ambiguities + PP predicted - semantics allows *She well plays chess

+ c-command effects (Barss & Lasnik) - constituency - semantic domains/zones unpredicted - same adverbs with different readings - flexible PPs unpredicted - complex movement stipulated (roll-up)

46  The Minimalist Framework While Cinque (1999, 2004) claims that syntax structurally determines the distribution of adjuncts by means of functional heads, Ernst (2004) assumes that semantics determines the order of adjuncts. We should add here that Larson (2004) refers to a Neo-Davidsonian semantics that, according to him, can explain a direct mapping of the VP-shell-structure to semantic terms as restriction/presupposition and scope/asserted material. Consider briefly the following examples in (57) and (58), illustrating the suggestion for (57) c. representatively. (57) a. Mary kissed John quickly. (manner) b. Mary kissed John for an hour. (duration) c. Mary kissed John in the park. (location) (58) There is a kissing of John by Mary and it is in the park. ∃e[kissing(e)∈ & Agent(e, m) & Theme(e, j) & in-the-park(e)] presupposition/restriction & asserted material/scope Thereby, Larson supplements the VP-shell-system with a semantic account of adjuncts. To conclude this chapter, we investigated approaches to adjuncts primarily based on positions in terms of the X-bar theory. The discussion revealed that basically, there are two opposing views in the literature, which are schematized in Figure 3.6. Either one assumes that syntax determines the semantics of adjuncts or that the interpretation of adjuncts is done by interpretational rules to be fleshed out at the interfaces SEM and PHON.8 Considering the two most frequently assumed and general Government & Binding options for adjuncts—namely, licensing by a functional head or adjunction to a maximal phrase (yielding segments)—it is also striking that these two have follow-up minimalist approaches. Take a look at the overview in Figure 3.7. We will turn to those developments in detail in the next chapter. Keep in mind the two challenges emerging from the discussion. What we need

SYNTAX

SEMANTICS and LINEAR ORDER of adverbs follow from structural constraints (Cinque 1999, 2004)

SYNTAX

Interpretational rules of SEMANTICS and PHONOLOGY (Ernst 2004)

Figure 3.6 The Relationship Between Syntax and Semantics/Phonology

The Minimalist Framework 47

Government & Binding (Chomsky-) adjunction of YPadjunct to XP: [XP [XP …X…] YPadjunct]

Licensing of adjuncts by functional heads: [XP YPadjunct [X’ Xfunct […]]

Minimalism Different dimensions

Feature projection and labeling

Figure 3.7 GB Precursors of Minimalism

is a minimal analysis that does not rely on X-bar theoretic positions (i) but keeps to the most minimal assumptions while accounting for the characteristic properties of adjuncts as outlined here (ii).

Notes 1. Structural ambiguities in languages also provide strong evidence for hierarchical decisions as opposed to linear ones. Take a look at a German example that can be easily translated into English showing that ambiguities resolve by relying on hierarchical order and that this is observable in any language. The italicized adjunct in i. can either refer to the man or to the event of following. i.  Er verfolgt einen Mann in einem Taxi. (He follows a man in a taxi.) 2. Thanks go to the reviewer who reminded of the important role of Speas (1990) in the discussion of projection. 3. In NS, syntactic relations (among existing terms) cannot be altered throughout the derivation (Epstein, Kitahara & Seely 2012: 256). 4. We will not discuss solutions based on parsing, such as Phillips (2003), who assumes that structures are built incrementally top-down from left to right because they cannot be reconciled with the derivational bottom-up conception of (Simplest) Merge. 5. If aspectual adverbs are interpreted at a higher position, which presupposes that this higher level is already present in the child’s language, this actually is no real argument in favor of a specifier head licensing but also can be explained by referring to interpretive domains. 6. A reviewer objects that (48) a. is predicted to be ungrammatical in a later paper since locative PPs would be located lower in the clause. However, Cinque’s account in the framework of cartography requires a complex feature system that is not in line with a free syntax and SMT.

48  The Minimalist Framework . See Bobaljik (1999) for a detailed discussion. 7 8. All accounts agree on the necessity to derive the interpretation of adjuncts, but they clearly differ in the relation that holds between syntax and semantics.

References Alexiadou, A. 2004. Adverbs Across Frameworks. Lingua 114: 677–682. Berwick, R. C., Bolhuis, J. J., Chomsky, N., Huybregts, M. A. C. and Everaert, M. B. H. 2015. Structures Not Strings: Linguistics as Part of the Cognitive Sciences. Trends in Cognitive Sciences 19 (12): 729–743. Berwick, R. C. and Chomsky, N. 2016. Why Only Us? Language and Evolution. Cambridge, MA: MIT Press. Bobaljik, J. 1999. Adverbs: The Hierarchy Paradox. Glot International 4 (9/10): 27–28. Boeckx, C., Hornstein, N. and Nunes, J. 2010. Control as Movement. Cambridge: Cambridge University Press. Bruening, B. 2014. Precede-and-Command-Revisited. Language 90: 342–388. Chomsky, N. 1986. Barriers. Cambridge, MA: MIT Press. Chomsky, N. 1995a. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, N. 1995b. Bare Phrase Structure. In Webelhuth, G. (ed.), Government and Binding Theory and the Minimalist Program: Principles and Parameters in Syntactic Theory, 385–439. Oxford: Blackwell. Chomsky, N. 2005. Three Factors in Language Design. Linguistic Inquiry 36: 1–22. Chomsky, N. 2007. Approaching UG from Below. In Sauerland, U. and Gärtner, H. M. (eds.), Interfaces + Recursion = Language? 1–29. New York: Mouton de Gruyter. Chomsky, N. 2008. On Phases. In Freidin, R., Otero, C. and Zubizaretta, M. L. (eds.), Foundational Issues in Linguistic Theory: Essays in Honor of JeanRoger Vergnaud, 133–166. Cambridge, MA: MIT Press. Chomsky, N. 2013. Problems of Projection. Lingua 130: 33–49. Chomsky, N., 2014. Minimal Recursion. In Roeper, T. and Speas, M. (eds.), Recursion: Complexity in Cognition, 1–15. Cham, Heidelberg, New York, Dordrecht, and London: Springer. Chomsky, N. 2015a. Some Core Contested Concepts. Journal of Psycholinguistic Research 44: 91–104. Chomsky, N. 2015b. Problems of Projections: Extensions. In di Domenico, E., Haman, C. and Matteini, S. (eds.), Structures, Strategies and Beyond: Studies in Honor of Adriana Belletti, 3–16. Amsterdam and Philadelphia: John Benjamins. Chomsky, N. 2016. What Kind of Creatures Are We? New York: Columbia University Press. Chomsky, N. 2017. The Language Capacity: Architecture and Evolution. Psychonomic Bulletin & Review 24 (1): 200–203. Cinque, G. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. Cinque, G. 2004. Issues in Adverbial Syntax. Lingua 114: 683–710. Epstein, S. D., Kitahara, H. and Seely, T. D. 2012. Structure Building That Can’t Be. In Uribe-Etxebarria, M. and Valmala, V. (eds.), Ways of Structure Building, 253–270. Oxford: Oxford University Press.

The Minimalist Framework 49 Epstein, S. D., Kitahara, H. and Seely, T. D. 2014. Labeling by Minimal Search: Implications for Successive-Cyclic A-Movement and the Conception of the Postulate “Phase”. Linguistic Inquiry 45: 463–481. Epstein, S., Kitahara, H. and Seely, T. D. 2015. From Aspects’ ‘Daughterless Mothers’ (AKA Delta Nodes) to POP’s ‘Motherless Sets’ (AKA Non-­Projection): A Selective History of the Evolution of Simplest Merge. In Gallego, Á. and Ott, D. (eds.), 50 Years Later: Reflections on Chomsky’s Aspects, 99–111. Cambridge, MA: MITWPL. Epstein, S. D., Kitahara, H. and Seely, T. D. 2017a. Is the Faculty of Language a “Perfect Solution” to the Interface Systems? In McGilvray, J. (ed.), The Cambridge Companion to Chomsky (sec. ed.), 50–68. Cambridge: Cambridge University Press. Epstein, S. D., Kitahara, H. and Seely, T. D. 2017b. Merge, Labeling and Their Interactions. In Bauke, L. and Blümel, A. (eds.), Labels and Roots, 17–45. Boston and Berlin: De Gruyter Mouton. Ernst, T. 2004. (sec. ed.). The Syntax of Adjuncts. Cambridge: Cambridge University Press. Ernst, T. 2014. The Syntax of Adverbs. In Carnie, A., Sato, Y. and Siddiqi, D. (eds.), The Routledge Handbook of Syntax, 108–130. London and New York: Routledge. Fukui, N. and Narita, H. 2014. Merge, Labeling, and Projection. In Carnie, A., Sato, Y. and Siddiqi, D. (eds.), The Routledge Handbook of Syntax, 3–23. London and New York: Routledge. Hauser, M., Chomsky, N. and Fitch, T. 2010. The Faculty of Language: What Is It, Who Has It and How Did It Evolve? In Larson, R. K., Deprez, V. and Yamakido, H. (eds.), The Evolution of Language: Biolinguistic Perspectives, 14–41. Cambridge: Cambridge University Press. Johnson, K. 2003. Towards an Etiology of Adjunct Islands. Nordlyd 31: 187–215. Kayne, R. 1994. The Anti-Symmetry of Syntax. Cambridge, MA: MIT Press. Larson, R. K. 1988. On the Double Object Construction. Linguistic Inquiry 19: 335–391. Larson, R. K. 2004. Sentence-Final Adverbs and “Scope”. NELS 34: 23–34. May, R. 1985. Logical Form—Its Structure and Derivation. Cambridge: MIT Press. Nunes, J. and Uriagereka, J. 2000. Cyclicity and Extraction Domains. Syntax 3: 20–43. Pesetsky, D. 1995. Zero Syntax—Experiencers and Cascades. Cambridge: MIT Press. Phillips, C. 2003. Linear order and constituency. Linguistic Inquiry 34: 37–90. Rizzi, L. 1990. Relativized Minimality. Cambridge, MA: MIT Press. Speas, M. 1990. Phrase Structure in Natural Language. Dordrecht: Kluwer. Speas, M. 1991. Generalized Transformations and the D-Structure Position of Adjuncts. In Rothstein, S. (ed.), Perspectives on Phrase Structure: Head and Licensing. Syntax and Semantics, Vol. 25, 242–257. San Diego: Academic Press.

4 Approaches to Adjuncts

4.1 A Different Dimension The first approaches in minimalism build on the arguments—adjuncts dichotomy in particular. The basic assumption here is that adjuncts basically enter the derivation on separate dimensions. Before discussing those approaches in detail, let us think about what dimensions are in the first place. One might speculate about a distinction along temporal lines. Take, for instance, simple clock time. If it is midnight in Germany, it will be 6 a.m. in Michigan, USA—so in some sense we have different times at the same time. Space and time (and space– time, if you think about relativity) are different though clearly related dimensions. In spatial terms, you can easily think of examples illustrating a distinct dimension. You can, for example, take a building with different floors. If one person is located on the first floor and another one is on the second floor, they are still in the same building but are in different dimensions. Those dimensions are real and part of the observable natural world. Therefore, it is not in the least surprising that some scholars try to deal with the dichotomy between adjuncts and arguments by referring to different dimensions, namely, temporal and spatial dimensions. A temporal dimension is basic for so-called Late-Merge approaches, building on Lebeaux (1991), who suggests that adjuncts can be inserted after wh-movement, thus circumventing Condition C of the binding theory, while complements do reconstruct. The relevant contrast can be seen in (1). (1) a. Which picture of Maryk did shej/*k sell? b. Which picture that Maryk took did shej/k sell? Stepanov (2000, 2001) assumes that adjuncts are merged post-cyclically (counter-cyclic or a-cyclic) after all non-adjuncts have been merged. Hence the adjunct is not part of the copy. By means of post-cyclic LastMerge of adjuncts, he explains the bleeding of Condition C in (1) b. The adjunct containing the R-expression is simply not there when the

Approaches to Adjuncts 51 wh-phrase moves to C; therefore, it will not cause a violation of principle C in the base position. The adjunct, the relative clause, is last-merged to the wh-phrase in the highest position. Also, some principle A effects (2), lack of intervention effects (3), the absence of blocking of PF-merger by adjuncts (4), and island effects (5) follow on his account. (2) a. *Which proof that showed that picture of each other are on sale did they deny? b. They denied the proof that showed that pictures of each other are on sale. (3) John seems [to Mary] to be smart. (4) a. John T quickly left. b. *John did quickly leave. (5) a. *What did John go to bed after Peter fixed? b. [C [T [John go to bed]]] [after Peter fixed what] The anaphor is contained in the adjunct; the relative clause that is lastmerged in (2) is not c-commanded by they. This results in a principle A violation. The example in (3) shows that the prepositional adjunct, the experiencer, does not block the raising of John, because the PP adjunct cannot intervene if it is merged after A-movement of the DP applied. The examples in (4) provide evidence for the fact that the adjunct does not disturb the adjacency between T and the verbal expression. Adjacency must hold for the morphological merger of inflectional elements and the verb to take place. Data such as in (5) exemplify the island property of adjuncts. There can be no movement across phrase-markers that are separately assembled, which is predicted to be impossible in the case of adjuncts as last-merged, hence not yet present when the matrix clausal spine is derived. It is crucial for Stepanov’s account to work that the NTC can be violated or needs to be weakened. Therefore, he formulates the Least Tampering Condition1 which allows for the last step of Merge to tamper. According to him, Merge preserves existing structures and Merge applies at the root when possible. Merge of adjuncts takes place at the end of a derivation because any further cyclic Merge to the segmented projection would change the c-command relations already established. Also, Stepanov defines an adjunct by resorting to features and projection (or its negation as in (6)) in order to define which phrases actually occur in a segmented syntactic object.

52  Approaches to Adjuncts (6) An adjunct = a non-projecting object A, merged with B, where A contains no active uninterpretable features to be checked. It follows that wh-adjuncts are different from non-wh-adjuncts. They undergo successive cyclic movement like arguments because they have uninterpretable features in their label, which, under this view, must then be checked against a matching C node cyclically: (7) a. How do you think that John said that Bill fixed the car ? b. Why do you say that they have chosen this paper ? We briefly considered the Last-Merge approach Stepanov proposed. We now turn to some problems this account has to face. One question concerns the architecture of the grammar and multiple Spell-Out. In order to allow for adjacent PF merger of T and vP, he resorts to non-cyclic Merge of a seemingly intervening adjunct, but, normally, Spell-Out makes the structural material inaccessible to further syntactic operations. Postcyclic Merge of the adjunct to spelt-out material is, however, what such an analysis would require. In general, Late-Merge increases the generative power by allowing counter-cyclic derivations in unbounded fashion. The question is whether counter-cyclic derivations are in line with minimalist reasoning, the SMT in particular. Late-Merge always complicates the computational procedure2 and raises the issue of computational efficiency. Furthermore, in Stepanov’s Last-Merge account, Merge needs to see the features of the elements in order to determine whether cyclic application (label with active features) or post-cyclic application (label with no active features) is needed. There are also empirical problems (A /A-bar differ in reconstruction and further embedding (see Biskup 2009; Sportiche 2019a, 2019b; Saito 2017 for a detailed discussion).3 Approaches explaining the behavior in terms of the respective timing of Merge are either based on counter-cyclic or post-cyclic derivation of adjuncts. Approaches based on a temporal distinction complicate the grammar and need to be supplemented with further mechanisms such as the label-inspection Stepanov posits in order to distinguish adjuncts from other phrases.4 An alternative view focuses on local dimensions in order to distinguish adjuncts from other syntactic objects. The standard alternative is to analyze adjuncts in terms of an operation called Pair-Merge. Again, the basic goal is to capture the asymmetry between arguments and adjuncts, which

Approaches to Adjuncts 53 Chomsky stresses in terms of a local dimension since “given the basic properties of adjunction, we might intuitively think of α as attached to β on a separate plane, with β retaining all its properties on the ‘primary plane’, the simple structure” (Chomsky 2004: 117–118). The asymmetric nature of adjuncts, namely, the observation that the whole structure—that is, the host plus the adjunct behaves like the host of adjunction—needs to be accommodated somehow in the grammar. According to Chomsky (2004), “an adjunction construction is plainly not the projection of a head: for NP-adjuncts for example, the constituent structure appears to be something like [NP XP]” (Chomsky 2004: 117). If the host of adjunction retains all its properties, then an NP or a VP with an adjunct still behaves as an NP or a VP respectively. Adjuncts show properties that place them in a parallel dimension. The point is that they are kept on a separate syntactic plane, which would exempt them from c-command relations. Consequently, they behave as if they were invisible, which empirically shows up in the properties in (8) we discussed earlier in Chapter Two. (8) “Invisible,” in a separate local dimension not interfering with the main clausal spine, hence, independent of the matrix clause insofar as

i. They don’t change selection. ii. They are not selected themselves. They neither get theta roles nor are they case-marked like arguments. iii. They may occur left or right. iv. PPs, in particular, are often unordered in syntax (though this is semantically constrained). v. They are immune to reconstruction. vi. They are islands to extraction.5

Note that we now deal with two syntactic operations in the computation. See (9) below and also the passage (10) taken from Chomsky (2004) explaining the basic idea. (9) Two Syntactic Operations

i. Symmetric Set-Merge: {X, Y} = {Y, X} = unordered pair ii. Asymmetric Pair-Merge: ≠  = ordered pair

(10) For structure building, we have so far assumed only the free symmetrical operation Merge, yielding syntactic objects that are sets, all binary: call them simple. The relations that come “free” (contain

54  Approaches to Adjuncts c-command, etc.) are defined on simple structures. But it is an empirical fact that there is also an asymmetric operation of adjunction, which takes two objects β and α and forms the ordered pair , α adjoined to β. Set-Merge and Pair-Merge are descendants of substitution and adjunction in earlier theories. Given the basic properties of adjunction, we might intuitively think of α as attached to β on a separate plane, with β retaining all its properties on the “primary plane,” the simple structure. Two questions arise about adjunction: (1) Why does it exist? (2) How does it work? (Chomsky 2004: 117–118) The why-question can be answered by looking at the semantic impact of adjuncts. Adjuncts enrich the expressive power at C-I, that is adjuncts take part in predicate composition. Hence, Chomsky (2004) suggests to associate syntactic operations with semantic relations as summarized in Table 4.1. The how is determined by the syntactic operation involved or, as Chomsky puts it in 2008, (11) The adjunct island subcase follows if an adjunct is not in the search domain of the probe. That in turn follows from the approach to adjuncts in Chomsky (2004), taking them to be entered in the derivation by Pair-Merge instead of Set-Merge to capture the fundamental asymmetry of adjunction. (Chomsky 2008: 146–147) Note that a positive consequence of this approach is that we can retain strict cyclicity, but we have different types of Merge now, which is of course a complication. Recall that though adjuncts behave as islands, they can be internally merged themselves as illustrated in (12). (12) a. *When haven’t you seen him [since ]? b. Since when haven’t you seen him ? c. Since when did you say that you think that you haven’t seen him ? Table 4.1 Different Syntactic Operation: Different Semantic Relation Syntactic Operation

Semantic Relation

External Set-Merge Internal Set-Merge Pair-Merge

argument structures discourse-related structures predicate composition

Approaches to Adjuncts 55 In (12) b.–c., the adjuncts must obviously be in the same dimension as the rest of the phrases unless we assume that all copies get internally pairmerged into the derivation. In (12) a. the full adjunct would be on a separate plane and therefore internally inaccessible to the other dimension. Later, as an optional part of Transfer to the interfaces, the operation SIMPL(ify) converts the ordered pair into a standard set. The operation incorporates elements from the separate plane into the simple structure for SEM and PHON interpretation to be applicable to the complete material the computation generates. The anti-reconstruction effect in (13) follows if the point of spelling out is determined by the host. (13) [Which picture [that Johnk likes]] did hek sell ? The operation Simplify applies to the copy merged to C, which explains that condition C is finally circumvented. Rubin (2003) addresses the question of how the system can decide between Set-Merge and Pair-Merge, that is when Set-Merge is applied, or when a phrase is pair-merged. He proposes that there is a silent Modhead for adjuncts which is subjected to Pair-Merge. Consider his reasoning in passage (14). (14) Given that adjuncts are endowed with a functional shell parallel to that which occurs in nominals and clauses (DP and IP/CP, respectively) under such an account, the simplest possible determination of Pair-Merge is available without search, just as we would want in the strongest minimalist theory. Any phrase headed by Mod is subject to Pair-Merge, not Set-Merge. (Rubin 2003: 664–665) In English, this stipulated head is always unexpressed. Also, it is questionable whether this functional extension would really be parallel to extended projections we find in the case of DP and clauses. Mod would have to select or extend completely different categories, namely, any category that occurs as an adjunct in the first place, for example AdvPs, PPs, and even clauses of various types. In contrast to this, it is normally assumed that nominal and verbal projections are functionally extended by means of a restricted set of heads. Irurtzun and Gallego (2007) argue further that Pair-Merge gives the right results in case of scopeless Markovian readings of right peripheral adjunct clusters and non-Markovian, scopal adjunct clusters too because, given distinct dimensions, the latter semantics would follow if each adjunct is separately formed by Pair-Merge, hence allowing for each to refer to a sub-event, which explains the possibility of framing effects as opposed

56  Approaches to Adjuncts to the paratactic Markovian reading. The interface-driven behavior of adjuncts thus would provide another argument for a Pair-Merge solution building on the integration into primary (or even secondary) planes. The Pair-Merge account covers many properties of adjuncts, yet it raises questions some of which we will address now. Empirical problems concern A-movement, which shows principle C effects and binding effects with adjuncts (compare with Biskup 2009; Sportiche 2019a, 2019b; Saito 2017). This is not predicted under the assumption that adjuncts are on a separate plane. The status of adjunct islands which are not absolute (see for instance Truswell 2007a, 2007b, 2011) should also be investigated, but this is not an objection applying particularly to the account at hand, but a general observation. Theoretical questions concern the general availability of External and Internal Pair-Merge parallel to Set-Merge. Richards (2009), for instance, suggests that internal Pair-Merge should be possible to handle scrambling data. Epstein, Kitahara and Seely (2016) argue for External Pair-Merge of heads and thereby derive differences among complements. They also predict some interesting pre-syntactic, morphological effects. Therefore, Pair-Merge might well be a theoretical and empirically attested option. However, the stipulation of a Mod-head for adjuncts, which Rubin (2003) suggests, seems ad hoc and probably is not needed at all. If adjuncts do not enter the derivation via Pair-Merge, one might argue that this would cause problems at SEM. Hence, it should rather be a matter of interpretation. Suppose an adjunct is set-merged instead of pair-merged (if we ignore the effects of Simplify). The result simply would not get a proper interpretation. But note that this assumption presupposes the idea that Pair-Merge yields syntactic objects that are interpreted in a specific way and are not translated into simple sets by Simplify at Transfer, undoing the distinction between Set- and Pair-Merge. However, one must discuss whether an additional operation accords with minimalist theorizing. If the asymmetric behavior of adjuncts is in fact interface-driven, one might wonder why they could not be treated in same fashion as other phrases in narrow syntax without being placed in a separate plane and whether one could account for the dichotomy as an interface effect only. Note that we observed that the internal structure of adjuncts is to be derived by Set-Merge anyway, and movement applying to adjuncts, that is internal Merge of complete adjuncts, are instances of Set-Merge too. Pair-Merge has been put forward then to handle (part of) the external distribution of unmoved adjuncts only. An obvious simplification of the computational procedure needed to account for adjuncts in general would be, however, to appeal to just one single operation. Further operations such as Pair-Merge and (sub‑)operations (of Transfer) like Simplify clearly complicate the grammar. Still, the overwhelming evidence for an asymmetry between arguments and adjuncts may call for differences in computation, which would have an effect on the interfaces too.

Approaches to Adjuncts 57 There are other minimalist approaches to adjuncts that are based on considerations of labeling and projection. Recall the discussion on labeling as projection and triggered versus un-triggered syntax in Chapter Three. In the following section, let us turn in detail to those accounts that make use of labeling and projection to cover adjuncts.

4.2 Labeling and Projection The question of labeling of a syntactic object can be addressed from different perspectives. One way to look at labeling is to make a distinction between labeled and unlabeled syntactic objects in the first place. Viewed this way, labeling is part of syntax but is not necessarily obligatory for all objects. One can also ask why labeling is needed at all and where labeling takes place. Recall the debate concerning labeling as being part of syntax or part of interpretation. Labels either trigger syntactic operations, determining the distribution of syntactic objects or, alternatively, serve as instructions to the interpretive interfaces only. Another question concerns the properties that project.6 In a binary process, two different objects may project their respective features. However, the issue might be raised what properties project, why (only) those, and how they project (relative to the derivational cycle). The following approaches to adjuncts address the issue of labeling in particular. Hornstein and Nunes (2008) and Hornstein (2009) question an independent operation Pair-Merge on minimalist grounds. Instead, they suggest that the difference between arguments and adjuncts can be viewed as a difference in terms of labels. Recall that BPS7 makes a distinction between complements, specifiers, and adjuncts pretty problematic. An account based on segments is in fact incompatible with BPS. Adopting the Inclusiveness Condition, Hornstein and Nunes propose an analysis that does not have to resort to ChomskyAdjunction but nevertheless captures the observation that adjuncts do not change the properties of the targeted phrase. The basic idea under Hornstein’s and Nunes’ view is to decompose Merge into the two operations in (15). (15) Two separate operations

i. Concatenate ii. Label

The first operation joins syntactic objects, while the second one provides a label to a syntactic object. According to the authors, adjuncts are sometimes integrated into the structure and sometimes not. Assuming that

58  Approaches to Adjuncts adjuncts are concatenated but do not need a label in every case, their behavior can be predicted. Consider (16) below. (16) a. [V eat ^ the-cake] ^ in-the-yard b. [V [V eat ^ the-cake] ^ in-the-yard] (Hornstein & Nunes 2008: 66) In (16) a. the prepositional adjunct is only concatenated with the verbal phrase while it is integrated into the verbal projection by means of labeling in (16) b. This would predict for the adjunct in (16) a. to be invisible to syntactic operations. Consequently, in this case, adjuncts are excluded from VPpreposing or VP-ellipses and Do-so-anaphora. Conversely, it follows that an adjunct being integrated into the VP as in (16) b. undergoes ­VP-preposing, VP-ellipses, and Do-so-anaphora with the rest of the verbal phrase. Semantically, adjuncts modify events in a Neo-Davidsonian sense according to the authors. They exemplify this with the example quoted in (17). (17) a. John ate the cake in the yard. b. ∃ e [eating(e) & subject(John, e) & object(the cake, e) & in— the—yard(e)] (Hornstein & Nunes 2008: 70) The gist of their reasoning here is that, in contrast to arguments, adjuncts do not need any additional mechanism to contribute to the proposition in (17) b. In this sense, the authors argue that adjuncts are the “normal” case, while arguments, against the common view, represent the “abnormal” one because they require support by means of grammatical roles in (17) b. Interpretation of phrases in terms of labeling would thus correspond to semantically embedding them into a proposition via the designated functions of subject and object in the Neo-Davidsonian format shown in (17) b. They further argue that their system accounts for the properties listed in (18). (18)

i. ii. iii. iv.

Optionality of adjuncts (unselected status) Multiple adjuncts, free ordering VP-ellipsis, VP-preposing, Do-so-anaphora Differences concerning the projection of focus

The target of syntactic operations must be labeled according to this view because the role of labels is to allow further computation and to identify relations. Consider (19) expressing this idea.

Approaches to Adjuncts 59 (19) Concatenation is defined over a set of atoms and labeling turns a non-atomic complex concatenate into a (complex) atomic element suitable for concatenation. In other words, what labels do is allow concatenation to apply to previously concatenated objects by bringing these complexes into its domain. (Hornstein & Nunes 2008: 65) Recall that it is an essential assumption of the authors that arguments require to be integrated by means of labels because the relations subject of, object of can only be interpreted at C-I with recourse to labels. Adjuncts modifying the event directly contrast with arguments in that they do not need labeling in order to receive an interpretation at C-I. This state of affairs is emphasized in the quote in (20). (20) To sum up the discussion so far, a labeled concatenate is a complex atom. Atoms have no accessible innards. By rendering a complex concatenate atomic, the label prevents the insides of the labeled elements from being targets of movement by the A/A condition. When adjuncts don’t move with the elements they modify, it is because they are not members of the labeled concatenate that has moved. [. . .] However, arguments can never be other than members of a labeled concatenate, for their interpretive lives depend on it. (Hornstein & Nunes 2008: 67–68) In what follows this quote, the authors address the consequences of their analysis for head movement, affix-hopping, relative clauses, and there-­ constructions. The reader is referred to Hornstein and Nunes for the details. As for the linear order of adjuncts, the authors stress that the surface order does not provide any information as to whether or not labeling has taken place. Also the mechanism for linearization seems to be distinct from those needed for arguments (Hornstein & Nunes 2008: 75). Thus, the view adopted here implies that adjuncts and arguments may not only be distinct in terms of syntactic computation (concatenate with or without labeling) but may differ on basis of their respective interpretation at the semantic and phonological interfaces with respect to the mechanisms needed. On the one hand, it seems to be good news not to have to resort to a specific operation, Pair-Merge, for adjuncts. On the other hand, the assumption about adjuncts being able to “dangle,” not needing to be integrated into the structure, clearly resembles the hypothesis of adjuncts

60  Approaches to Adjuncts existing in another dimension, understood here as an unconnected. This increases the computational procedure, especially in view of the fact that further interpretive rules are needed at PHON and SEM to fully capture the dichotomy of arguments and adjuncts. Another difficult issue is provided by the topic of movement. First, the question arises if there is a third operation called Move or if displacement should be analyzed in terms of internal concatenate. A follow-up question would be whether there is evidence for internal concatenate with and without labeling, which would of course be a logical implication. Also, it is questionable whether a revival of the A/A condition makes sense. If labeling produces atoms, one might also wonder whether this rules out extraction from VP in general. When labels along the clausal spine atomize the phrases they include, how can anything move to the top? Consider (21). (21) a. [DP Which book] did John [VP read ]? b. [PP In which garden] did John read the book ? Why does the VP-label not make its innards inaccessible? Is the label inexistent when the DP moves, or is this atomizing only effective in case of the A/A-condition? If syntactic operations require labeling, then whmoved adjuncts need labeling; this would not necessarily be the case with non-moved adjuncts. The adjunct in (21) b. must be labeled or it could not be involved in the further computation. In order to be interpreted correctly, there must be an integrated copy, hence a concatenated and labeled adjunct copy too. The interpretation of an adjunct in a copy position such as in (22) a. does not seem to be different, though, from a nonmoved adjunct as the one in (22) b. The same applies to (23). (22) a. In the garden John read the book . b. John read the book in the garden (and Bill did so in the park). (23) a. How quickly did he leave the room ? b. How quickly do you think that he left the room ? c. He left the room very quickly (and Mary did so very slowly). Another question concerns the interpretation of unlabeled adjuncts at SEM. It makes sense to say that labels are needed at the interfaces for interpretation. It is, for instance, uncontroversial that DPs may be

Approaches to Adjuncts 61 interpreted as entities, and this is not the case with VPs or APs. Hence, recourse to labels is crucial for the interpretation of a syntactic object. So, how can unlabeled objects be interpreted after Transfer? The NeoDavidsonian interpretation concerns the direct modification of an event by adjuncts. Still, the adjunct itself needs to be interpreted as well. Consequently, a label might be needed after all.8 Summarizing briefly, this seemingly simple approach, which deals with the asymmetric behavior of adjuncts in terms of labels, increases the computational procedure and raises nontrivial issues of interpretation. Furthermore, it leaves open the question of how to handle displacement in more general terms. A quite different approach sticks to the traditional assumption that projection follows from selection in the syntax. The basic issue here is which features project and thus label syntactic objects. Zeijlstra (2019) suggests that formal features determine the syntactic computation. Uninterpretable features (uFs) trigger the operation Merge in a way similar to proposals made in categorical grammar. A feature uF carried by X signals a dependency between X and F. Both are removed in a sister relation and do not project. This idea is exemplified in (24). (24) [DP D [uN] [NP [N]] The category D selects a noun, which is expressed by means of an uninterpretable feature [uN]. Upon Merge, [uN] and [N] cancel each other out and only the remaining feature, the interpretable [D], projects and thereby labels the product of Merge. Zeijlstra proposes a solution to adjuncts, according to which those phrases are not selected but select the syntactic object they modify. It is worth noticing that the general idea that adjuncts select the element they modify is not new. Bowers (2001) puts forward the hypothesis that modifiers explicitly select the phrases they modify. Starting with the observation that the two expressions walk slowly and eat lunch are both phrases consisting of two words forming a constituent, Bowers argues that the relation between them yet is different because slowly is a modifier of walk, whereas lunch is an argument of eat. As observed by him, it complicates the analysis of syntactic relations when the first is handled in terms of Set-Merge and the second relation would be established by means of a different operation, Pair-Merge. Conceptually, an approach that treats all relations on a par with each other is much simpler. Under the hypothesis that there is an operation forming relation (FR) yielding ordered pairs only, Bowers concludes that in eat lunch, the verb eat is the first coordinate of a binary relation (eat, lunch), whereas in walk slowly, the verb walk is the second coordinate of a binary relation (slowly, walk). Modification is thus the reversal of the order we find in the case of the

62  Approaches to Adjuncts sub-categorization relation. Consider the following passage taken from Bowers (2001), in which he explains the basic idea. (25) I propose next [. . .] that just as verbs are marked with features indicating the category of the LIs they select or subcategorize, so adverbs are marked with features indicating the category of the LIs they modify. Thus an adverb such as ‘perfectly’ has a selector of the form [__V], indicating that in an ordered pair of the form (perfectly, x), it must be the case that x ∈V. An adverb such as ‘probably’, on the other hand, has a selector of the form [__T], while an adverb such as ‘stupidly’ has one of the form [__v/Pr]. Notice that this immediately accounts for the otherwise puzzling fact that adverbial modifiers are always optional. Whether or not an adverb occurs in a given sentence simply depends on whether or not it is present in the lexical array. If it is, then it occurs; if not, not. The relational theory thus accounts in the simplest possible way for the difference between modification and subcategorization/selection within a unified theory of selection, while at the same time accounting for the fact that modifiers are always optional. (Bowers 2001: 43–44) Essentially, by adopting assumptions about immediate Spell-Out, which has a crucial impact on the way constituents and movement are analyzed, Bowers develops a rather non-standard but interesting approach. What is important in the course of our discussion here is the fact that Bowers provides a different view on modification with the direction of selection, reflecting the main difference between selection (including arguments) and modification (adjuncts). Similarly, Cormack and Smith (2005) argue from the perspective of categorial grammar that “it now becomes straightforward to assign a complex category label to an adjunct like ‘very’. If the item selects for an adjective and returns the category adjective after merge, then we must put Y = A and X = A in the tree in (7) [= [X X/Y X]], to yield (8) [= [A [A/A very] [A red]]]. More generally, we have the categories and tree in (8b) [= [X [X/X] = adjunct [X] = host]] for a simple head which is adjoined to a category X” (Cormack & Smith 2005: 119). More in the spirit of standard minimalism, Zeijlstra (2019) also argues for adjuncts as selectors. Building on the ideas taken from categorial grammar, he newly suggests that adverbs have symmetric features: they carry an interpretable and an uninterpretable feature of the same kind, namely [V] and [uV]. Consequently, there is no extra operation needed for adjuncts. There is just Merge and projection of features. Strikingly

Approaches to Adjuncts 63 the [V] feature of the adjunct (26) projects because the uninterpretable selection feature uV of the adverb and the interpretable V feature of its merged sister cancel each other out. The remaining feature of the adverb [V] percolates and labels the result. (26) [V] [uV], [V] = {often, sleeps} From this perspective, it follows that adjuncts do not change the category of their “host,” since they have a V feature which survives and projects upon merging.9 For PP adjuncts that select VPs under his analysis, Zeijlstra has to posit a feature composition similar to adverbs, but with an additional uD feature for the DP-object of the preposition. This is illustrated in (27). (27) [V], [uV], [uD] (= in) When P merges with D, uD being checked, D does not percolate. The remaining features on P, namely V and uV, project as the label. After merging with a verb, it is again the interpretable V feature of the preposition that projects, ultimately labeling the syntactic object. He further argues that the order of feature projection is flexible in principle and that the alternative order would be attested with particle verbs. According to the author, unwanted orders can still be ruled out at the interfaces as being not interpretable. We have repeatedly emphasized the dichotomy of arguments and adjuncts, which is reflected in the data. What does the account at hand predict concerning this observation? The account implies that PP arguments and PP adjuncts are syntactically identical in their feature composition. Under the premise that prepositions are composed of the features V, uV, and uD, one wonders how adjuncts differ syntactically from arguments if in both cases the V feature of the preposition projects. (28) a. He [VP went [V] [into [V, uV, uD] [DP the garden]]]. b. He [VP worked [V] [in [V, uV, uD] [DP the garden]]]. The examples in (28) would get the same projection structure. There is no obvious syntactic distinction between the prepositional object of (28) a. and the PP adjunct in (28) b. Under this view, prepositions can never be selected but must always select. The question is how to justify this direction of selection while explaining the differences between prepositional adjuncts and prepositional arguments. Prepositional adjuncts clearly contrast with prepositional arguments with regard to the possibility of

64  Approaches to Adjuncts Internal Merge, as illustrated in (29) for arguments and (30) for adjuncts with A- and A-bar- movement, respectively. (29) a. They talked to the man. b. The man was talked to c. Which man did you talk to ? (30) a. They met after the break. b. *The break was met after c. *Which break did they meet after ? To account for the optionality of adjuncts which contrasts with the obligatory status of prepositional arguments, one might well resort to interpretation at the interfaces and capture the distinction between, for example, He stared *(at the crowd) versus He relaxed (in the garden) semantically, but this would not account for (29) and (30). Though Zeijlstra argues for a principally flexible and free order, he now posits a difference in the order of projection to accommodate for the distinction: a predicate must first merge with its arguments before it can merge with an adjunct. The author suggests that arguments are distinguished from adjuncts by means of the mother node: if the mother carries an uD feature, it is an argument; if not, it is an adjunct. Since the distinction of argument versus adjunct is crucial for the interpretation at SEM, this raises the nontrivial question of how the interfaces can see uninterpretable features, which should have been checked and actually already have been deleted in the syntax. Recall that the distinction between arguments and adjuncts leads to a basic distinction in the respective analyses in the previous approaches we discussed. On the contrary, they are syntactically non-distinct unless one could derive the order of projection in a more principled way. In particular, one should investigate whether islands can be derived in this account. The following data in (31) clearly show a contrast in grammaticality. (31) a. b. c. d.

*Which course did he sleep during? Which course did he talk about? Which table did he put the book on? *Which conference did he meet her after?

Although DPs can easily be extracted from prepositional arguments, the examples involving Internal Merge out of adjuncts have to be ruled out somehow.

Approaches to Adjuncts 65 Multiple adjuncts as in (32) can be derived by the percolation mechanism, but adjuncts to adjuncts like those in (33) raise difficulties. (32) a. He sleeps calmly in the garden. b. He sliced the cake with a knife on his birthday. (33) a. He left the room surprisingly quickly. b. He left the room obviously on purpose. We get a derivation such as (34) for an example like (32) a. (34)

i. [V], [V, uV] = {sleeps}, {calmly}; feature V of calmly projects ii. [V] [V, uV, uD] [D] = {sleeps calmly}, {in the garden}; feature V of in projects

One can easily check that multiple prepositional adjuncts like (32) b. can also be derived. In all those cases, it would be the V feature of the adjunct that percolates and labels the syntactic object. What about adjuncts to adjuncts in (33)? Consider (35) exemplifying (33) b. for instance. (35)

i. [V, uV, uD], [D] = {on}, {purpose}; features of on (= [V, uV]) project ii. [V, uV], [V, uV] = {obviously}, {on purpose}?

What could be the percolating feature here? Since V uV on both Merge partners would cancel each other out, there is no feature left for the label of the resulting syntactic object. The same applies to the merging of surprisingly (V, uV) with quickly (V uV) in (33) a. Consequently, the approach is incompatible with adjuncts modifying other adjuncts. Stipulating that checking can be prevented, and both feature complexes percolate in order to label the merge product raises problems as soon as the projection goes on. Alternatively, one would have to assume that the joined feature set {VuV, VuV}, the features being identical, reduces to {VuV} which is checked in the next step when it is merged with a verbal projection. Nevertheless, this would be problematic because preventing checking in (35) ii. goes against Zeijlstra’s assumption that each Merge operation must have a purpose. Another set of data raises more serious questions. The idea that adjuncts project and label turns them into non-maximal projections. Under this view, wh-movement of adjuncts turns out to be problematic. Consider the examples in (36).

66  Approaches to Adjuncts (36) a. In which room did he work? b. How slowly did he leave the room? c. Where in Göttingen can you buy computer software? In the first example, the maximal projection is [V-max work [in which room]] because the preposition with the feature complex [V uD uV] projects its V-feature. Hence, a non-maximal projection would be internally merged to C, which is not possible and unattested under standard assumptions (see 37). (37) a. b. c. d.

*Which did he meet girl? *Brother of Mary John met the nice Which girl did he meet ? The nice brother of Mary John met

In essence, the analysis predicts that neither adverb phrases nor prepositional phrases are maximal projections. Since they both select their “host,” they only form maximal phrases containing their host. Hence, as being non-maximal themselves, they are predicted to be immobile, contrary to the facts. The tree diagram in Figure 4.1 illustrates this situation. A further empirical question is raised by adverbs and prepositional phrases, which modify or select modifees other than verbal expressions. Recall our discussion of distinct layers vP–TP–CP in Chapter Two. In (38) a. the adverb unfortunately takes a prepositional phrase for Mary as a complement. The whole adjunct is placed in a high position where it scopes over the full sentence. In (38) b. the same applies to the adverb– adverb combination. (38) a. [Unfortunately for Mary] they could not fix the car. b. [Quite obviously] they did not fix the car.

Vmax Vmax left

the room

V, u V non-max (how) slowly

Figure 4.1 Non-Maximality of Adjuncts Under Zeijlstra’s Analysis

Approaches to Adjuncts 67 One would have to assume that C (and also T) constitutes a verbal projection too, perhaps, with C (and T) as sub-features. In addition, one would have to stipulate that V uV in the case of adverb–PP/adverb– adverb combinations would not cancel each other out, but percolate to label the result. The checking would be forced to be delayed for the merged syntactic object to get a label (see the previous discussion of (35)) and for uV to be still available to select the extended verbal projection.10 The required featural composition for each example in (38) is illustrated in (38). (38) a. {{{V uV},{VuV uD}}, {V C}} {{{Unfortunately}, {for Mary}}, {they could not fix the car}} b. {{{Vu V},{V uV}}, {V C}} {{{Quite}, {obviously}},{they did not fix the car}} Whether we should argue for a delayed checking to handle data like this is doubtful. The stipulative flavor of such a solution becomes clear as soon as we consider this as a general option. First, Merge must always have a purpose in this approach. Second, one would expect other categories to occur recursively in the above sense, too. Yet it does not seem possible to merge two SOs bearing identical features: for instance, C-C, T-T or D-D does not seem to be empirically attested. Since PPs can also modify AP and NPs as in (39), Zeijlstra discusses the possibility of predicative super-categories11 carrying the symmetric features [Pred, uPred]. With those features getting valued in the course of the derivation as either V or N shown in (40), it would follow that prepositional phrases can occur in nominal and verbal phrases. (39) a. the [NP man [PP in the garden]] b. They [VP met him [PP in the garden]]. (40) P = Pred, uPred, uD a. merged with [pred: N] b. merged with [pred: V] Roots thus turn into [Pred:_] under this analysis and get the feature specification [V] or [N] if they are merged with an item bearing uPred: [V] or [N], respectively. In order to derive attributive adjectives modifying nouns, he extends the approach to inflectional elements. The overall system is clearly complicated by those features and values. If roots bear a [pred] feature, they have to be valued by either [V] or [N].12 Consequently, [V] and [N] are values, not features. Are there features [V] or [N], or are

68  Approaches to Adjuncts those elements feature values in all cases? What is the difference between a feature, a feature value, and a sub-feature, if there is any? Besides, some adjuncts cannot be easily subsumed under adverbs or prepositional phrases. Consider the set of examples in (41). (41) a. b. c. d.

The man [that/who/ø Mary talked to ]. He bought a new car [so that he did not need to take the bus]. He left [to do his homework]. They met him [yesterday].

In (41) a. we have a relative clause which is standardly analyzed as a CP. Analyzing the infinitival particle to on basis of the homophonous directional preposition to as a preposition is at least questionable. In (41) d. the adjunct is nominal in nature. Expressions such as yesterday’s lecture provide evidence for this observation. Do we have to resort to a silent preposition selecting the DP? In short, shall we analyze any adjunct as bearing the feature [VuV] and any non-adjunct of the clausal spine as bearing a verbal feature? A general theoretical issue concerns the relation between syntax and semantics. Zeijlstra explicitly argues for a free, flexible ordering (Zeijlstra 2019: 16). In those situations in which the order seems to be fixed, the fixedness shall follow from extra-linguistic factors at the interfaces (Zeijlstra 2019: 17). Orderings as in (42) a. and b. with P uV uD first merging with V and then with D, and T first merging with D, and then with the verbal projection would thus both be ruled out semantically at the interfaces. (42) a. *He [[in worked] the garden]. or *He [[worked in] the garden] b. *[reads the book slowly] [T uV, uD [the man]]. Under an approach resting on free Merge instead, without labeling taking place in syntax, Merge is completely blind to content. Interpretation is related to the interfaces only. It is theoretically much simpler to have the structure-building device apply freely and distinguish this from interpretation at the interfaces in the first place than to posit a trigger for Merge, assign labels in syntax contributing to interpretation, and rule out orders by means of interpretation at SEM, as suggested in the present account. Furthermore, the author constrains checking to elements standing in a sisterhood relation. Sisterhood of uF and iF can, in principle, be obtained in three different ways under this account. Consider (43), showing the options upon which the system is based.

Approaches to Adjuncts 69 (43)

i. Via percolation of uF ii. Via Internal Merge of iF iii. Via External Merge of uF and iF

Now, if we first merge an element with a uF, and at a later stage we merge a corresponding element with iF, then we can apply percolation, but if iF is merged earlier (lower), we apply Internal Merge later. In addition, percolation seems to be a further operation, next to Merge, and next to Agree. Depending on the respective order in which elements are merged, one gets a distinction in the interaction of the operations at hand, namely (44). (44)

i. Applications of Merge → checking (Agree) ii. Checking (Agree) → applications of Merge

Viewed this way, Merge restricts checking and checking restricts Merge. The order in which the features are merged thus would be relevant as well. Given the general percolation of uF, a rather complicated system emerges.13 To sum up, we discussed several highly interesting and smart approaches to adjuncts. There is empirical coverage in each of them, but they also raise questions and problems open to future research. For the discussion at hand, it is noteworthy that the analyses we addressed in this chapter either require an extra-operation (like Pair-Merge or Late-Merge), or they build on labeling (no labeling of adjuncts or labeling by adjuncts), which presupposes a purpose for Merge (requiring additional feature and percolation mechanisms to hold). However, if we aim at a theory compatible with the strong minimalist thesis, one that allows for UG to be as simple as possible, we should keep the following two aspects in mind: (45)

i. Different dimensions always complicate the computation. ii. Triggering by features complicates the computation too.

SMT suggests that there is a system with an efficient and simple computation perfectly satisfying interface conditions leading to SEM and PHON interpretations. Table 4.2 presents an overview of the approaches we have discussed so far in terms of simplicity only, in order to evaluate how close they get to the SMT requirements we observed earlier.

70  Approaches to Adjuncts Table 4.2 The Simplicity of Minimalist Approaches Two Dimensions

Same Dimension

Labels in syntax/ syntax = triggered

Stepanov (2000, 2001) Late-Merge of adjuncts

No labels in syntax/ syntax = free

Chomsky (2013, 2015) Pair-Merge of adjuncts

Zeijlstra (2019) projecting adjuncts Hornstein and Nunes (2008) unlabeled adjuncts

There is an obvious gap here. Adjuncts are either analyzed with the complication of two dimensions, or they are analyzed with the complication of triggering syntax, which is a departure from the idea that the basic property of human language can be derived from Simplest Merge without resorting to any interpretational aspects such as featural asymmetries. We also find the doubly complicated analysis of adjuncts, namely, assuming featural triggers and separate dimensions. Is it possible to fill the blank field of the table? Is there a way to analyze adjuncts in the same dimension as other syntactic objects and keep to Simplest Merge as set formation without labeling? This would, of course, be a welcome simplification in accord with SMT, which we will investigate more thoroughly in the next sub-chapter.

4.3 A Simplification Ahead? As remarked at the end of the previous section, investigations of adjuncts either refer to different dimensions or analyze them in terms of features/ labels determining the syntactic assembling procedure. So far there is no account for adjuncts which adopts a free syntax and implements adjuncts without positing timing or local distinctions. It is also worth noticing that we have observed that two main directions in minimalist theorizing have immediate precursors in Government & Binding as shown in (46). (46) GB  Minimalist Approaches

i. Functional heads → feature triggered syntax/labeling ii. Adjunction → Pair-Merge

Before we consider the question as to whether there are minimalist approaches to adjuncts which point into the direction of SMT-compatible simplifications, let us take a look at some quotations covering an impressive time span of ten years of research in minimalism.

Approaches to Adjuncts 71 (47) Nobody seems to know exactly what to do with adverbs. The literature of the last 30 years in formal syntax and semantics is peppered with analyses of the distribution or interpretation (or both) of small classes of adverbs but has few attempts at an overall theory; there have been popular proposals for other phenomena based crucially on assumptions about adverbial syntax that have little or no foundation; and almost everyone who has looked at the overall landscape has felt obliged to observe what a swamp it is. (Ernst 2004: 1) (48) More fundamental questions arise at this point. There has never, to my knowledge, been a really satisfactory theory of adjunction, and to construct one is no slight task. That has been recognized since the earliest efforts to construct a reasonable version of phrase-structure grammar, what came to be called X-bar theory. (Chomsky 2004: 117) (49) The question of how adverbs/adjuncts are integrated into clause structure constitutes a long-standing puzzle. (Alexiadou 2004: 1) (50) It is fair to say that what adjuncts are and how they function grammatically is not well understood. (Hornstein & Nunes 2008: 57) (51) If the study of adverbs once represented a “swamp,” as was sometimes remarked forty years ago, it is perhaps now low-lying marshy ground, messy and prone to flooding, but with a few solid paths meandering through it. (Ernst 2014: 108) It is particularly interesting that everyone acknowledges that adjuncts are puzzling and have to be assigned a proper analysis in the given framework. So far, we also have seen that a lot of promising research has

72  Approaches to Adjuncts been done on adjuncts. Besides the contributions explicitly referred to and commented on, far more scholars have addressed this controversial issue than are picked up in the discussion at hand. Adjuncts are a longdebated issue in the generative tradition, and still the only real agreement between linguists is that those optional SOs called adjuncts resist a generally accepted treatment. What can we do today with this common linguistic phenomenon? Is it reasonable to restate the GB-analyses in a more recent minimalist setting? Evidently, the hypotheses put forward are conceivable, and each captures a lot of the patterns attested with adjuncts. However, it should be noticed that an account that could fill the gap in the table at the end of the previous section would be tempting. Hence, is there a further simplification ahead? What would an approach to adjuncts that has the property described in (52) look like? (52) Adjuncts are built in the same dimension as any other phrase is in an un-triggered, free syntax with Simplest Merge as the only syntactic operation available. The following discussion serves to only highlight some recent ideas in minimalist theorizing in search of an analysis that satisfies the characterization given in (52). First let us recapitulate in (53) some relevant aspects which should be kept in mind as we go on. (53)





i. There is only one syntactic operation, Simplest Merge, in accord with minimal computation. Merge applies freely and it is not triggered. ii. The distinctive properties of adjuncts must be captured. The dichotomy of adjuncts and arguments, for instance, the island behavior of adjuncts, and the fact that adjuncts occur unbounded in principle, are real. iii. Semantically, an event can be specified further by assembled adjuncts in terms of, for instance, manner or locative, spatial, temporal pieces of information. Similar observations hold of the modification of nominal entities. iv. There is an interaction between Merge (Narrow Syntax), and the conceptual-intentional system (C-I)/SEM. Either Narrow Syntax satisfies the conditions of C-I or Narrow Syntax shapes the conditions of C-I, or this is conceived of as a real interaction that works in both direction (see (54). This is quite plausible on the assumption that Merge yields binary sets which are interpreted. Full Interpretation, a condition of the interfaces, requires that the output of Merge is interpretable.

Approaches to Adjuncts 73

v. Are adjuncts outside any c-command domain in a derivation? C-command relates SOs and should be defined in terms of the derivation as elaborated by Epstein (1999) who points out that X c-commands the terms of the category Y with which X was merged externally or internally in the course of the derivation (see Epstein 1999: 329).

(54)

i. Merge → SEM ii. Merge ← SEM iii. Merge ← → SEM

Consider the sets of sentences in (55) taking up the points mentioned above. (55) a. John is working on a porch. b. What is John working on? c. John is working there. The sentence in (55) a. is structurally ambiguous, between on a porch being analyzed as the prepositional object of work, or an adjunct providing some locative information. The examples in (55) b. and (55) c. provide evidence for this distinction to be made in structural terms, because in b., under extraction, only the argument reading survives, and in c., it must be an adjunct that is referred to by the pronominal expression there. Therefore it seems reasonable to maintain the distinction between adjuncts and arguments. The data in (56) point in the same direction. (56) a. John is waiting on a train. b. John is waiting on a friend. c. John is waiting on a friend on a train. Again, the a-example is ambiguous. John could either be sitting on a train and waiting there for whatever you can think of, or it is the train that he awaits. One might, however, argue that this is just a semantic issue, and the semantic content of the DP would be responsible for the two readings. Compare with (56) b., which strongly suggests the reading that he is looking out for his friend, the local reading being pretty weird. Yet, this would oversimplify the facts, because there are structural effects, as (56) c. illustrates. Adding a prepositional phrase, we can only analyze the second one as an adjunct (either related to the clause or to the noun friend). Structural ambiguities provide clear evidence for structural distinctions. In fact, they serve as a standard diagnostic. This can be further demonstrated with (57).

74  Approaches to Adjuncts (57) a. b. c. d. e.

The children saw the man standing at the window. Where did the children see the man standing? The children saw that the man was standing at the window. The children saw the man while they were standing at the window. The children saw the man who was standing at the window.

The linguistic expression standing at the window can be an adjunct to the man (with the reading in (57) e.) or to the children (see (57) d.), or it can be a non-adjunct, namely, the predicate of an embedded clausal expression taking the man as its subject (compare with (57) c.). Interestingly, any adjunct reading is banned in (57) b. which follows from the standard assumption that adjuncts in contrast to complements are islands to extraction. The example in (58) confirms the observation that multiple adjuncts occur in natural language. We can stack adjuncts; iterate them; and, in principle, add any number of adjuncts (which is only restricted in performance, but not in terms of the competence of native speakers). (58) John bought the painting reluctantly on Monday in the gallery at the airport during his break before he met Jane in the garden next to the university. The question of whether adjuncts are somehow syntactically isolated, so that they do not participate in any (c-command) relations is problematic. Take a look at (59). (59) a. Johnk looked for a camera [to take a picture of himselfk]. b. Hek looked for a camera [to take a picture of John*k]. c. Theyk gave John a camera [to take a picture of themk/j]. d. They gave himk a camera [to take a picture of John*k]. If principle A ((59) a.), principle C ((59) b. and d.), and principle B (59 c.) can be invoked here to explain the binding co-references, the adjunct, that is, the purpose clause must stand in a relation to the main clause at the point where those expressions are interpreted.14 The binding data investigated in the literature are rather complex, and they may not necessarily lead to clear and unequivocal results as we stressed out earlier. We have, however, repeatedly emphasized that there is good evidence for a distinction between adjuncts and arguments. Given the five points noted in (53) and the ultimate goal of minimizing UG, it makes sense to look at some more recent accounts which focus on a syntax that does not contain

Approaches to Adjuncts 75 a specific operation for adjuncts, and which aim at explaining the distinction between adjuncts behaving as being “invisible” to syntax and arguments which behave as syntactically integrated into a structure. It is worth stressing that in this section, we will only focus on very central ideas and concepts suggested by the respective linguists. For a detailed presentation of the analyses, the reader is referred to the scholars themselves. The research considered here is explicitly committed to SMT. Contrary to the approaches in the previous section, the scholars lay a focus on the architecture of the grammar to implement adjuncts. Strikingly, they refer to the labeling algorithms we presented in Chapter Three and to Transfer (or Spell-Out). As we will show, the first three accounts—though they build on new ideas—rather look like new versions of previous accounts. They will be presented in chronological order before we take stock. One new approach can be found in Hunter (2015). He basically suggests that for YP to be adjoined to XP means to be inserted into the structure and spelt out later. Crucially, YP does not have to be merged, which under his analysis also means being connected to the structure by features. Consequently, the syntactic operation Merge is again viewed as triggered. According to him, the computation uses the operations Insert, Merge, and Spell-Out. Under this view, movement of a phrase XP is reanalyzed as XP waiting to be merged, until it combines with a feature. Adjuncts are not merged, unless they have a feature triggering the integration into the structure. His approach is interesting in so far as he can capture the optional status of adjuncts, their iterability, and the island effect observed with adjuncts by means of the way the computation takes place. This is briefly illustrated in (60) and (61). (60) Three Operations Insert → Merge → Spell-Out = putting into syntax → connecting by means of features → PF/ LF-interpretation Applying or not applying the operations to the italicized linguistic expression, yields the options given in (61). The syntactic object operated on is marked in italics. (61) Application of the Operations

i. ii. iii. iv.

Insert, Merge, Spell-Out: see Mary Insert, Spell-Out: sleep quietly Insert: see who (to be moved = merged later) Merge, Spell-Out: who C

The ideas Hunter puts forward here are in some sense similar to the analysis proposed by Hornstein and Nunes. In their view, Merge is decomposed

76  Approaches to Adjuncts into Concatenate and Label. Adjuncts are not really integrated into the structure since, according to Hornstein and Nunes, integration takes place by means of labeling only. Hunter resorts to features requiring or triggering the integration by Merge and assumes that Insert is a further operation. Since feature projection is basically labeling, the parallels are evident: Merge is again triggered by features in Hunter’s analysis. Hence, this seems to be a new version of the non-integration account we found in the non-labeling analysis of Hornstein and Nunes (2008). Oseki (2015) wants to eliminate Pair-Merge. He combines aspects of three analyses, namely, the principles of labeling based on Chomsky (2013), the suggestion of “two-peaked” structures inducing Transfer by Epstein, Kitahara and Seely (2012) and Hornstein’s (2009) Label Accessibility Condition which says that only the label of a syntactic object is accessible to Merge. Oseki basically argues that on the one hand, PairMerge is too-restrictive since some adjuncts are actually transparent for extraction, but it is too loose on the other hand because some adjuncts get reconstructed with respect to binding principle C. This amounts to saying that some adjuncts are visible to syntax and others are invisible. If a two-peaked structure with an intersecting XP and YP (an adjunct) results in an unlabeled structure, as in Figure 4.2, it follows that the adjunct YP is inaccessible, hence invisible. Transfer is needed to allow for the derivation to go on. Oseki further assumes that there are also visible adjuncts in a XP, YP configuration that can be labeled through a prominent shared feature. Two predictions follow on this view. First, the derivation goes on since the set {XP, YP} is labeled, and second, the YP adjunct is transparent for extraction. We summarize the empirical predictions in the Table 4.3.

ZP Z

? Transfer XP

YP (adjunct)

Figure 4.2 Two-Peaked Structure

Table 4.3 Predictions of Oseki’s Account Attested

no extraction from adjunct Attested extraction from adjunct Not possible no extraction from adjunct Not possible extraction from adjunct

no reconstruction invisible adjunct reconstruction reconstruction

visible adjunct visible and invisible

no reconstruction visible and invisible

Approaches to Adjuncts 77 It seems quite plausible that the paradox of being invisible and visible cannot be solved in any obvious way. Thus, Oseki explains why adjuncts with doubly-peaked structures behave as islands and are inaccessible to reconstruction effects, while those dominated by a labeled node are not. Note, however, that under this view labels are part of syntax again. Hornstein’s Accessibility Condition drives the computation and makes labeling part of Merge again. Another question concerns the Feature Sharing in cases like What did John [feature? [arrive] [whistling ]. Oseki resorts to accusative case to get a Feature Sharing with transparent adjuncts. Alternatively he also mentions a possible agreement via an ASP head. Since the former account is concerned with accusative case, the question arises of whether it is justified to apply this to the example at hand. There is no accusative case with an uncaccusative item such as arrive. Consequently, only the latter option might be investigated further. As for the empirical predictions illustrated in Table 4.2, it says that the impossibility of wh-movement out of an adjunct goes hand in hand with the invisibility of an adjunct in terms of the binding principle C. Consider (62) now. (62) a. *Who did he cry [after John hit ]? b. *[Cry after Johnk hit him] hek did Obviously, these data correspond to what should be impossible, namely, an adjunct which is invisible (no extraction from the adjunct via twopeaked, unlabeled structure) and visible (reconstructed in term of principle C of the binding theory) at the same time. Oseki’s analysis thus falsely predicts the pattern in (62) to be unattested. In general, a solution according to which adjuncts are not implemented by means of an extra syntactic operation Pair-Merge would have a desirable and simplifying impact on the syntactic computation. However, though there is no Pair-Merge operation, Oseki needs distinct kinds of syntactic objects, namely, normal sets and intersecting sets (two-peaked structures), and he puts forth an analysis with labels needed in the syntactic computation. Consequently, it seems that we deal here with a new version of the local dimension account combined with features and labels triggering the derivation. The next account is based on Nakashima (2017). He combines the labeling algorithm of Chomsky (2013) with a specific label-based conception of the No Tampering Condition (LNTC), allowing counter-cyclic Merge. According to the author, Merge of X and Y leaves the two SOs unchanged if X or Y is dominated by a labeled node, but Merge can countercyclically apply when the dominating node is unlabeled. Interestingly,

78  Approaches to Adjuncts this nicely captures A- and A-bar asymmetries. Consider Nakashima’s (14) a.–b. given in (63) a. –b. (63) a. *[Q [? Q Which] [NP claim that Johnk was asleep]] [C[Q] was hek willing to discuss]]? b. [? [? Q The] [NP claim that Johnk was asleep] [T seems to himk to be correct]]. Assuming that nominal expressions are actually composed of Q, D, and N, with N bearing phi-features, the example in (63) b. allows for LateMerge of the NP containing the R-expression because the dominating node would be unlabeled at this stage. Q and D are merged but do not share a feature, and this object being merged with the TP would also lack a label since there is no shared feature at this stage either. The NP can be counter-cyclically merged under the LNTC, which would yield a φφ label in the end. A counter-cyclic merging of NP is not possible in the case of wh-movement in (63) a. because the dominating node is already labeled via feature [Q] shared by [Q D] and the CP. Note that the effect of the LNTC is to restrict the possibility of Late-Merge which decreases the computational burden counter-cyclic Merge normally implies. According to Nakashima, Late-Merge is ruled in when a node is not labeled. One objection deals with the search for a label. How deep can Minimal Search actually be? In (63) b., the φ is found inside the embedded NP. Another objection concerns the empirical coverage. If we keep to the LNTC and the analyses suggested for the data in (63), we will not be able to explain the adjunct–argument asymmetry we actually started with. To see this, consider (64). In terms of labeling and the LNTC, the examples in (64) are not different. In both cases, the dominant node would be labeled with Q-Q, hence ruling out Merge inside of the node. (64) a. *Which [claim that Johnk was asleep] was hek willing to discuss? b. Which [claim that Johnk made] was hek willing to discuss? The general idea to allow Late-Merge, but to prevent wild counter-cyclic Merge, is actually in line with a minimal setting, but we face serious problems with Late-Merge. Basically, this is still just a new version of the temporal dimension accounts we discussed earlier, though it builds on the most recent labeling algorithm. Similarly, Saito (2017) attempts to constrain the option of Late-Merge by resorting to the more recent concept of a phase. He suggests revising the PIC, so that complements of a phase head are still accessible though not internally. This has the interesting effect of ruling in iterated

Approaches to Adjuncts 79 Late-Merge obeying cyclicity. However, the question arises why spelling out of the complement renders the material internal to the complement inaccessible, but leaves the complement, or the label, intact for operations to apply here. The next set of analyses illustrates that labeling and Transfer might be applied in specific ways to handle adjuncts. Again, we are concerned with the basic ideas of the respective accounts which will be briefly investigated in a chronological fashion. The first scholar to consider, Gallego (2010), refers to the second LA of Chomsky (2008) that we presented in Chapter Three and Talmy’s concepts of figure and ground. Building on the latter, he starts with the observation that data like (65) a. and b. (his (13) and (14): 191) are similar in that they share a structure such as (66): (65) a. Peter works in New York. b. Peter is in New York. (66) [pP = applicative head [V*P Peter works] = figure [p’ p [PP in New York] = ground]]] Gallego assumes that a VP (as a figure/subject) is merged to an adjunct (a ground/predicate) which actually yields an unstable XP, YP product, in the case at hand, VP, PP. He further argues that the results cannot be labeled by the LA, since there is neither a head nor a movement target involved in those XP, YP cases. The resulting labeling failure is, according to him, the very reason for adjuncts (PPs) as behaving like being on a separate plane. Besides explaining the island behavior of adjuncts, the labeling failure would also account for the fact that adjunct do not get theta roles like arguments. The suggestion in turn is that little-v, as a phase head, merges with the unstable, unlabeled product of VP and PP. VP and PP are transferred together. Multiple adjuncts are simultaneously merged to VP in a parallel fashion on his account. Hence, this requires a form of multi-dominance or Parallel Merge. Take (66) as an illustration of the simple case with one adjunct only. (66) [vP v [? [VP works] [PP in New York]]

i. The node? is transferred (complement of phase head v) ii. The whole set is labeled by vP which dominates the? iii. Little-v searches ambiguously, but due to feature inheritance between v-V it is not PP that is searched into. The vP is the goal, the PP is the island.

The necessary assumption (66) iii. shows that it is not just labeling failure that can derive the properties of adjuncts. The question also arises how

80  Approaches to Adjuncts the set labeled with the question mark can finally get a proper interpretation. Note that integration of adjuncts in the structure is, according to Gallego (2010), possible by Internal Merge of the PP to the edge of the construction. In this case, adjuncts can participate in binding relations. The view adopted here does not, in fact, resort to Pair-Merge. Though it is interesting to deduce the properties of adjuncts from the failure of labeling, new questions arise. How can the SEM interface deal with unlabeled objects? If the interfaces could handle unlabeled syntactic objects, a follow-up question would be what calls for labels in case of labeled objects in the first place. What syntactic status, if any, does Parallel Merge have? The analysis requires it to be a syntactic operation in order to account for multiple adjuncts. Adopting Parallel Merge for multiple adjuncts basically means that we return to the separate dimension which clearly complicates the computational system as emphasized on the outset.15 Still another view is presented in Narita (2012, 2014). Narita argues for a projection-free syntax. He assumes that there is a condition on Merge, the H-Alpha-schema, requiring that every Merge operation involves (at least) one head. Apparent XP, YP mergers reduce to the H-Alpha-schema if either XP or YP, or even both have a phase head that induces Transfer of the complement before being merged with a SO. This way, apparent XP YP merged SOs can be handled if either X or Y (or both) are phase heads. Transfer leaves the phase head in the syntax. In line with the H-αschema, Merge takes a head and combines it with another SO (either head or non-head). See the three options we get in (67). (67) XP YP Three Options:

i. X and Y are phase heads (apparent XP, YP = X, Y) ii. X is a phase head (apparent XP, YP = X, YP) iii. Y is a phase head (apparent XP, YP = XP, Y)

Note that in some cases phase heads have edges (“specifiers”) which are not transferred, hence in those cases the first option comes into play. What is interesting to us now is that the H-Alpha-schema requires that with adjuncts, one of the SOs must have a phase head. Either the adjunct or the host, the target of adjunction, would have to be a phase head inducing Transfer. Narita proposes that the adjunct has a phase head and its complement is transferred, which automatically turns it into an “island” for syntax. This is illustrated in (68) below. (68) *The girl John failed the test [because he was thinking about ]  apparent (TP, CPadjunct) = (TP, Cphase)

Approaches to Adjuncts 81 So, if in fact, adjuncts are composed of Hphase XP, and XP gets transferred before Merge applies to Hphase, the island property of adjuncts automatically follows. Note, however, that a projection-free syntax does not completely abandon labels from the architecture of the grammar. Though projections or labels do not determine the syntactic computation, they are still needed in the sense of what Chomsky calls object-identification necessary at the interfaces for interpretation. If the adjunct head in Narita’s analysis is a phase head which remains in the derivation, then we are left with a SO consisting of C becausephase head and TP (the “main” clausal spine). What is the label of this syntactic object in terms of interpretation? If a label is identified by means of head-detection (Minimal Search)16 this would actually make the head of the adjunct the label of the entire set. It seems a bit counter-intuitive to assume that a prepositional phrase (the head of PP) merged with some XP (the main clause) identifies the syntactic object at the SEM interface. This reasoning applies to both, adjuncts occurring in clauses and in nominal entities as illustrated with (69) and (70). (69) a. He left the room after the lecture. b. TP after= phase head? = {after {TP, after}} c. *Which lecture did he leave the room after ? (70) a. She is very fond of the man with the dog. b. NP with= phase head? = {with {NP, with}} c. *Which dog is she very fond of the man with? The adjuncts are islands in both cases. Hence, according to Narita’s analysis, SOs inside of the adjunct have been transferred, leaving the phase head posited for adjuncts behind. This adjunct-head would thus ultimately identify the newly formed object. Narita (2014) explicitly argues against the most recent LA of Chomsky (2013) by stressing that {XP, YPadjunct} products are stable, which is basically the reason why the asymmetric behavior of adjuncts is analyzed as a consequence of Pair-Merge by Chomsky and others. In general, Narita’s analysis stands and falls with endocentricity taken as a universal property of structure-building and the question of whether (69) and (70) can be made sense of. Another quite recent approach is Jakielaszek’s (2017). Rightly, he stresses that it would be an enrichment of the theory to have orderedpairs with a built-in mechanism for a specification about which of the objects is to be the host. Yet, the latter point can be argued to be solely

82  Approaches to Adjuncts part of the interpretation and not part of syntax in the first place. He adopts a theory without Pair-Merge and resorts to the most current LA (Chomsky 2013, 2015). According to this LA, merging of an adjunct XP to YP yields a labeling conflict. Jakielaszek suggests now that adjuncts are internally self-merged. Consider Figure 4.3 and (71). Since the higher copy can never contain the lower occurrence in case of Internal SelfMerge, it behaves as non-existent according to the author. Also, lower copies are said to be invisible for the LA. Consequently, both copies are predicted to be invisible under Jakielaszek’s (2017: 63) reasoning. (71) Operations in Phrasal Adjunction

i. Take XP and YP ii. Self-Merge XP iii. Merge the output of ii. with YP

Self-Merge takes place before the two syntactic objects are merged in step iii. Consequently, adjunction structures are acceptable because the entire adjunct is made invisible for the search mechanism, and the lack of label is accepted by the procedure. Jakielaszek assumes that this invisibility also captures the fact that there is no featural relationship between the internally self-merged object and its sister. The copy-like character implies that “its interpretive import should come to the import of other kinds of copies as close as possible—without being entirely identical, for the C-I component cannot be blind to the absence of a head of such a putative chain” (Jakielaszek 2017: 66) The semantic interpretation of this extremely local chain is determined by a semantic function (for details see Jakielaszek 2017: 64–76). Another effect he proposes is that adjuncts are expected to be frozen, because displacement, under his view, destroys the configuration necessary for the labeling algorithm (Jakielaszek 2017: 75). This last point seems doubtful on empirical grounds because, as emphasized throughout, adjuncts can be internally merged to C. Hence, they are not frozen in place and the configuration the interpretation rests on would be destroyed without causing any problems. The approach advocated here implies that a lack of label can be tolerated if Minimal Search does not see the adjunct, and the LA does not yield a conflict at all. In order for this analysis to go through, one needs to accept his conclusion that the adjunct is simply invisible. Under standard assumptions,

XP Figure 4.3 Internal Self-Merge

YP

Approaches to Adjuncts 83 (in)visibility implies that higher occurrences properly contain lower ones, which is said to make the lower ones invisible. Jakielaszek’s view requires that containment need not obtain. This makes the whole object invisible for the LA. Containment of lower copies is in the standard case a consequence of Internal Merge applying cyclically to a higher position not violating the NTC. Consequently, the basic question is whether or not Self-Merge is a general option in the system. First, merging in this reflexive fashion can only be internal. It is a kind of vacuous operation. Note that copies follow from the third factor–driven NTC/LCR because a relation established by means of Merge shall not be changed or destroyed later. However, this reasoning does not apply to vacuous, internal SelfMerge. Why should there be a copy in the first place? XP being internally self-merged to XP does not change any relation XP has already entered. Is the relation of XP to itself changed without the copy? This sounds a bit weird. What effects can internal Self-Merge have at the interfaces? Another issue would be the phenomenon of adjuncts to adjuncts. One internally self-merged object would be merged with another internally self-merged object—both being invisible? Two basic insights of Jakielaszek (2017) are correct, the first being that the most recent LA does not deal with adjuncts, and the second one acknowledging that Pair-Merge is an additional burden for the syntactic component, but one should further investigate whether vacuous Self-Merge can really be an option in minimalist syntax in line with computational efficiency. The Table 4.4 concludes our discussion in this section. Strikingly, in one form or other, labeling or Transfer or both play a crucial role in the accounts presented in Table 4.3. In general, there is a clear tendency in all the approaches to favor a syntactic analysis of adjuncts that does not resort to specific operations, which would only be necessary to capture adjuncts. This general direction accords well with the SMT and the history of generative grammar, which is nicely characterized by Epstein, Kitahara and Seely (2017) stressing the discussion on labels. Table 4.4 Overview of SMT-Accounts to Adjuncts

Hunter (2015) Oseki (2015) Nakashima (2017) Gallego (2010) Narita (2012, 2014) Jakielaszek (2017)

new “version” of a previous account

relating to Labeling Algorithms (Chomsky 2008 or 2013)

relating to Transfer (or Spell-Out)

yes yes yes no no no

no yes yes yes yes yes

yes yes no yes yes no

84  Approaches to Adjuncts (72) In part, this is a history of the simplification of a central aspect of syntactic theory, namely that labels were explicitly represented in the syntactic objects that constitute the representational output of the structure building mechanism(s). But, over time, labels and label projection were eliminated from the syntax. The structure building mechanisms have changed over the course of the development of generative grammar and, with these changes, we find different notions of label and label projection. A number of researchers, including Collins (2002) and Seely (2006), argue for the elimination of labels and labeling entirely. In Chomsky’s most recent work, however, the effects of labels, which are not explicitly represented in syntactic representations, are derived from the application of independently motivated, third-factor mechanisms (specifically minimal search), and with interesting empirical consequences. (Epstein, Kitahara & Seely 2017: 18) Aiming at the most minimal version of UG, one first and foremost needs set formation by Merge in the syntax and Transfer of the sets to the interfaces, where sets can be interpreted by PHON and SEM rules respectively. Third-factor principles of minimal computation, which are independent of the faculty of language, ensure an efficient computation. Consequently, an additional operation like Pair-Merge would need empirical and theoretical justification. Do we need a second structurebuilding operation Pair-Merge? If there is a second operation, Pair-Merge besides Set-Merge, the question to be raised is whether we can find more evidence for the options of Pair-Merge in language. Consider Table 4.5. Richards (2009) suggests that internal Pair-Merge should be possible to handle scrambling data. Epstein, Kitahara and Seely (2016) argue for External Pair-Merge of heads17 to derive differences among complements (DP versus CP) and to get some pre-syntactic morphological effects. Thus, it is a theoretical and an empirical question whether to allow for Pair-Merge too. Nevertheless, Pair-Merge (plus Simplify) complicates the computational procedure and it would be simpler if the properties could be derived within a system that relies on Set-Merge only. In general, multiple dimensions are indeed conceivable in a threedimensional world like ours. We have hierarchy as a vertical relation and Table 4.5 Pair-Merge and Set-Merge

Internal External

Pair-Merge

Set-Merge

Richards (2009) Epstein, Kitahara and Seely (2016)

√ √

Approaches to Adjuncts 85 we have linear order, a horizontal relation. So, we might think about adding a third dimension to get adjuncts. However, since we still need a mapping which includes adjuncts too, this is actually no real argument for a separate dimension. The goal of this sub-chapter has been to point out to the reader that recent debates basically revolve around the issue of labeling. This expressively has an impact on the analyses proposed for adjuncts, but none of the accounts discussed so far is formulated in terms of Simplest Merge. Let us therefore take another look at the status of labels in search of a new direction that is in line with SMT and does not adopt extra mechanisms for adjuncts. Stockwell (2016) argues for labels in syntax18 in terms of efficiency. Since labels are needed at both interfaces, it would be efficient to create them in syntax before the derivation splits into SEM and PHON. However, we need an operation that transfers the output of syntax to the interfaces anyway. Hence, labeling might well take place at Transfer; that is, after syntax via a labeling algorithm as Chomsky (2013) suggests it. Labeling need not be part of syntax proper. Seely’s discussion in 2006 leads the way to Simplest Merge in a labelfree syntax and provides clear arguments against labels as being part of Merge. Let us consider his reasoning in more detail. He shows that labels are inert, actually even “non-existent,” because given that they are no terms they cannot participate in derivational c-command (Seely 2006: 187). Considering the operation Merge, he observes that the label is not produced by the structure-building procedure itself but created by two additional steps (see Seely 2006: 185). This is shown in (73). (73) Merge including labels is a complication:

i. It creates {A, B}. ii. It makes a copy of A. iii. It creates a labeled set {A {A, B}}.

Set formation, copying and labeling turns Merge into a very complex operation. Step i and ii should be eliminated in favor of a simple structurebuilding conception of Merge. Consequently, Seely convincingly argues as follows in (74). (74) Rather, what is actually eliminated are two sub-operations of the ‘complex’ operation Merge. It is a consequence of adopting the “simplest” version of Merge [{A, B}] [. . .] that there are no phrasal labels nor projections; i.e. it is a consequence of the simplification of Merge. (Seely 2006: 193)

86  Approaches to Adjuncts Why assume labels in the first place if they complicate syntax and cannot be used in syntax? Looking inside the set and finding the information for labeling must be possible but is only needed after syntax. If the information needed is encoded inside the set, to put it into labels in the syntax would be redundant according to Seely. He also stresses rightly that labels are “ ‘look back’ mechanisms which gives us the crucial information about what happened earlier” (Seely 2006: 207). One could add here that labels would also imply some kind of look ahead, because they only serve interpretation that takes place after structure-building. It does not come as a surprise, then, that labels are “representational” in nature, since both SEM and PHON interpretation have to consider the complete derivation. This also fits Seely’s observation that labels show parallels to moved elements and their copies, which both provide interpretational pieces of information at the SEM and PHON interfaces. For those reasons, let us suppose instead that Set-Merge forms unlabeled sets, which also contain “adjuncts.” Adjuncts can be complex sets themselves and can be merged with other sets. Labels are not part of syntax but play a role in the interpretation of those complex sets that are the output of Merge. Consequently, we arrive at the picture we present in (75). (75) Transfer

i. From Syntax  to the interfaces ii. From non-labeled sets  to labeled sets

What exactly happens at Transfer? Chomsky’s most recent labeling algorithm might well apply at Transfer and yield labels. Conversely, Richards (2017) argues against the recent labeling algorithm. Basically, he formulates two interesting counter-arguments. First, he stresses that cyclic Transfer derived by phase heads already has the effect of successive cyclic movement and intermediate steps. Consequently, deriving it from labeling too would be redundant. Note that the LA does not predict the exact intermediate landing site. Phases predict those to be at the edge of the phase head. Second, he states convincingly, if two opposing views yield the same result, this is quite unsatisfactory. On the topic of islands, scholars are divided into those who assume that the lack of a label produces an island effect (compare Hornstein & Nunes 2008, see also Goto 2016) and scholars who argue the opposite and say that it is the label that causes opacity and freezing effects (see Fukui & Narita 2017 arguing for a symmetry-driven syntax where phrases sharing features are stable). For those reasons, Richards (2017) suggests an alternative in his position paper. Under his view, neither the lack of a label, nor the existence of a label, but a doubly labeled syntactic object would lead to an anomalous

Approaches to Adjuncts 87 interpretation. Furthermore, he does not assume labeling by means of a specific LA but labeling by phase heads. So, it would be quite possible to reduce all effects to phases. Thus, he proposes a “two-label effect” for the sets in (76). (76) {XP, YP} both phases = islands

i. ii. iii. iv.

Subject {DP, vP} Adjunct {vP, CP} Free relative islands {DP, CP} NPC {nP, CP}

This looks like a very simple alternative to the LA and it seems to nicely predict the island behavior of phrases. Consider the adjunct case first. What exactly predicts that it is the CP in (76) iv. that is turned into an island? Why not the other phase, the vP? The same applies to the other cases. So, why is the DP phase in (76) i. an island, but not the vP phase? Unless derived from some independent fact, this solution looks a bit ad hoc. Moreover, how can we interpret two phases—say, a {DP, vP} set—at all, if they are anomalous in interpretation? Richards argues that non-uniform composite labels cause confusion at the interfaces and ultimately lead to a deviant interpretation (Richards 2017: 6). It is quite obvious that this seemingly simple system has to be supplemented with further assumptions. Richards alludes to the possibility of active versus non-active workspaces in this connection (Richards 2017: 7). The island phase would be inactive while the workspace of the other phase, that is, the one the island phase is merged to, would still have to be open. Nevertheless, the stipulation of active and non-active workspaces adds to the complexity of the overall system. Dealing with different workspaces does not answer the question of which workspace is already closed and which one is still left open. Moreover, distinct workspaces resemble the different dimensions again. Also, if only labeled syntactic objects contribute to the meaning at the interfaces, as we argued before, one might wonder how exactly the interpretation of phase internal material can be accommodated. Richards explicitly denies sub-phasal material an interpretive impact. Consequently, one would have to rethink Full Interpretation, which normally requires any symbol to enter the rules of interpretation. So, why are ONLY phases labeled? According to the cases in (76), DP, CP, nP, and vP are phases. What status do prepositional phrases have? Since they form islands (at least in adjunct positions) they must count as phases too. What about adjectival phrases and adverbial expressions? Do they also compose phases? Note that irrespective of islands, the difference between concrete labels might also play a role in the respective interpretation. See for example the data in (77).

88  Approaches to Adjuncts (77) a. b. c. d.

He drove home [quite nervous] He drove home [quite nervously] He met the woman [quite nervous] He met the woman [quite nervously]

The respective attachment may differ too, of course, but the label differences ((77) a. and c. vs. (77) b. and d.) seem to make interpretive distinctions too, since the secondary predicate reading is only available with an AP, and the event modifying reading is only possible with the adverbs. Hence, it would be important to clarify the concept of a phase. However, it is worth observing whether and how Richards will elaborate on his alternative account in the future. For the time being, we will limit ourselves to pointing out that he states some interesting ideas on phases and labels in his position paper. Where are we now? Basically, we want to keep to a couple of assumptions in consonance with SMT, repeated here in (78) for convenience. (78) SMT-Based Assumptions





i. UG must be simple in order to have evolved. Therefore, it must capture the basic property of human language, that is to say, hierarchical structure: Simplest Merge, unordered set formation, fulfills this task nicely. ii. Interpretation is restricted to the interfaces SEM and PHON. Since labeling or identification of a syntactic object is interpretive, labeling is to be restricted to the interfaces: labels are no syntactic objects, i.e., they are neither output of a syntactic operation, nor are they manipulated by a syntactic operation. iii. Third factor-considerations should apply: the computation must be efficient.

Considering the architecture of language, we have observed that we minimally need an inventory of linguistic expressions the syntax may operate on by means of Merge. Merge is not deterministic, it does not need a purpose or a trigger to apply, and it is blind to content. Content or interpretation solely takes place at the respective interfaces PHON and SEM. The link between Syntax and the interfaces is Transfer. Figure 4.4 provides a simple overview. So, what is Transfer? Crucially, third-factor considerations imply that, once you reach the interfaces, you cannot go back to syntax. Hence, once transferred, a syntactic object becomes inaccessible to further syntactic operation. It is also part of an efficient computation to transfer syntactic objects in chunks or phases. Once transferred, those objects are invisible to syntax proper. Viewed this way, Transfer sets an endpoint to syntax and a starting point for interpretation.

Approaches to Adjuncts 89

LEXICON: inventory of items SELECT

INPUT to SYNTAX SYNTAX:

Merge (External and Internal) TRANSFER

OUTPUT of SYNTAX

INTERFACES SEM & PHON: Interpreta on Figure 4.4 Overview of the Architecture

Ishii (2016) suggests subsuming Transfer under Merge, as in (79). (79) Transfer = Self Pair-Merge He argues that transferred domains remain in the workspace, but they are made invisible and inaccessible to the syntactic computation through SelfPair-Merge. Viewed this way, transferred domains are sent from a ‘primary plane’ to an opaque ‘separate plane’ (adjunct plane) (Ishii 2016: 159). Transfer under his conception would thus be conceived of, as in (80) below. (80)

i. [XP YP [X’ X ZP]] → Transfer → [XP YP [X’ X ]] ii. Simplify → X {ZP, ZP} → X {ZP}

Although this is an interesting idea, it still has the cost of assuming PairMerge as an operation besides Set-Merge. Also, we would again need

90  Approaches to Adjuncts Simplify to apply at the interfaces. Since Self-Pair-Merge results in an identical pair, application of Simplify would ultimately lead to a singleton set. Recall that Jakielaszek (2017) suggests an analysis of adjuncts in terms of Internal Self-Merge, which involves copies. The question arises as to whether Self-Pair-Merge is actually the same, since self-Merging necessarily involves identical elements, hence copies. The XP being self-merged is already constructed. Also, if the elements are identical copies, in what sense could they be ordered at all by Pair-Merge? However, we can draw the conclusion that there seems to be some connection between adjuncts and Transfer then because those parallels in the analyses cannot be a mere coincidence. A reduction is a welcome result, but if we, in fact, reduce Transfer to a version of Merge, as Ishii suggests, this would imply that there is structure-building taking place after syntax because Merge (in any version) is structure-building. The parallels between adjuncts (islands) and transferred elements (inaccessible to syntax) are nevertheless worth noticing. Nunes and Uriagereka (2000) building on the Multiple Spell-Out hypothesis developed in Uriagereka (1999) account for the island behavior of adjuncts by explaining the inaccessibility of the internal material of adjuncts in terms of Spell-Out. The label remains for further deterministic computation, but the internal parts of adjuncts (and subjects too) must be spelled out in order to get properly linearized in consonance with Kayne’s LCA. The lack of c-command enforces Spell-Out. Note that this analysis accounts for the island property, but the authors assume syntactic labeling, and the computation is further complicated by the availability of sideward movement. In sum, the fruitful debates addressed in this sub-chapter demonstrate that focusing on labeling and Transfer may lead to new insights. The exciting side of recent developments consists in the possibility of minimizing and simplifying the computation. If it would be possible to include adjuncts properly in the new system, this step would, in fact, be progress within the most recent minimalist approach. The task of the next chapter is to investigate this possibility further and propose a new analysis in line with SMT.

Notes 1. The NTC in the sense of the Law of Conservation of Relations (LCR) (see Epstein, Kitahara & Seely 2012: 256) would normally block any form of counter-cyclic Merge, which changes the syntactic objects already formed. Late-Merge actually represents a form of counter-cyclic Merge. 2. If Late-Merge exists, one would have to allow it to apply recursively too, which would lead to a massive complication. Moreover, the question arises of whether it would be possible to rule out Internal Late-Merge as a further natural option, which would basically rule in island violations. See Sportiche (2019a, 2019b) for details. 3. There are scholars who argue against differences between adjuncts and arguments with respect to reconstruction (cf. Georgi, Salzmann & Wierzba 2018

Approaches to Adjuncts 91 and the works cited therein). If reconstruction takes place in any case, and all copies remain in place, this would provide further evidence against Late-Merge. 4. The analysis of Takahashi and Hulsey (2009), based on Late-Merge of a complement known as “wholesale Late-Merge,” will not be discussed here. What is relevant to our discussion is that Late-Merge of the NP complement of NP is conditioned in this case. However, a conditioning or restriction on Merge goes against the conception of free Merge anyway. The same basically applies to Stepanov’s label inspection which is clearly not compatible with free Merge either. 5. A reviewer points out that there are languages (the reviewer mentions Finnish and Korean) in which adjectives, relative clauses, various PPs, and some manner adverbs are case-marked. He also objects that reconstruction is sometimes possible, and adjuncts are not always islands. First, the properties under (8) are an approximation. Second, modifying elements that are casemarked also exist in German, embedded into an argument-DP. They are not independently case-marked but agree with the D-N, and we probably deal with agreement at PHON. Third, we agree that the reconstruction data are tricky; still, as far as wh-movement is concerned, there is an observable distinction between arguments and adjuncts. As for islands, this issue will be addressed in Chapter Five. 6. Chomsky (1995: 244) discusses three options for label construction. If projection takes place when X and Y merge, the intersection or the union of X and Y, or properties of either X or Y, may label the product. 7. Given BPS, projection is taken to be a relational concept. A projection is maximal if it no longer projects. Hence, adjuncts cannot preserve a (bar-) level information. 8. Moreover, any (unmoved) adjunct would syntactically be ambiguous between concatenated and concatenated plus labeled. Since syntactic outputs get interpreted at SEM, one would, however, expect this ambiguity to correlate with some distinction in the respective interpretation. Still, there seems to be just a single interpretation available for a sentence like John met Mary at the party. 9. This would actually capture what Hornstein (2009: 109) calls the “double personality” of adjuncts, namely, that they seem to be simultaneously inside and outside a given phrase. Consequently, they can participate in VP-ellipsis or be stranded. 10. However, it is worth noticing that the idea of extended projections dating back to Grimshaw (1991 [2005]) is in fact considered by Bowers (2018), who elaborates on it as bottom-up selection yielding clausal and nominal projections in his approach. As noted previously, Bowers (2001, 2018) also analyzes adjuncts as selecting their “host.” Thus, the idea of an extended projection would presumably be applicable to Zeijlstra’s approach. Adverbs and prepositional phrases with the V uV specification would thus be able to select aspectual verbal projections, too. In (i), one must argue that recently bearing the features V uV uPerf selects perfective have composed of the features V Perf uV. Have in turn would select the verbal participle. Note, however, that under this view, the need for more features, or more sub-features would arise, rendering the system of triggered Merge more complicated. i.

a. b. c. d. e.

He has been reading the book recently. [V] [uV] [Perf] [uT] (has) [V] [u Perf] [uV] [Prog] (been) [V] [uProg] (reading) [V] [uV] [uPerf] (recently)

92  Approaches to Adjuncts It must also be guaranteed that the verbal inflections are expressed on the correct items (at PHON). For example, as can be seen in i c. and e. the adverb has the same features as the auxiliary be (with the exception of the assumed [Prog] (sub-)feature), still, it is the auxiliary that will be inflected for the participle feature in the end and not the adverb. We might argue that since, according to Zeijlstra, -ly values the Pred of the root recent as V, there could be no second affix due to a ban on two endings, but this does not help in cases where the adverb lacks -ly, for instance, fast. 11. In a similar vein, Panagiotidis (2015: 116) assumes that there is a functional head, a supercategory called [perspective] that categorizes roots. Zeijlstra (2017: 319) differs in that he abandons the categorizer and assigns the supercategory, contentive, now [pred], directly to roots. 12. The necessity of additional features becomes evident if we look at the following examples:

i. John numbered the examples. ii. the number of cookies

Since Zeijlstra does not assume a categorizer, but assigns the Pred feature to roots directly, number has to be valued as Pred:V in i. and Pred:N in ii. Consequently, T must have a V feature, and D an N feature that values the respective root. Now, we also need to add the respective selectional features of T and D which are needed in order to guarantee that Merge has a purpose. Does T have uV, V and D uN, V? Note that uV, V are the same features adjuncts shall be assigned. If T also selects a subject-DP, it will have the same features as a preposition, that is, V, uV, uD—the only difference being tense. 13. It is worth mentioning that Zeijlstra (2019) starts his reasoning by pointing out that the first LA (Chomsky 1995, 2000) would rest on an unmotivated assumption that the selector projects. His analysis would actually give a reason why this seems to be the case. Though matching under sisterhood and percolation of features result in the desired labeling, it has to be stressed that he actually only substitutes the “selector projects” assumption by two other stipulations. The first one is that Merge must always have a purpose (his rule 2), and the second one is that there is a specific, obligatory percolation mechanism (his rule 1). Uninterpretable features always percolate up until they find a matching feature in a sister, whereas interpretable ones would never percolate beyond their maximal projection. Note also that the latter part of the percolation mechanism does not define projection as a relative aspect of structure-building as is normally the case since the beginning of minimalism and BPS. 14. Arguing for a PRO as the subject of the infinitival adjunct mediating the binding options so that an adjunct internal solution is possible will not help, because PRO itself needs to be interpreted. In (59) a.–b., PRO refers back to the subject of the matrix clause, and in (59) c.–d., it is controlled by the object of the main clause. Both phrases occur outside the adjunct. 15. Gallego’s proposal implies that adjuncts have to be traced back to two quite distinct situations:

i. unlabeled syntactic objects ii. double, —triple, etc., peaks created by means of Parallel Merge

16. Endocentricity shall, in fact, be taken as a pure interface phenomenon based on Minimal Head Detection. See Narita (2014: 191ff) for details. 17. Daniel Seely (p.c.) notes that Pair-Merge is a potentially serious problem for the evolutionary argument about Merge, since two types of Merge significantly weaken the evolutionary argument.

Approaches to Adjuncts 93 18. Stockwell’s account builds on the idea that children acquire labeling in syntax by means of parallel derivations, leading the way to crash-proof grammar. Taking optional labeling as a preliminary hypothesis, the child exposed to a set of (AP, NP), for instance, would assume that the label could either be AP or NP, but, since the derivation goes on by merging a determiner, the label leading to an ultimately crash-proof result, would be NP.

References Alexiadou, A. 2004. Adverbs across frameworks. Lingua 114: 677–682. Biskup, P. 2009. Adjunction, Phase Interpretation, and Condition C. In Grohmann, K. K. (ed.), InterPhases: Phase Theoretic Investigations of Linguistic Interfaces, 114–148. Oxford: Oxford University Press. Bowers, J. 2001. Syntactic Relations. Manuscript, Department of Linguistics, Cornell University. Bowers, J. 2018. Deriving Syntactic Relations. Cambridge: Cambridge University Press. Chomsky, N. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, N. 2000. Minimalist Inquiries: The Framework. In Martin, R., Michaels, D. and Uriagereka, J. (eds.), Step by Step: Essays in Minimalist Syntax in Honor of Howard Lasnik. 89–155. Cambridge, MA: MIT Press. Chomsky, N. 2004. Beyond Explanatory Adequacy. In Belletti, A. (ed.), Structures and Beyond: The Cartography of Syntactic Structures, Vol. 3, 104–131. Oxford: Oxford University Press. Chomsky, N. 2008. On Phases. In Freidin, R., Otero, C. and Zubizaretta, M. L. (eds.), Foundational Issues in Linguistic Theory: Essays in Honor of JeanRoger Vergnaud, 133–166. Cambridge, MA: MIT Press. Chomsky, N. 2013. Problems of Projection. Lingua 130: 33–49. Chomsky, N. 2015. Problems of Projections: Extensions. In di Domenico, E., Haman, C. and Matteini, S. (eds.), Structures, Strategies and Beyond: Studies in Honor of Adriana Belletti, 3–16. Amsterdam and Philadelphia: John Benjamins. Collins, C. 2002. Eliminating Labels. In Epstein, D. S. and Seely, T. D. (eds.), Derivation and Explanation in the Minimalist Program, 42–64. Oxford: Blackwell. Cormack, A. and Smith, N. 2005. Linearization: Adjuncts and Arguments. UCL Working Papers in Linguistics 17: 111–129. Epstein, S. D. 1999. Un-Principled Syntax: The Derivation of Syntactic Relations. In Epstein, S. D. and Hornstein, N. (eds.), Working Minimalism, 327–345. Cambridge, MA: MIT Press. Epstein, S. D., Kitahara, H. and Seely, T. D. 2012. Structure Building That Can’t Be. In Uribe-Etxebarria, M. and Valmala, V. (eds.), Ways of Structure Building, 253–270. Oxford: Oxford University Press. Epstein, S. D., Kitahara, H. and Seely, T. D. 2016. Phase Cancellation by External Pair-Merge of Heads. The Linguistic Review 33: 87–102. Epstein, S. D., Kitahara, H. and Seely, T. D. 2017. Merge, Labeling and their Interactions. In Bauke, L. and Blümel, A. (eds.), Labels and Roots, 17–45. Boston and Berlin: De Gruyter Mouton. Ernst, T. 2004. (sec. ed.). The Syntax of Adjuncts. Cambridge: Cambridge University Press. Ernst, T. 2014. The Syntax of Adverbs. In Carnie, A., Sato, Y. and Siddiqi, D. (eds.), The Routledge Handbook of Syntax, 108–130. London and New York: Routledge.

94  Approaches to Adjuncts Fukui, N. and Narita, H. 2017. Merge and (A)symmetry. In Fukui, N. (ed.), Merge in the Mind, 9–34. London and New York: Routledge. Gallego, A. 2010. An L-Syntax for Adjuncts. In Duguine, M., Huidobro, S. and Madariaga, N. (eds.), Argument Structure and Syntactic Relations—A CrossLinguistic Perspective, 183–202. Amsterdam: John Benjamins. Georgi, D., Salzmann, M. and Wierzba, M. 2018. Reconstruction in German A’Movement. An Experimental Investigation. Paper presented at the CGSW 33, Göttingen, September 28. Goto, N. 2016. Labelability = Extractability: Its Theoretical Implications for the Free-Merge Hypothesis. In Proceedings of the 46th Annual Meeting of the North East Linguistic Society, 335–348. Amherst, MA: GLSA, University of Massachusetts. Grimshaw, J. 1991 (2005). Extended Projections. In Grimshaw, J. (ed.), Words and Structure. CSLI Lecture Notes, Vol. 151, 1–73. Stanford, CA: CSLI Publications. Hornstein, N. 2009. A Theory of Syntax—Minimal Operations and Universal Grammar. Cambridge: Cambridge University Press. Hornstein, N. and Nunes, J. 2008. Adjunction, Labeling, and Bare Phrase Structure. Biolinguistics 2: 57–86. Hunter, T. 2015. Deconstructing Merge and Move to Make Room for Adjunction. Syntax 18: 266–319. Irurtzun, A. and Gallego, A. 2007. Consequences of Pair-Merge at the Interfaces. ASJU XLI-2: 179–200. Ishii, T. 2016. Transfer as Self Pair-Merge. School of Arts and Letters, Meiji University, Bungei Kenkyuu (Studies in Literature) 129: 5–33. Jakielaszek, J. 2017. A Minimalist View on the Syntax-Semantics Relationship— Turning the Mind into a Snowflake. Frankfurt am Main: Peter Lang GmbH. Lebeaux, D. 1991. Relative Clauses, Licensing and the Nature of the Derivation. In Rothstein, S. (ed.), Perspectives on Phrase Structure. Syntax and Semantics, Vol. 25, 209–239. New York: Academic Press. Nakashima, T. 2017. On Strict Cyclicity and Label: Toward Elimination of Late Merge. Explorations in English Linguistics (EEL) 31. Narita, H. 2012. Phase Cycles in Service of Projection-free Syntax. In Gallego, A. J. (ed.), Phases—Developing the Framework, 125–172. Boston and Berlin: Mouton de Gruyter. Narita, H. 2014. Endocentric Structuring of Projection-Free Syntax. Amsterdam and Philadelphia: John Benjamins. Nunes, J. and Uriagereka, J. 2000. Cyclicity and Extraction Domains. Syntax 3: 20–43. Oseki, Y. 2015. Eliminating Pair-Merge. In Steindle, U., Borer, T., Fang, H., Pardo, A. G. Guekguezian, P., Hsu, B., O’Hara, C. and Ouyang, I. C. (eds.), Proceedings of the 32nd West Coast Conference on Formal Linguistics, 303– 312. Somerville, MA: Cascadilla Proceedings Project. Panagiotidis, P. 2015. Categorial Features—A Generative Theory of Word Class Categories. Cambridge: Cambridge University Press. Richards, M. 2009. Internal Pair-Merge: The missing Mode of Movement. Catalan Journal of Linguistics 8: 55–73. Richards, M. 2017. Problems of ‘Problems of Projection’: Breaking a Conceptual Tie. Paper presented at the Generative Syntax 2017: Questions, Crossroads, and Challenges workshop, Barcelona, June 23. To appear in Catalan Journal of Linguistics, July 2019.

Approaches to Adjuncts 95 Rubin, E. 2003. Determining Pair-Merge. Linguistic Inquiry 34: 660–668. Saito, S. 2017. The Relation Between Late Merge and Phases. Explorations in English Linguistics (EEL) 31. Seely, T. D. 2006. Merge, Derivational C-Command and Subcategorization in a Label-Free Syntax. In Boeckx, C. (ed.), Minimalist Essays, 182–217. Amsterdam and Philadelphia: John Benjamins [reprinted in Epstein, S. D., Kitahara, H. and Seely, T. D. 2015. Explorations in Maximizing Syntactic Minimiziation— Routledge Leading Linguists, 116–154. London and New York: Routledge]. Sportiche, D. 2019a. Neglect (or Doing Away with Late Merger and Countercyclicity). Available at Lingbuzz/002775. Sportiche, D. 2019b. Somber Prospects for Late Merger. Linguistic Inquiry 50: 416–424. Stepanov, A. 2000. The Timing of Adjunction. In NELS 30, 597–611. Amherst: Mass GLSA. Stepanov, A. 2001. Late Adjunction and Minimalist Phrase Structure. Syntax 4: 94–125. Stockwell, R. 2016. Labelling in Syntax. COPIL-Cambridge Occasional Papers in Linguistics 9: 130–155. Takahashi, S. and Hulsey, S. 2009. Wholesale Late Merger: Beyond the A/Ā Distinction. Linguistic Inquiry 40: 387–426. Truswell, R. 2007a. Extraction from Adjuncts and the Structure of Events. Lingua 117: 1355–1377. Truswell, R. 2007b. Tense, Events, and the Extraction from Adjuncts. Proceedings from the Annual Meeting of the Chicago Linguistic Society (CLS) 43: 233–247. Truswell, R. 2011. Event Structure and A’-Locality. In Folli, R. and Ulbrich, C. (eds.), Interfaces in Linguistics—New Research Perspectives, 17–24. Oxford: Oxford University Press. Uriagereka, J. 1999. Multiple Spell-Out. In Epstein, S. D. and Hornstein, N. (eds.), Working Minimalism, 251–282. Cambridge: MIT Press. Zeijlstra, H. 2017. Contentives: A Lexical Supercategory Above Nouns and Verbs. In Mayr, C. and Williams, E. (eds.), Wiener Linguistische Gazette (WLG) 82. Festschrift für Martin Prinzhorn, 313–321. Wien: Universität Wien, Institut für Sprachwissenschaft. Zeijlstra, H. 2019. Labeling. Selection and Feature Checking. To appear in Smith, P., Mursell, J. and Hartmann, K. (eds.), Agree to Agree: Agreement in the Minimalist Programme, 137–174. Berlin: Language Science Press.

5 A New Proposal Labeling Is Transfer

5.1 Adjuncts and Labeling The proposal we will advance in this chapter is based on a minimalist syntax couched in terms of Simplest Merge. The computational system shall be maximally simple, as laid out in Chapter Three, and this necessarily implies that there must not be any specific device for adjuncts. Upon careful inspection of the existing accounts in Chapter Four, we are led to conclude that up to this point, there is no approach to adjuncts satisfying the analytical requirement in (1): (1) Adjuncts are syntactic objects built in the same dimension as any other syntactic object by means of a syntactic computation that is un-triggered; Simplest Merge is the only syntactic operation available applying freely to build syntactic objects in form of unlabeled sets. As we have seen in the previous chapter, attempts to simplify the computation starting with Chomsky (2013) focus on labeling and Transfer. In light of this observation, it seems fair to say that these operations deserve a closer look. Labels are, roughly speaking, assigned to syntactic objects independently of the syntactic operation Merge, while Transfer hands the syntactic objects over to the interpretive components. Viewed this way, there is no direct connection between labeling and Transfer. In fact, we will see that an alternative conception is conceivable, which helps to further simplify UG and the design of the language faculty. What is the role or function of a label to begin with? A non-linguistic example may help to clarify. Think about a bookshelf with books on it, each having a specific title—a label, so to speak. What is the function of this “label?” Basically, it has two functions. First, the title of a book gives some hints about its content, namely, what the book is about and what kind of book it is to begin with (a scientific investigation in a sub-field of linguistics, in mathematics, or in physics; a novel; a drama; a collection

A New Proposal 97 of short stories; and so on). Second, the title helps to put an order to the books on the shelf. Each one needs a concrete place. This dual function of “labels,” namely, pointing to the content on the one hand and to linear order on the other hand is exactly what we observe with labels in linguistics too. Labels have a function at SEM, pointing to asymmetric relations such as predicate/argument or operator/scope. For instance, to identify predicates and arguments properly, the labels VP and DP are needed in a sentence such as The man plays the guitar virtuously. At PHON, labels serve to regulate the asymmetric relations of precedence and prosodic rules. One example would be the respective order of V and DP in VP with V preceding or following the DP (compare English versus German, respectively). Another example is given by the stress difference visible in permit depending on a verbal or a nominal label. Identifying the syntactic object is part of the interpretive rules. A transferred syntactic object must have a label in order to be interpretable. This echoes the discussion we held in Chapter Three, where we observed that labels should not be part of the syntax (labeling ≠ projection ≠ part of syntax), but are only needed later for interpretation (labeling = Minimal Search = part of Transfer). By logical reasoning, then, we can propose (2) to hold and, in fact, to simplify the system further by reducing to (2) iv. (2)

i. Transfer makes objects inaccessible to syntactic operations. ii. Labels are not syntactic objects. Hence labels are inaccessible for syntax. iii. Labeling makes objects inaccessible to syntactic operations. iv. Labeling = Transfer It makes sense to say that syntax cannot have access to elements that are interpretive in nature. Recall too the insights of Seely (2006), who we discussed in the previous chapter. Merge only creates sets but no labels. Labels are syntactically inert, even non-existent in the computation. Notice also that the computational procedure in its simplest conception is no longer label/feature-driven. Structure-building (External Merge) and movement (Internal Merge), under this view, are necessarily completely free. Consequently, if we assume that there are labels at the interfaces that serve as instructions, or which are elements that interpretative rules act on, then we actually should draw the conclusion that labeling is Transfer, because syntax has no access to labels. (Cyclic) labeling yields fully labeled objects at the interfaces that can be semantically interpreted and linearized, as well as pronounced at PHON. If labeling is Transfer, the architecture of the grammar is crucially simplified. Instead of two rather unconnected operations, there is only one operation serving the purpose of translating unlabeled sets (syntactic objects) into labeled sets (interface objects).

98  A New Proposal Note that this suggestion sounds similar in a sense to a route taken by Takita, Goto and Shibata (2016), who propose an analysis in terms of Spell-Out. Arguing against the invisibility of lower copies in Chomsky’s LA of 2013, they assume that labeling through Spell-Out is mediated through a phase head in cases of {XP, YP} so that, in a {wh-DP, CP} set for instance, the complement of the phase head C is spelled out and Minimal Search can find the head C in the remaining set {wh-DP C} to label it. However, this solution of labeling through Spell-Out would not help us with adjuncts, unless one would stipulate that all adjuncts are complements of a phase head. With this in mind, let us re-examine the latest labeling algorithm by Chomsky to see whether we might revise it accordingly to deal with adjuncts as well. The LA based on Minimal Search must deal with conflicts {XP, YP}. Given a complex set, XP, being merged with a complex set, YP, the result needs to be labeled to be identified at the interfaces. Recall that in order to unambiguously label a syntactic object, Minimal Search must find a single head. In {XP, YP} cases, it finds two heads, X and Y, which leads to the labeling conflict. Chomsky (2013, 2015) proposes two strategies to resolve the labeling conflict in the set {XP YP}, repeated here for convenience in (3). (3)

i. Feature Sharing if XP and YP have a feature in common ii. Internal Merge of one of the two, XP or YP

As we have mentioned already, adjuncts being merged with another phrasal object cause a labeling conflict. This is evident in the relevant {XP, YP} cases shown in (4) for adjuncts. We also add a related semantic aspect to (4) a.–c. in order to show that the respective label Minimal Search needs to detect serves the interpretation of the set in question. (4) a. {NP, PP} entity—property/modifier b. {vP, PP} event—predicate/modifier c. {TP, PP} proposition—predicate/modifier The first strategy of disambiguation under (3) cannot be used in the case of adjuncts, because there is no obvious feature in (4) shared between the two phrases. Consider the concrete examples in (5). (5) a. {the man} {in the black shirt} b. {read the book} {in the garden} c. {they will meet her} {tomorrow at 10 a.m.}

A New Proposal 99 Resorting to some kind of predicate feature in both phrases is not really convincing, because being a predicate is basically a function or a relation that a phrase can enter in a given structure but not a feature carried by a lexical item. Also, it is already a part of the interpretation. Inventing additional features that {XP, YP} might share seems to be a mere ad hoc stipulation. Sharing a prominent feature as it is the case with subject–predicate relations and wh-movement in (6), where we can indeed observe that agreement overtly expressed simply does not hold for adjuncts and their host. (6) a. DPφ TPφ: subject—predicate: The man likes/*like Mary. b. XPQ CPQ: wh-OP—proposition: Which man/*The man does she like? The second strategy of (3) is applicable in cases like (7), where a full adjunct undergoes Internal Merge leaving a copy indicated by the angled brackets behind. (7) a. In what inning did they lose the ball game b. In the first inning they lost the ball game c. . . . {set {lose the ball game} } With the lower copy, , being invisible to Minimal Search as Chomsky assumes, the set can be labeled by the head of {lose the ball game}. However, the question remains how to deal with “stable” {XP, YP} cases which are unlabeled under the most recent LA. We observed that syntax creates structures by symmetric set formation. Any asymmetry between SOs is interpretive in nature and must be part of the interfaces. Ignoring for the time being the feature-shared, symmetric label, labels assign asymmetry to a set. Recall that we proposed the simplification in (2) that labeling is Transfer. How can the label be determined when the adjunct does not move? If we are on the right track and labeling is Transfer, we can label an adjunct upon Merge and it is transferred; hence, automatically invisible to syntax by its label. Assigning a label would thus render the syntactic object invisible to the syntax and at same time visible to the interface. (8) a. Merge {lose the ball game} {in the first inning} → {set {lose the ball game} {in the first inning}} b. Label adjunct with PP → {set {lose the ball game} {PP in the first inning}}

100  A New Proposal The strikethrough shall indicate that the label PP renders the set inaccessible to further operations of Merge. The oft-cited island property of adjuncts follows automatically now as well just by labeling the adjunct. Since the adjunct-set is labeled and thereby no longer accessible to syntactic operations, it is only accessible to the interfaces now. Internal Merge cannot apply to anything inside the PP, which accounts for the island behavior of adjuncts illustrated again in (9). (9) *What inning did they lose the ball game [PP in ]? The PP adjunct is labeled (= transferred) to the interfaces and this makes the DP embedded inside invisible to the syntax. Extraction from a labeled adjunct would thus be ruled out. We will elaborate on this later. Suffice to say here that the example in (9) will not be interpretable and is predicted to be as ungrammatical as a sentence as *What inning did they lose the ball game in what inning? As a first approximation, we extend the labeling algorithm to include this third way shown in (10) iii. as a further option to solve the labeling conflict arising upon merging XP YP. (10) LA & Conflict Resolving Strategies

i. Internal Merge: (XP, YP)  YP . . . [XP, ] label = XP ii. Feature Sharing: (XP, YP)  [XP[F] YP[F]] label = [F] iii. Labeling (= Transfer): (XP, YP)  (XP, YP) label = XP

Either an adjunct is internally merged (leaving an invisible copy) or labeled (made invisible to syntax) after being externally merged. In both cases, the remaining XP serves as the label for the full set in the end. Consider (11) as an illustration with the respective labels eventually assigned. (11) a. John will [vP [vP read the book] [pp in the garden]] (= (10) iii) b. [In which garden] will John [vP [vP read the book][PP in which garden]] (= (10) i) The interpretation in the position in bold at SEM would be the same. In both instances, the adjunct is invisible for Minimal Search at some stage of the derivation, because of labeling after being externally merged in the first case and due to Internal Merge in the second case. The interfaces differ only in that in (11) b., there are copies of the PP in the full set(s) of the sentence. If labeling marks the stage at which syntactic objects are no longer accessible to syntax, this means that only non-labeled objects are accessible to further Merge operations in syntax. In (11), we label (transfer)

A New Proposal 101 PP before we label vP. The vP must still be accessible to further computation because, with the next Merge operation, a tense element gets introduced into the derivation. Suppose we would label vP first—rendering vP ­invisible—the derivation would go on with the set {set {. . .} {in the garden}} containing the adjunct at this point still unlabeled instead. Merging the T (will) with in the garden in a next step would finally yield gibberish at SEM and would be ruled out there as violating the selection of Tense. This is illustrated in (12). Again, the strikethrough is indicating that labeling is Transfer. (12) T {? {vP read the book} {in which garden}}}: Label of ? = PP → SEM *T selecting PP The basic idea is to make syntactic objects inaccessible and thus unavailable to syntactic operations. We do not mean that labeling/Transfer implies literal removal from the structure since this would yield an NTC/ LCR violation.1 The LA of Chomsky (2013) assumes that contrary to the highest copy, lower copies are invisible to Minimal Search. We suggest an alternative to this view here, basically, because the main idea of copies is that they are a single, discontinuous object. We therefore further argue that all copies get labeled simultaneously rendering them invisible to syntax and distinguishing them from repetitions at the interfaces. That means that identical elements which are labeled (= transferred) at the same time are copies (for example, John was seen ) and not repetitions as in John saw John. It must be stressed, though, that this step has crucial consequences for Chomsky’s (2013) LA because it renders all copies (including the highest one) invisible and inaccessible simultaneously. This step would make all copies equal and remove the distinction between visible and invisible copies. Notice that this can be regarded as a further simplification of the minimal system. Copies are conceived of as a single discontinuous object, not two distinct objects (see Chomsky, Gallego & Ott 2019: 10) and, for this reason, it makes good sense to say that a single object is labeled once and not several times (or in pieces). Also note that both copies are interpreted in a sentence such as Who do you like ?, and must therefore be identified by the label in both positions. If all copies are visible (unlabeled in syntax) or invisible (labeled, hence inaccessible to syntax) at the same time, the conflict solving strategy in (10) ii. can be no option at all, because as soon as the lower copy is inaccessible, the highest is too and as such invisible. Under this view, a prominent shared feature may not label a set. Instead in (13), the heads C[Q] and T[φ] label the respective sets in the end in a rather traditional fashion as shown below. (13) [CP [DP which book] [C’ C did [TP [the man] [TT [DP the man] [V’ v read [DP which book]]]]]]

102  A New Proposal With all copies being likewise invisible, only C, T and v remain as the available labelers. Thus, labeling would always be asymmetric. In syntax, only symmetry plays a role which accords well with the fact that Simplest Merge yields only symmetric sets capturing the basic property of language (structure-dependency). Asymmetry does not play any role in the syntax but is needed only later at the interfaces—as pointed out earlier—to get a proper interpretation of the sets formed by Merge. Notice that this further simplifies the perspective on the computation which only constructs symmetric sets. There is no asymmetry in syntax. Furthermore, Minimal Search never looks below head level to search for features within members of sets. The general picture given in (14) emerges. (14) • •

Syntax = Simplest Merge = symmetric set formation Labeling = Transfer: i. It makes sets inaccessible to syntax. ii. It turns symmetric sets into asymmetric set.

•

Interfaces = SEM and PHON = asymmetric set interpretation

The considerations about copies are rather sketchy so far and need to be investigated further. Other issues to be addressed more closely as we go on are successive cyclic movement and phases in particular. We will come back to these topics but first consider how labeling of an adjunct works in a sentence. The labeling of adjuncts needs to take place immediately after Merge. Let us consider a simple derivation by focusing on the points that cause a labeling conflict. In a sentence such as John met Mary in the garden, several {XP, YP} sets occur, causing a labeling conflict. Take a look at (15) and Figure 5.1, where three conflicts are indicated accordingly.2 (15) Conflict 1: {John} {met Mary} Conflict 2: {John met Mary} {in the garden} Conflict 3: {John} {T John met Mary in the garden} Crucially, all the remaining nodes (except those marked as causing a conflict) can be labeled by Minimal Search finding the first head in {H, XP}. One would have to say more about roots, but since the status of roots is actually a general question that is not particularly related to adjuncts, we will ignore it here. How do we resolve those conflicts and get fully labeled sets at the interfaces which are interpretable? The first conflict and the third conflict can both be resolved upon internal Merge

A New Proposal 103

C ph

conflict 3

n

John T

conflict 2

conflict 1

n

John

in the garden

v ph met n

Mary

Figure 5.1 Conflicts

by labeling the copies, thus making them invisible and inaccessible to syntax. The second conflict resolves when we label the adjunct directly after it has been merged into the structure (thus making the adjunct invisible and inaccessible too). Note that conflict two could alternatively be resolved by internal Merge of the adjunct to C in a sentence like In the garden John met Mary . If we do not resolve the conflicts, the nodes will not be labeled at all and, consequently, not transferred to the interfaces. If we resolve the conflicts, the rest of the nodes also can be labeled (by C, T, v). In essence, Minimal Search finds a single head in every remaining case. Under wh-movement with both copies being labeled at once, for example, immediately after Internal Merge, what remains to be labeled is the matrix clause. This is no longer problematic now because Minimal search finds the head C, and the entire sentence can be labeled (= transferred). There is no need to assume Transfer to be triggered by phase heads.

104  A New Proposal Recall the difficulties previous approaches had encountered in accounting for adjuncts to adjuncts, which can now be explained in the same way. If we have a sentence like John left certainly on purpose, we label the AdvP immediately after Merge (to the PP), then we label the remaining set containing the PP as soon as the full adjunct has been merged to the main clausal spine. We can assume that labeling (Transfer) applies freely, but if nodes cannot be labeled (= not transferred), they will not be interpreted at the interfaces. Alternatively, if we would label (= transfer) the “wrong” XP, say, the sister of an adjunct instead of the adjunct, this would yield an uninterpretable set at the interfaces, because the adjunct would remain accessible to further syntactic operation in the syntax but the “actual host” would not. Basically, we can reduce labeling to the following LA, which is clearly a simplification since Minimal Search only looks for heads now. (16) The Labeling Algorithm {H, XP} → label = H Any conflict {XP, YP} can be resolved by means of applying (17) in the course of the derivation. (17) The labeling conflict {XP, YP} resolves under i. or ii.

i. Label XP or YP immediately after External Merge ii. Label XP or YP immediately after Internal Merge

After labeling one of the two sets, the remaining set serves as the label. The decision as to which of the sets gets labeled (and rendered inaccessible to syntax) is part of the interpretation at SEM. “Choosing the wrong set” yields a violation at the interpretive level. This is further exemplified with the example in (18): (18) John likes the picture on the wall. a. Merge {the picture} {on the wall} → {? {the picture} {on the wall}} b. Labeling one of the sets, XP or YP, resolves the conflict: i.  Labeling/Transfer of {the picture} leaves {on the wall} as a labeler for the set-? violating the selection of like at SEM. ii.  Labeling/Transfer of {on the wall} leaves {the picture} as a labeler for the set-? yielding a proper interpretation at SEM. In sum, labeling conceived of as a means to make symmetric syntactic objects inaccessible to syntax by turning them into asymmetric entities only visible to the interfaces allows us to resolve the conflicts by the strategies shown in (17).

A New Proposal 105 If all copies are invisible simultaneously because they form one single syntactic object, it also follows that any set created by Internal Merge gets the label of the “target” or “host” (to put it in traditional terms). This is particularly desirable in cases where there is no immediately obvious candidate for a prominent feature that, under the former LA, might possibly be shared and label the set. The data in (19) illustrate this situation. (19) a. [[Each part of the system] [C we have to examine carefully ]] (topicalization) b. They said that we should leave the room and [leave the room] [C we finally did Neg > v. At PHON, we only need to require that T is overt in negative expressions in English. Hence, the higher copy of T is realized as does and we get the externalized sentence John does not like books. Since Merge, Internal and External, is always structure-building, and we do not have substitutions as with Move-Alpha in the GB framework,

T DP

T T

v

v Figure 5.11 T-v

v √

DP

A New Proposal 145 we cannot analyze the inversion in Yes/No-questions (and wh-questions) as T-to-C movement. An alternative view in the veins of the analysis sketched here would yield the following externalization for S-A-inversion. T has been internally merged at the edge of the entire set. Then C is externally merged. We have to assume now that at PHON a C [+Q] T-sequence gets pronounced as does in the example at hand. (94) C [+Q] T John T John T v root books → Does John like books? Basically, there must be some kind of T-v-X-“alliance” at PHON similar to an extended projection which allows the pronunciation of the element instantiating the sequence at different positions of the T-copy as a member of the alliance. X stands for the categoryless √root and functional heads such as Perf or Prog (auxiliaries). Take for instance, T-v-Perf (has) or T-v-Prog (is), which both are overtly pronounced higher in Yes/Noquestions since there is a higher copy of T (see (95) and are expressed lower otherwise (compare with (96)). The highest copy of the subject is overtly expressed as usual. (95) C-[+Q] T Is John T John T v Prog John Prog v read the book? C-[+Q] T Has John T John T v Perf John Perf v read the book? (96) C John T is John T v Prog John Prog v read the book. C John T has John T v Perf John Perf v read the book. In essence, this boils down to the assumption that cross-linguistic variation concerns the respective lexical resources and pronunciation statements in terms of T-v-X positions at PHON. Considering other languages, we observe that the respective pronunciation rules differ in the positional respect. In German, for instance, we find V/2 and inversion applies to the T-v-root complex. In the ideal case, this is just a PHON phenomenon and does not relate to the syntactic computation (and SEM). Let us briefly consider this option. Linear orders are subject to cross-linguistic variation and are settled at PHON. Syntax does not care about linear order. In the German, verbal phrase the linear order is generally taken to be OV and T is also head-final. This would yield the following sequence for an embedded sentence like (97). (97) C-dass T Hans T Hans ein Buch √root v T dass Hans ein Buch las that Hans a book read that Hans read a book

146  A New Proposal Under the assumption that the sequence must be overtly expressed in the highest T position in German V/2, we could derive the respective examples as in (98) and (99). A German Yes/No-question is illustrated in (100).  (98) Hans C T las Hans T Hans ein Buch √root v T Hans las ein Buch Hans read the book  (99) Ein Buch C T las Hans T Hans ein Buch √root v T Ein Buch las Hans A book read Hans Hans read a book. (100) C-[+Q] T las Hans T Hans ein Buch √root v T Las Hans ein Buch? Read Hans a book? Did Hans read a book? Comparing English, German and French, we observe that the T-v-X string gets realized in distinct T-copy positions in each case (unpronounced copies of XPs are presented in angled brackets in (101). (101) a. C John T often T-v-√eats apples. b. C Jean T mange souvent T-v-√ des pommes. c. Äpfel C T isst Hans T oft T-v-√ . Note that the syntactic computation would basically be the same in all cases. First, the heads T and v are (externally) merged and, later, T is internally merged to a higher set, yielding the labeled/transferred structure shown in Figure 5.12. In German, T is internally merged again before C is externally merged, and in a last application of Merge, the object Äpfel is internally merged to the entire set. Let us stress that the implications of the ideas on head movement briefly sketched out here need to be critically elaborated and put under severe scrutiny. What we hope to have shown in this sub-chapter is that we can in fact argue that heads, by being syntactic objects, undergo Simplest Merge (External and Internal Merge likewise). Set formation applies freely and labeling as Transfer can be conflict-free for a successful interpretation to take place finally at SEM. The proper externalization of head sequences and linguistic expressions (and copies) in general underlies language

A New Proposal 147

C C

T

subj

T T

v

v

v √

Obj

Figure 5.12 Labeled/Transferred Structure

particular rules which are not part of the syntax but should be captured by phonological rules. Consequently, the opposition of uniformity (syntax, SEM) and diversity (PHON) would hold in the domain of heads too. We can draw the conclusion that neither adjuncts, nor head movement (“adjunction”) require specifically designed mechanisms. Both phenomena can be handled by means of Simplest Merge. The proposal that labeling is Transfer put forward in this chapter results in a very simple design of the language faculty in line with SMT, summarized in Box 5.2.

Box 5.2  Overview of the General System SYN Simplest Merge forms unordered, unlabeled sets 1. {H, H} 2. {H, XP} 3. {XP, YP}

Labeling = Transfer Labeling is needed for interpretation only. The labeling conflicts (in 1. and 3.) have to be resolved after Merge. At SEM, semantic interpretation rules apply to labeled sets. At PHON, language particular morpho-phonological rules apply, and linear order is established.

148  A New Proposal

5.4 A Final Remark Throughout this book, we capitalized on the hypothesis that there is no structure-building mechanism beyond Simplest Merge. It is conceptually attractive to conceive of UG in the most minimal way, providing a principled explanation for the emergence of the human capacity for language. Assuming two formally distinct operations, Set-Merge and Pair-Merge, would complicate the picture drastically. Reducing to Simplest Merge, one can assign uniform properties to the syntax. This implies that modifying phrases (adjuncts) and complex heads (head adjunction) must be revisited in terms of Simplest Merge unless one would analyze the phenomena in question as being located outside the syntax. Consider Figure 5.13, which illustrates the basic situation. Throughout this book, we advocated for an account of adjuncts building on Simplest Merge. Two phrases are merged and form a symmetric set {XP, YP} that needs to be assigned an asymmetric interpretation. This is the gist of our Labeling is Transfer proposal. We extended this to the possibility of having symmetric {X, Y} sets that undergo the same translation operation from symmetry to asymmetry. In order to resolve conflicts, labeling renders one or the other member of the set inaccessible to syntax (and automatically accessible to the interfaces). This yields the maximally simple system, shown in (102). (102) symmetric syntax  Labeling = Transfer  asymmetric interpretation A final example shall illustrate the basic effects. Recall that we deal with sets, though we represent the following derivation in form of a tree diagram (Figure 5.14) for the sake of illustration. (103) Which boy has the man not seen since that party? We have several labeling conflicts here. First, there are {XP, YP} labeling conflicts and second, we find {X, Y} ambiguities. We indicate them in the lists below using the letters we used in the tree. Capital letters signal phrasal conflicts lowercase letters show where head-conflicts arise. The rest of the nodes not raising any conflicts are simply numbered.

Set-Merge: Simplest Merge only:

symmetric (unordered) sets minimized to

Pair-Merge: asymmetric (ordered) sets

Figure 5.13 Minimization of Operations

symmetric (unordered), unlabeled sets

A New Proposal 149 F Wh-XP

6

C(omp)

E DP

5 T

D b

4 v

Perf

3 Neg

C

B

YP(adjunct)

a

v

A

2

1 v

√

Wh-XP

Figure 5.14 Phrasal and Head Conflicts

(104) {XP, YP} conflicts A = {DPsubj lower copy, 2} B = a, A C = {B, YP adjunct} D = {b, 4} E = {DPsubj higher copy, 5} F = {wh-XP, 6} (105) {X, Y} conflicts a = {Perf lower copy, v} b = {T lower copy, v} Notice first that in set 1 we do not deal with a head conflict, because the root is no category and hence invisible to Minimal Search (head detection only regards categories). Consequently, there is no conflict in set 2 either. Minimal Search will find the head (category) v in both sets (1 and 2). The sets can be labeled (= transferred). We label the adjunct after External Merge, which resolves the conflict C and leaves B accessible to the syntax. Labeling the subject copies simultaneously resolves the conflicts A and E. When the derivation has

150  A New Proposal reached the stage F and the wh-copies get labeled at the same time, we can also label all other nodes just by head inspection. There is no conflict that cannot be resolved. Labeling Perf resolves conflicts a and B, labeling T resolves the conflicts b and D. The sets 1–6 get labeled by the heads v, Neg, Perf, T, and C(omp). Recall that labeling is basically free. This means that any order is permitted in principle, but it might yield a crash at the interfaces. Notice also that Merge is free and could join any potential elements in a set. The assembling and the labeling of sets are both free, but the respective results may fail a proper interpretation at the interfaces. Ultimately, at SEM, we get the labeled sets in Figure 5.15. Flattening the sets and inserting concrete material at PHON might look like (106). (106) Which boy CQ the man Tfin vpres Perf not v-en vtrans seen since that party? Pronouncing the highest copies of XPs, which boy and the man, and pronouncing the head-“alliances” CQ-T-Perf at the highest position as has, C Wh-XP C

C T DP

T T

v v

Perf

v

Perf

Neg Neg

v

v

PP(adjunct)

v

v

v

v v

v Figure 5.15  Labeled Result

√

DP

A New Proposal 151 and ven-vtrans-root as seen41 will eventually lead to an externalization of the structure as the sentence Which boy has the man not seen since that party? For strong islands, we assumed in this chapter that, broadly speaking, a higher conflict must not contain a lower one, which would be confirmed by adjunct islands and subject islands in (107). (107) a. *Which party has the man not seen the boy [since ]? b. *Which man has [the father of ] not seen the boy since this party? The attentive reader will have noticed that resolving head conflicts, {X, Y}, would need a closer look. The two possible conflict resolving strategies, namely, labeling immediately after External Merge or later after Internal Merge play a major role in the analysis of {XP, YP} conflicts, but we only applied labeling after Internal Merge to {X, Y} conflicts. A full account of head movement would thus require considering both options for headconflicts and seeing what implications follow for particular languages. Cross-linguistic differences might, in fact, be linked to the application of one or the other strategy. The choice of the respective strategy might also be the source of differences among languages concerning {XP, YP} conflicts. This general idea needs to be investigated further. We just mention it as a possible direction for future research. Another point that calls for careful examination and explanation concerns islands42 and the interaction between {XP, YP} and {X, Y} conflicts. As can be seen in the example above, the sets involving the conflicts B and D show that head conflicts and phrasal conflicts are connected. We leave it here at it is, but stress that this is an effect which has to be explored further. The reader should also keep in mind that there are good reasons to follow a solution based on a free syntax. We summarize them in (108). (108) GENERAL REASONS for a FREE, UN-TRIGGERED SYNTAX



i. Conceptual simplicity of a non-deterministic computation allows for the three general options {X, Y}, {X, YP}, {XP, YP} we encounter. ii. Under triggered syntax, the target of a Merge operations remains problematic (recall that locality favors XP, economy favors X), but is not an issue of debate under free Merge (take either X, or YP, both are SOs). iii. Roots get properties in the structures they are merged into but are featureless (non-categories) before they enter the computation. It would thus be counter-intuitive to assume that Merge of a root is driven by features. In a free syntax, there is no problem in merging a category and a non-category {X,√}.43

152  A New Proposal

iv. Adjuncts are, in traditional generative grammar, optional, not selected, not subcategorized, hence, they do not fit easily into a projectional system based on featural triggers but are predicted to exist in a free syntax as {XP, YP}.

We aimed for the most natural way to combine Simplest Merge with the phenomenon of adjuncts. We hope to have shown in this chapter that labeling conceived of as Transfer appears to offer a very simple explanation of the phenomenon at hand.

Notes 1. Note that SOs still present (like copies) but inaccessible to syntax are different than inaccessible meaning removed from the syntax as in the sense of Narita (Daniel Seely p.c.). 2. We remain agnostic as to whether nominal expressions such as proper names involve a D-node. The possibility of the John I know point into that direction. What is relevant to our discussion is that we deal with XP in this case and not with a single head. 3. A reviewer objects that, given Rizzi’s conception of criteria, both C and the moved XP could bear [+Foc]. It would be quite easy to posit a feature, and something must trigger the movement, for example, license Internal Merge. First, the reviewer overlooks the point that we are searching for a solution in terms of Simplest Merge. Under this simpler conception, there are no triggers determining the computation. External and Internal Merge being two sides of the same coin, apply freely. Second, postulating a lexical feature for various distinct categories remains stipulative and considering Minimal Search, it is questionable how deep it can search into a set. Given that heads are atoms, it is counter-intuitive to search beyond head level. Anyway, a free syntax and a labeling algorithm that is unequivocally based on heads is much simpler than a deterministic syntax building on heads and features. 4. Note that since, labeling applies freely, you also could label both sets at once before you merge them, but then they would both be inaccessible to any further Merge operation. Hence, we only look at the labeling possibilities after Merge. There is no labeling before Merge, which follows from the intrinsic ordering of syntax taking place before semantics can apply. However, one has to investigate the option of labeling both sets at once after merge. What would be left in the syntax in this case is the unlabeled set containing the bynow labeled and transferred sets. This unlabeled set is the only thing further operations of Merge can be applied to, and whatever gets merged to it would serve as a labeler. So, in essence the result would be truly exocentric sets of the form i. or ii. at SEM.

i. . . . Z {Z {XP, YP}} ii. . . . {Z WP} {Z {XP, YP}}

This possibility has to be investigated further, but it is questionable whether sets like those would be interpretable. If not, they are ruled out at SEM. 5. Notice that this does not mean that labeling is driven by interpretation. The system is not crash-proof. Some derivations yield an interpretable result, others do not.

A New Proposal 153 6. Note that sentences like the following are not ruled out by labeling:

i. *Which dog do you wonder C John likes ? ii. * John thought which dog that John likes

Epstein, Kitahara and Seely (2015: 230) argue that the first example is ruled out because the embedded question can only be interpreted properly if there is a WH-Q C-Q label. Since we do not assume that there is labeling by means of feature sharing, we cannot resort to their solution. Instead, one might think about a solution in terms of the PHON interface here requiring a WH-Q C-Q agreement and ruling out WH-Q C-decl. as a violation of this agreement. 7. For various examples, consider Adger (2003: 377–388). 8. Complex copies such as welches Buch (‘which book’) must not be pronounced in i. here but, as ii shows, this seems to be a phonological restriction because the simple, expletive was (‘what’) can occur instead:

i. *Welches Buch meinst du welches Buch er gelesen hat? (‘Which book think you which book has he read’) ii. Was meinst du welches Buch er gelesen hat? (‘What think you which book has he read’?)

9. The lowest copy at SEM is of course always semantically interpreted. External Merge necessarily has an impact on interpretation. Still, it makes sense to consider further effects at intermediate positions, provided that a phase head is involved. 10. Of course, this raises questions about v as a phase head. If we want to resort to the duality of semantics, it should be one. But do we find similar effects attested at PHON for v? Also, one has to investigate A-movement further. Under the assumption that only C is a phase head, there would only be two copies in John T seems to John like her. Moreover, the role of v in connection with roots must be further examined in this connection because roots only get categorized if they are embedded in structure. Consider i. and ii. The root record either occurs in a verbal or in a nominal environment. This observation implies that roots have no category and cannot label. The label has effects on the interpretation at SEM and PHON (compare the difference in meaning and stress patterns of i. and ii. here).

i. to record a song ii. the record

11. Note that VP-fronting just depends on which set gets copied (see below). The respective labeling can still be derived accordingly, since the PP-adjunct in either case must be labeled to resolve the conflict of {{read the book} {in the garden}}.

i. {Read the book in the garden} John did . ii. {Read the book} John did in the garden.

12. As Daniel Seely (personal communication) crucially observes, there is no locality violation here. What goes wrong with this derivation has nothing (directly) to do with movement from an adjunct. Rather, what goes wrong (ultimately) is that T does not select PP (it is an interpretive problem at the interface in the analysis suggested). 13. This also accounts for the status of subjects as islands. In *{Who} did {a picture of {who}} T {a picture of {who}} please Mary? a copy would contain a copy, which makes it impossible to resolve the conflict. Copies need to

154  A New Proposal be labeled (= transferred) simultaneously. To label higher copies (who/Crelated) before lower copies (a picture of/T-related) would violate cyclicity. If the wh-copies are labeled first, this leaves the conflict XP (subject), YP unresolved. If the subject copies are labeled (= transferred) first in accord with cyclicity, this makes who inside the subject invisible. The wh-element in the CP would thus be taken as a repetition, which violates full interpretation, too. Because of cyclicity, it is not possible first to label the wh-copies and next to label the “subject’ copies. 14. In German, the grammaticality of data like Von wem gefällt dir das Foto (Of whom did a picture please you) seems to go against the analysis proposed in the text. But further data (see i. here) support the view that in German, the PP von wem is actually an adjunct outside the subject.

i. Von wem gefällt es dir? (Of whom do you like it)

15. In an example like who did you fall asleep [before meeting who], we assumed so far that extraction from the before-adjunct fails because with the conflict being resolved by labelling the adjunct, the wh-phrase gets inaccessible. Alternatively, in the other derivations, we argued that the conflict is either resolved by assigning the “wrong” label (crash at the interfaces T-PP), or the conflict remains unresolved (no label/no Transfer) which yields a kind of non-generability. Daniel Seely (p.c.) rightly asks how we can have any intuitions regarding the sentence if it is not generable. In particular, he points out that we can interpret it as for which x, you fell asleep before meeting x and that we don’t get just noise at the eardrum. Since the analysis is not framed as part of a crash-proof grammar and since we indeed have this judgment, we would expect there to be a fifth derivation fulfilling the following criteria in order to explain Daniel Seely’s observation: i. The wh-phrases must be analyzable as copies yielding the reading in ii. ii. As for which X, you fall asleep before meeting X. iii. The example cannot be judged as perfect but violates some condition at the interfaces. There is in fact a derivation with these properties. Suppose we have the set {before meeting who} and we merge it with the head ‘asleep,’ which would also label the set. Hence, this would be a derivation without the conflict {XP, YP} because we would simply merge the before-set as a complement. The derivation goes on and we can finally label the wh-copies simultaneously. All sets can be labeled/transferred and we can assign the intended interpretation to the structure, but it is degraded because ‘asleep’ does not select anything. Hence, the only derivation which allows extraction yields a selectional violation (crashes at SEM) which corresponds to our judgment. So there is no derivation left that would rule in extraction from adjuncts as perfect. 16. It should be stressed here that an analysis of the that-t-Filter (see i.), or ECPviolations, in terms of labeling as in Chomsky (2015) is not available. Since we can observe cross-linguistic variation in those cases too, it would be tempting to resort to a PHON-solution here as well. It is at least marginal in German, for instance, which favors V/2 in those cases (see ii. and iii).

i. *Who do you think who that T comes to the party? ii. ??Wer denkst du dass T zu der Party kommt? iii. Wer denkst du kommt T zu der Party

17. Another pool of data normally addressed in terms of agreement is the so called there-construction in i. here. We would suggest that the associate is actually no subject but an object or predicative related to the copula. This

A New Proposal 155 would capture the fact that the associate is no island and the phrase following it is, in fact, an adjunct, hence, an island. This is confirmed by the data in ii. and iii., respectively.

i. There are [pictures of John] on the table. ii. Who are there pictures of on the table? iii. *Which table are there pictures of John on ?

We can also cover data such as the following in iv. by analyzing the bracketed material as an adjunct. The sentence in v. is not grammatical because the example *There is likely a man is not interpretable in the first place.

iv. There is a man [likely to be in the garden]. v. *There is likely a man [to be in the garden].

18. Recall that another consequence of the proposal is that labels are assigned by inspecting the head only. We just deal with heads and non-heads in syntax. Heads label sets: there is no need for searching deeper into an atom or go beyond the head level to look for features. 19. It would actually be necessary to investigate the role of roots more closely. It seems obvious that they point to conceptual content and that the combination of roots and categories in sets make available complex thoughts that enable human beings to move away from the here and the now. In general, this yields the capacity for mind-dependent, infinite thought mediated by recursive syntax. A full discussion of roots though relevant goes beyond the scope of this book. 20. Instead of roots and categories one could simply posit non-categories and categories. Linguistic atomic expressions can thus be divided into non-­ categories and categories. Only the latter can label sets. Crucially, meaning is conveyed in structure. Consider i. and ii.

i. He has a dream. ii. He dreams of a better world.

D-n√dream T-v-√dream

Basically, roots (non-categories) have no properties in isolation. They are simple units without internal structure. They occur obligatorily at the bottom of a syntactic derivation (“first Merge”), always embedded under categories. Any property is encoded in the structure (or set). Hence, meaning in general is conveyed in sets which results in a flexibility that cannot be bound to a single item. Data like iii. and iv. illustrate this.

iii. John shouted. iv. John shouted the mouse out of the room.

Because meaning in this sense is always composed, human thought can only be mind-dependent. There are no elements referring to the external world. 21. The relation can be causal or temporal, what they share is the single event character, where one event necessarily either causes the other, or precedes the other. Truswell concludes that the Single Event Condition applies to complements and bare Present Participial Adjuncts likewise. It conditions A-bar-­dependencies since they are legitimate only if the minimal constituent containing head and foot of the chain describes a single event (Truswell 2011: 24). 22. Notice that German does not have complements of the —ing-type. The prediction would thus be that German does not have transparent, participial adjuncts. This seems to be correct because only the full adjunct can be fronted. Extraction from the adjunct is banned:

i. Welches Lied singend, kommt er an? (which song singing arrives he prt.?) ii. *Welches Lied kommt er singend an? (*Which song comes he singing prt.?)

156  A New Proposal 23. Similar considerations might hold for double objects since, as Johnson observes, either the direct object or the indirect object behaves as an island, blocking either focus projection or movement (compare Johnson 2003: 213). We will have little to say about double objects here. Double objects would be necessary to investigate in detail in terms of labeling and the proposal outlined. 24. Rizzi (1990: 4), illustrates wh-islands with data like the following:

i. ??Which puzzle do you wonder how John could solve t t? ii. *Which student do you wonder how t could solve the puzzle t? iii. *How do you wonder which puzzle John could solve t t?

He analyzes the asymmetry in terms of his Relativized Minimality condition plus binding (via referential indices). Lasnik and Saito (1984) suggest an account based on lexical versus antecedent-government which is no longer applicable in minimalism since government has been completely eliminated. The same applies to the accounts of Chomsky (1986) and Lasnik and Saito (1984). Still, the asymmetry must be explained. Consider iv. and v. further underpinning its relevance (see Lasnik & Saito 1992: 174–175):

iv. *Why do you wonder who left? v. ??What do you wonder who wrote?

Lasnik (2018) presents subtle data suggesting that binding of subjects into the wh-domains influences the acceptability of wh-islands. He puts forth an interesting view based on feature valuation. However, the strong violation of adjunct extraction as opposed to objects as in the data vi-vii (see Lasnik 2018: 69) cannot be captured this way:

vi. What problemi did you figure out [howj [PRO to solve ti tj]]? vii. *Howj did you figure out [which problemi [PRO to solve ti tj]]?

Lasnik puts this question aside for future investigation (Lasnik 2018: 69). 25. Note that we can apply this account of islands to complex-NP islands as in i. as well since relative clauses are standardly analyzed as adjuncts:

i. *Which man did they meet [the girl who likes ]

6. Thanks go to a reviewer for mentioning Sprouse and Hornstein (2013). 2 27. We refer the reader to the framework of distributed morphology (see for instance, Halle & Marantz 1993; Embick & Noyer 2007; Embick 2015; Harley 2014), the exo-skeletal account of Borer (2003, 2013, 2014) or a derivational solution to roots suggested by de Belder and Craenenbroeck (2015). 28. Recall that we suggested that categories and non-categories (roots) are atoms of the computation which get instantiated at the point of externalization. The sets created in the computation can only be labeled by categories. 29. There are, of course, also examples demonstrating that a complement can be merged involving the same root but distinct categories as i and ii show:

i. This is similar to his analysis. ii. There is some similarity to his analysis.

30. Recall that we assume that the atoms of human language are categories (such as C, T, D, v, a, n) on the one hand and non-categories (roots) on the other hand. 31. That labeling takes place in cyclic fashion is assumed for independent reasons. Subject copies have to be labeled prior to copies which have an instance merged to higher C, for instance, wh-phrases. Remember that if higher wh-phrases would be transferred before the lower DP-subject copies in i., subject-islands would be ruled in. Since it applies cyclically, the subject

A New Proposal 157 copies are labeled and transferred first. Consequently, there cannot be a whcopy contained in a subject-copy. Since the derivation is bottom-up, it is no strange assumption that labeling takes place in a cyclic fashion, too, though apart from this it is free in principle and underlies interpretation at SEM.

i. *Who does [a picture of who] T please John?

32. Note that “cyclic” refers to the level of the respective sets. So, for instance, DP and root-v copies are actually sets at the same level since their respective copies are contained in sister sets. Reversing the order of 1–2 would not violate cyclicity, but would simply yield no felicitous result at SEM. However, labeling of wh-copies before the DP-copies would violate cyclicity. 33. See Dékány (2018) for an excellent, critical overview of approaches to head movement. 34. A syntactic approach to head movement is further complicated by reasons of locality. Carstens, Hornstein and Seely (2016) discuss a problem of equidistance arising with both T-to-C and v-to-T because the D-head of the external argument should be available too. Therefore the question arises as to what guarantees that it is v or T that move, but not D, which should be found by Minimal Search, too. The authors discuss a possible solution by assuming feature-inheritance or feature-copying. Features and triggers would, of course, complicate the picture. 35. A good candidate for a language-particular rule applying at PHON is dosupport. It is necessary in English (see i.) but not in standard German (see ii.), to insert do in certain contexts though there is some minor variation in German. Some non-standard dialects allow a supportive element tun (see iii.).

i. Does he work a lot? ii. Arbeitet er viel? iii. Tut er viel arbeiten?

36. Positional effects might of course also be traced back to alternative copypronouncement if Internal Merge has taken place in syntax. 37. A suggestion might appeal to features and heads at PHON. If Agree which is highly parametric in nature applies at externalization and the linear order of heads is also treated here, the following patterns might be explicable.

i. ii. iii. iv.

Who C T v-saw John? Who C John T John v-see ? Where C John T John v-see Mary ? Who C T Neg v-see John?

In i., there is one uninterrupted or adjacent sequence of phi and Q features and do-support does not apply. C in ii and iii. share the Q feature with the whphrase, but not the phi feature with John/T. Hence, at C the sequence is interrupted, which needs to be repaired by insertion of a form of do. In negated sentences, the Neg head has the same interrupting effect because it does not share any of those features. Consequently, do needs to be inserted (see iv.).   Comparing English and German, we need to refer to the specific heads (categories) and to the positions on the sequence. In English, only the highest head with a Q-feature needs to be pronounced by a category (not including the lexical root). In German, the highest head must be overt in any case and this includes categories with roots. Thus, we either have V/2 or an overt complementizer dass (that).

i. Gestern C-arbeitete er (Yesterday worked he/He worked yesterday). ii. C-dass er gestern arbeitete (that he yesterday worked/that he worked yesterday).

158  A New Proposal This would mean that, in principle, any head of the sequence C-T-v-root could be overt, but that it is an idiosyncratic decision of particular languages at which position the respective element bearing specific features is expressed phonologically. 38. At the end of Chapter Four, we thought about an evolutionary scenario based on the following steps i. atoms: root, categories ii. recursive Merge: SO, SO → labels (=Transfer) → SEM (uniformity) iii. Map: SEM, PHON features → Externalization iv. Agree: feature values: Fv, Fv (diversity). The idea behind this is that cross-linguistic variation is associated with the externalized PHON side only. We assume that all languages have categories like C, T, v, D, n, and so on, and those categories can label = transfer. Semantic features like (non-)finiteness, (in)definiteness, q(uestion), and so on can be externalized. PHON pronounces SEM. “Nominal” forms (like case) and “verbal” forms (any inflections) would under this view be PHON-members. The same applies to agreement (in feature values) and copy pronouncing. These properties are supposed to be highly language-particular in nature. Consider some examples that would be violations at PHON in English under this view:

i. *Him likes she. (*Acc < Nom) ii. * They tried to liked her. (*to Past-ed) iii. *They likes him. (*phi-pl-phi-sing) iv. *They wonder Bill C Mary likes. (*non-Q-C-Q)

39. In general, PHON deals with linearization and with realization. In the same vein, the idea of sequences in the case of do-support should be understood. 40. Narita (2012) suggests an account in terms of phases to explain the complementarity of head and phrasal movement: T/*TP, DP, *D). 41. At PHON, case forms are probably related to the respective heads Tfin— nominative and vtrans accusative. 42. The cyclicity of labeling and the respective order needs to be scrutinized as well. The conflicts are said to be resolved immediately after External Merge or immediately after Internal Merge. The latter (“after movement”) necessarily implies a delay because copies need to be labeled simultaneously. The examples with head movement would point into the direction that labeling in this case is in fact delayed further—we cannot, for instance, label the PerfP in the example we discussed in this section before we reach the top of the derivation because this would render the PerfP invisible and inaccessible to syntactic operations. Neither the DP subject(-copies) nor the wh-copies would thus be accessible. This is clearly a false prediction as the reader will have already recognized. 43. De Belder and Craenenbroeck (2015) even assume that first Merge joins a categorizer and an empty slot that is replaced by root material at PHON. It also is impossible, under this view, to assume that this operation is triggered by features.

References Adger, D. 2003. Core Syntax: A Minimalist Approach. Oxford: Oxford University Press. Arregi, K. and Pietraszko, A. 2018a. Generalized head movement. Proceedings of the Linguistic Society of America 3: 1–15. Arregi, K. and Pietraszko, A. 2018b. The Ups and Downs of Head Displacement. Available at Lingbuzz/004096.

A New Proposal 159 Bailyn, J. F. 2018. Cost and Intervention: A Strong Theory of Weak Islands. Paper presented at the 13th European Conference on Formal Description of Slavic Languages (FDSL 13), Göttingen, December 5. Berwick, R. C. and Chomsky, N. 2016. Why Only Us? Language and Evolution. Cambridge, MA: MIT Press. Biberauer, T. and Roberts, I. 2010. Subjects, Tense and Verb Movement in Germanic and Romance. In Biberauer, T., Holmberg, A., Roberts, I. and Sheehan, M. (eds.), Parametric Variation: Null Subjects in Minimalist Theory, 263–302. Cambridge: Cambridge University Press. Bobaljik, J. D. 1995. In Terms of Merge: Copy and Head Movement. MIT Working Papers in Linguistics 27: 41–64. Boeckx, C. and Stjepanović, S. 2001. Heading Toward PF. Linguistic Inquiry 32: 345–355. Borer, H. 2003. Exo-Skeletal vs. Endo-Skeletal Explanations: Syntactic Projections and the Lexicon. In Moore, J. C. and Polinsky, M. (eds.), The Nature of Explanation in Linguistic Theory, 31–67. Stanford, CA: CSLI Publications. Borer, H. 2013. Structuring Sense Vol III Taking Form. Oxford: Oxford University Press. Borer, H. 2014. The Category of Roots. In Alexiadou, A., Borer, H. and Schäfer, F. (eds.), The Syntax of Roots, 112–148. Oxford: Oxford University Press. Bowers, J. 2018. Deriving Syntactic Relations. Cambridge: Cambridge University Press. Carstens, V., Hornstein, N. and Seely, T. D. 2016. Head Movement in Problems of Projection. The Linguistic Review 33 (1): 67–86. Cecchetto, C. and Donati, C. 2015. (Re)labeling. Cambridge, MA: MIT Press. Chomsky, N. 1957. Syntactic Structures. The Hague, Netherlands: Mouton de Gruyter [reprinted in Hornstein, N., Lasnik, H. Patel-Grozs, P. and Yang, C. (eds.), Syntactic Structures After 60 Years: The Impact of the Chomskyan Revolution in Linguistics, 1–130. Mouton: de Gruyter]. Chomsky, N. 1986. Barriers. Cambridge, MA: MIT Press. Chomsky, N. 2013. Problems of Projection. Lingua 130: 33–49. Chomsky, N. 2015. Problems of Projections: Extensions. In di Domenico, E., Haman, C. and Matteini, S. (eds.), Structures, Strategies and Beyond: Studies in Honor of Adriana Belletti, 3–16. Amsterdam and Philadelphia: John Benjamins. Chomsky, N., Gallego, A. and Ott, D. 2019. Generative Grammar and the Faculty of Language: Insights, Questions and Challenges. To appear in Gallego, A. and Ott, D. (eds.), Generative Syntax: Questions, Crossroads, and Challenges. Special Issue of Catalan Journal of Linguistics, 2019. Cinque, G. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. Citko, B. Missing Labels: Head Movement as Project Both. 2008. In Chang, C. and Haynie, J. H. (eds.), Proceedings of the 26th West Coast Conference on Formal Linguistics, 121–128. Somerville, MA: Cascadilla Proceedings Project. De Belder, M. and Craenenbroeck, J. 2015. How to Merge a Root. Linguistic Inquiry 46: 625–655. Dékány, É. 2018. Approaches to Head Movement: A Critical Assessment. Glossa—A Journal of General Linguistics 3: 1–43. deVos, M. 2014. Head Movement Is an Artefact of Optimal Solutions to Linearization Paradoxes. Stellenbosch Papers in Linguistics. 44: 23–48.

160  A New Proposal Embick, D. 2015. The Morpheme—A Theoretical Introduction. Boston and Berlin: de Gruyter Mouton. Embick, D. and Noyer, R. 2007. Distributed Morphology and the Syntax-­ Morphology Interface. In Ramchand, G. and Reiss, C. (eds.), The Oxford Handbook of Linguistic Interfaces, 290–324. Oxford: Oxford University Press. Epstein, S. D., Kitahara, H. and Seely, T. D. 2015. What Do We Wonder Is Not Syntactic? In Epstein, S. D., Kitahara, H. and Seely, T. D. (eds.), Explorations in Maximizing Syntactic Minimizing, 222–239. New York and London: Routledge. Epstein, S. D., Kitahara, H. and Seely, T. D. 2016. Phase Cancellation by External Pair-Merge of Heads. The Linguistic Review 33: 87–102. Fukui, N. and Narita, H. 2017. Merge and (A)symmetry. In Fukui, N. (ed.), Merge in the Mind, 9–34. London and New York: Routledge. Halle, M. and Marantz, A. 1993. Distributed Morphology and the Pieces of Inflection. In Hale, K. and Keyser, S. (eds.), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, 111–176. Cambridge, MA: MIT Press. Harizanov, B. and Gribanova, V. 2019. Whither Head Movement? Natural Language and Linguistic Theory (NLLT) 37: 461–522. Harley, H. 2004. Merge, Conflation, and Head Movement: The First Sister Principle Revisited. In Moulton, K. M. (ed.), Proceedings of NELS 34, 239–245. Amherst, MA: UMass, GLSA. Harley, H., 2014. On the Identity of Roots. Theoretical Linguistics 40: 225–276. Hofmeister, P. and Sag, I. 2010. Cognitive Constraints on Island Effects. Language 86: 366–415. Hornstein, N. and Nunes, J. 2008. Adjunction, Labeling, and Bare Phrase Structure. Biolinguistics 2: 57–86. Johnson, K. 2003. Towards an Etiology of Adjunct Islands. Nordlyd 31: 187–215. Koeneman, O. 2000. The Flexible Nature of Head Movement. Doctoral Dissertation, University of Utrecht. LaCara, N. 2018. Head Movement and Ellipsis Licensing. Paper presented at the 92nd Meeting of the Annual LSA Meeting, January. Lasnik, H. 2018. Two Families of Questions. In Gallego, A. and Martin, R. (eds.), Language, Syntax and the Natural Sciences, 58–76 Cambridge: Cambridge University Press. Lasnik, H. and Saito, M. 1984. On the Nature of Proper Government. Linguistic Inquiry 15: 235–289. Lasnik, H. and Saito, M. 1992. Move-α—Conditions on Its Applications and Output. Cambridge: MIT Press. Matushansky, O. 2006. Head Movement in Linguistic Theory. Linguistic Inquiry 37: 69–109. McCloskey, J. 2016. Interpretation and the Typology of Head Movement: A ReAssessment. Handout of a talk presented at the Workshop on the Status of Head Movement in Linguistic Theory. Müller, G. 2004. Verb-Second as vP-First. Journal of Comparative Germanic Linguistics 7: 179–234. Narita, H. 2012. Head-Detection, Phases, and the Complementarity of XP- vs. X0-Movement. Online Proceedings of GLOW in Asia Workshop for Young Scholars.

A New Proposal 161 Platzack, C. 2013. Head Movement as a Phonological Operation. In Lai-Shen Cheng, L. and Corver, N. (eds.), Diagnosing Syntax, 21–43. Oxford: Oxford University Press. Rizzi, L. 1990. Relativized Minimality. Cambridge, MA: MIT Press. Rizzi, L. 2016. Labeling, Maximality and the Head—Phrase Distinction. The Linguistic Review 33: 103–127. Schoorlemmer, E. and Temmerman, T. 2012. Head Movement as a PF Phenomenon: Evidence from Identity under Ellipsis. In Choi, J., Hogue, E. A., Punske, J., Tat, D., Schertz, J. and Trueman, A. (eds.), Proceedings of the 29th West Coast Conference on Formal Linguistics, 232–240. Somerville, MA: Cascadilla Proceedings Project. Seely, T. D. 2006. Merge, Derivational C-command and Subcategorization in a Label-Free Syntax. In Boeckx, C. (ed.), Minimalist Essays, 182–217. Amsterdam and Philadelphia: John Benjamins [reprinted in Epstein, S. D., Kitahara, H. and Seely, T. D. 2015. Explorations in Maximizing Syntactic Minimization— Routledge Leading Linguists, 116–154. London and New York: Routledge]. Sprouse, J. and Hornstein, N. 2013. Experimental Syntax and Island Effects: Toward a Comprehensive Theory of Islands. In Sprouse, J. and Hornstein, N. (eds.), Experimental Syntax and Island Effects, 1–18. Cambridge: Cambridge University Press. Stepanov, A. 2000. The Timing of Adjunction. In NELS 30, 597–611. Amherst: Mass GLSA. Stepanov, A. 2001. Late Adjunction and Minimalist Phrase Structure. Syntax 4: 94–125. Surányi, B. 2005. Head Movement and Reprojection. In Annales Universitatis Scientiarum Budapestenesis de Rolando Eötvös Nominatae. Sectio linguistica, Tomus XXIV, 313–342. Budapest: ELTE. Takita, K., Goto, N. and Shibata, Y. 2016. Labeling Through Spell-Out. The Linguistic Review 33: 177–198. Truswell, R. 2007a. Extraction from Adjuncts and the Structure of Events. Lingua 117: 1355–1377. Truswell, R. 2007b. Tense, Events, and the Extraction from Adjuncts. Proceedings from the Annual Meeting of the Chicago Linguistic Society (CLS) 43: 233–247. Truswell, R. 2011. Event Structure and A’-Locality. In Folli, R. and Ulbrich, C. (eds.), Interfaces in Linguistics—New Research Perspectives, 17–24. Oxford: Oxford University Press. Zeijlstra, H. 2019. Labeling. Selection and Feature Checking. To appear in Smith, P., Mursell, J. and Hartmann, K. (eds.), Agree to Agree: Agreement in the Minimalist Programme, 137–174. Berlin: Language Science Press. Zwart, C.J-W. 2001. Syntactic and Phonological Verb Movement. Syntax 4: 34–62.

6 Conclusions and Outlook

Adjuncts still pose serious questions today, but we hope to have shown that a fuller understanding of the relation between structure-building conceived of as Simplest Merge and interpretation at the interfaces mediated by labeling (= Transfer) provides us with an analysis of adjuncts in accordance with SMT. We observed that there is a clear distinction between adjuncts and arguments reflected in the data in terms of the criteria listed in (1) and presented in Chapter Two. (1)

i. Theta-marking ii. Limitation of the number iii. Positional flexibility iv. Target of operations (ellipsis, extraction . . .) v. Reconstruction and binding

We also discovered loose ends. For instance, we found a few gaps in the dichotomy of adjuncts and arguments casting some doubts on the sharp distinction made above. Some adjuncts do not cause island violations and some arguments (subjects) form islands too. The binding data raise questions in this connection too (Pesetsky’s paradox, asymmetries between A- and A-bar-movement), but we might well argue that binding relations are determined at the semantic interface by rules applying at this stage only. Sportiche (2019a, 2019b) presents an interesting analysis building on a rule called Neglect, nicely covering some of the asymmetries observed here. Another question to be raised is whether the difference between adjuncts and arguments is a syntactic one, whether it is located in semantics, or whether the answer lies somewhere in the middle. Basically, adjuncts underlie the same structure-building operation as any other phrase does. It goes without saying that adjuncts are semantically interpreted. Consider the data in (2), (3), and (4) for some aspects of semantic information.

Conclusions and Outlook 163 (2) a. John walked home slowly on Monday. b. John walked home slowly. c. John walked home. (3) a. John jumps over the fence carelessly. b. *John has a knife carelessly. c. John has a knife in his pocket. (4) a. John walked home slowly. b. John walked home thoroughly exhausted. The sentences in (2) show entailment relations with a. entailing b. and b. entailing c. in turn. The examples in (3) illustrate the fact that the meaning of adjuncts relate to the meaning of the event or state expressed. A manner adjunct like carelessly, for instance, cannot be construed with a state, but a locative adjunct can be combined with either an event or a state. The pair in (4) shows that manner adverbs modify the verbal domain, while adjectival adjuncts form a secondary predicate (or a small clause).1 Naturally, adjuncts are also interpreted phonologically because they have to be linearized with the rest of the structure, and they are pronounced in certain positions. Prosody, stress, and focus are additional factors located at the PHON interface. But how does syntax deal with adjuncts? How do they enter the derivation before they can be interpreted? In the ideal case, they enter the syntax as any other element does, as we repeatedly emphasized in this book, since this accords with the most recent minimalist framework outlined briefly in Chapter Three. As we demonstrated in the forgoing discussion, the various GB-analyses of adjuncts cannot be used in minimalism since they refer to the X-bar format, which is not in line with minimalist structure-building. Nevertheless, we found that there are minimalist “successors” to the GB-accounts. This is roughly presented in (5). (5)

i. Chomsky-adjunction  Pair-Merge (untriggered) ii. Specifier-licensing  feature-triggered projection

We also saw that many recent approaches in minimalism which we addressed and put under critical scrutiny in Chapter Four deal with labeling questions and range from adjuncts as being un-labeled (Hornstein & Nunes 2008; Hornstein 2009) to adjuncts as labelers (Zeijlstra 2019).

164  Conclusions and Outlook Accounts addressing the labeling conflict {XP, YP} raised by stable host-adjunct structures in particular either resolve it by means of the operation Merge (Narita 2014: H-Alpha on Merge, Jakielaszek 2017: Self-Merge, Oseki 2015: double-peak/Transfer) or do not resolve it at all (Gallego 2010: feature-inheritance makes the difference). Given the wealth of interesting and elaborated work on adjuncts by both syntacticians and semanticists over many decades, adjuncts should be a topic in ongoing minimalism too. As we stressed throughout our argumentation, we should keep to the goals of SMT and not add any extra mechanism to deal with adjuncts. A minimal computation should lay the foundation of our analysis. A complete overview of the minimalist approaches that we discussed in the previous chapters is hard to present in a single table. Abstracting from the details, we nevertheless try and compare the accounts (without the one we proposed) in terms of two minimalist concerns given in (6) that are in accordance with SMT. We ignore any specific assumption of the respective accounts and compare them only in terms of simplicity as defined in (6) i–ii. (6)

i. Simplicity: one syntactic operation is simpler than two (or more) different types of syntactic operations. ii. Simplicity: free, un-driven syntax with interpretation located at the interfaces only is simpler than triggered syntactic operations.

With the two criteria for simplicity in (6), we arrive at the simplified overview of the approaches discussed in the previous chapters given in Table 6.1. It seems that we have two candidates “winning the game,” namely, Narita (2014) and Jakielaszek (2017). But the former has to put a restriction Table 6.1 Overview of the Minimalist Approaches to Adjuncts

Syntactic Operations Free syntax

Hornstein (2009)

Gallego (2010)

Chomsky (2013)

Narita (2014)

Concatenate & Label no

(Set-)Merge & Agree no

Set-Merge & Pair-Merge yes

(Set-)Merge

Jakielaszek (2017)

yes

Oseki (2015)

Hunter (2015)

Nakashima (2017)

Zeijlstra (2019)

Syntactic Operations

(Set-)Merge

Free syntax

no

Insert & (Set-) Merge no

(Set-)Merge (Set-)Merge (Set-) Merge & Agree no yes no

Conclusions and Outlook 165 on Merge itself—the ESC, the endocentric structuring constraint—and the latter resorts to Internal Self-Merge, a vacuous operation that is doubtful in terms of computational efficiency. The hypothesis we put forth in Chapter Five suggests reducing Transfer to labeling. It keeps to (6) i–ii: there is, in fact, only one syntactic operation, Simplest Merge, that applies freely without any featural triggers. Structure-building is always done by Simplest Merge, whether we consider adjuncts or any other phrase in the derivation. Labeling mediates between syntax and interpretation and can also be viewed as applying freely. If a “wrong” label is assigned, the result does not receive a proper interpretation. If no label is assigned, the set is not accessible to interpretation at all. In a nutshell, we propose to define labeling as follows. (7) Labeling = Transfer Inspect the set for a head (H, XP) (immediately) after (External or Internal) Merge with HP labeled (= transferred) in i. and ii first.

i. {HP, XP} ii. {HP {XP . . . HP . . .}

Notice that this view implies a further advantage because, instead of three different operations—Merge, Labeling, and Transfer—we reduce to two—Merge, as the syntactic structure-building device, and Labeling, relating to the interfaces. For the labeling conflict XP, YP to be resolved we concluded that for head inspection to work properly, this basically means that one (XP) or the other (YP) is to be labeled at once or is a copy to be labeled when the last copy is ultimately merged. Hence, either labeling after External or after Internal Merge resolves the conflict. Where are we right now? In a recent paper with the title Generative Grammar and the Faculty of Language: Insights, Questions and Challenges, Chomsky, et al., discuss MERGE as taking place in a workspace. Take a look at the following quote. (8) MERGE operates over syntactic objects placed in a workspace: the MERGE-mates X and Y are either taken from the lexicon or were assembled previously within the same workspace. [. . .] We assume that MERGE is strictly binary: given that this is what is minimally necessary to create hierarchical structure, we assume that this is the only operation defined by UG (although adjunction structures may necessitate a separate operation. (Chomsky, Gallego & Ott 2019: 9)

166  Conclusions and Outlook Recall our discussion on structure-building and the comparison with the creation of a painting. We might again look for an analogy. The workspace of creation in art is the canvas, and the painter has access to the (external) colors she uses and the objects in the canvas she has painted already. Similarly, the lexicon can be freely accessed, and the syntactic objects in the workspace are accessible too. The painting is finished when the space is exhausted, and the derivation stops when the workspace contains a single object. Nevertheless, the analogy does not go through because the procedures differ in a crucial way. In the creative act of painting the mechanism is additive rather than recursive. Again, we observe that human language is unique in the basic property of recursive generation. Consequently, it is this very property that needs to be implemented in a UG-based view on language. So, we have a workspace and recursion. What else can be said about these ingredients? (9) All syntactic objects in the lexicon and in the workspace WS are accessible to MERGE; there is no need for a SELECT operation (as in, e.g., Chomsky 1995). WS represents the stage of the derivation at any given point. The basic property of recursive generation requires that any object already generated be accessible to further operations. WS can contain multiple objects at a given stage, so as to permit formation of {XP, YP} structures (subject-predicate constructions) by EM. A derivation may, but need not, terminate whenever WS contains a single object; if it terminates in any other situation, no coherent interpretation can be assigned. (Chomsky, Gallego & Ott 2019: 19) We get a fairly simple picture. MERGE operates on elements taken from the lexicon (some mental store that must be accessible anyway) and on those sets assembled in the workspace (WS). Syntactic objects or sets can be of the form {XP, YP}. Besides subjects and predicates mentioned in the quote, this form, under our view, also subsumes sets with operator and sets with adjuncts. At the semantic interface, those sets need to be distinguished, and this can arguably be done by means of labeling (= Transfer) so that we get (10) at SEM. (10) At SEM a. {vP {DP, vP}}, {TP {DP, TP}} b. {CP {wh-DP, CP}} c. {vP {vP, PP}} Assuming, as we do, that labeling is Transfer and that it makes the sets ready for interpretation, this might well be implemented into the

Conclusions and Outlook 167 WS account. The derivation there is said to terminate if a single object remains in the WS. This would have to be revised along the following lines under our proposal. A derivation basically terminates if no accessible syntactic object remains in the WS, because a single, labeled object is no longer accessible to MERGE. There is no syntactic object left in the WS for further syntactic operations, but a single labeled set composed of other labeled sets is visible now at SEM and subject to the rules of interpretation. We do not assume literal removal. The idea is to make the sets created in syntax by means of Merge inaccessible to Merge as soon as they are labeled/transferred. The WS is reduced insofar as Labeling/Transfer reduces the number of SOs available for Merge. There is no NTC/LCR violation, but we basically deal with the situation given in (11). (11)

i. Syntax: unlabeled sets are accessible to Merge ii. SEM: labeled sets are accessible to interpretation

Of course, the reversal holds too. (12)

i. Syntax: labeled sets are inaccessible to Merge ii. SEM: unlabeled sets are inaccessible to interpretation

The derivation comes to an end when there is a single labeled/transferred set at SEM (containing labeled/transferred sets). Crucially, the quote from the paper mentioned above goes on to raise the question of whether items of the WS are replaced by sets or whether sets are added. The former view is more restrictive in terms of computational efficiency and determinacy because the resources are restricted, and further operations unambiguously refer to the sets created (see Chomsky, Gallego & Ott 2019: 19). Hence, MERGE is computationally efficient and replaces objects by creating sets. Notably, the paper also addresses adjuncts as a challenge for future research. Pair-Merge is still mentioned as an option, but the authors stress that another solution would be preferred. Consider the quote below for this. (13) PAIR-MERGE is a formally distinct operation from Simplest MERGE, hence raises problems of evolvability. Ideally, it could be shown to be dispensable. We do not take up the challenge here. (Chomsky, Gallego & Ott 2019: 24)

168  Conclusions and Outlook In the forgoing chapters, we took up the challenge. If, indeed, the analysis we suggested and developed in Chapter Five holds, a Simplest MERGE account is not only strongly favored but also possible. In two recent talks, in Arizona and in Reading, Chomsky (2017a, 2017b)2 elaborated on the ideas touched upon in the aforementioned paper. He presents a crucial modification of the operation Merge (14) and states seven general conditions (consider (15)) that must be satisfied. (14) MERGE is an operation on the workspace (WS). (15) Seven conditions

i. Descriptive adequacy ii. SMT iii. Restrict computational resources! iv. Determinacy v. Coherence (stability of semantic interpretation) vi. The Faculty of Language involves recursion vii. Binary sets

Merge, now MERGE, an operation on the workspace rather than an operation on syntactic objects, can be represented as in (16) and (17) with crucial reference to WS. (16) WS = sequence i. Σ = (X1, . . . Xn) ii. Each X is accessible (recursion) iii. Σ exhausts the WS (17)  MERGE:

i. Take any element from the workspace (= recursion) ii. Keep the workspace from expanding (= internal Merge) iii. Reduce the workspace (= external Merge)

An interesting and welcome side effect of the new conditions is that Parallel Merge and Sideward Merge are completely eliminated, which simplifies the computation accordingly. The idea of a workspace is quite useful and convincing. Let us play a pretty bold mind game and compare this conception with our universe. What is minimally needed to create a universe? We basically need three things: matter (or mass), energy, and space to which the laws of nature apply, that is, gravity, motion and the rules of quantum mechanics. Now, what is minimally needed to create language? We need three ingredients here as well: atoms, sets (MERGE), and the workspace. Crucially, we know from Einstein’s theory of relativity that mass and energy are basically the same (E= mc2). Can we go one step further with our comparison? Suppose we would try something

Conclusions and Outlook 169 similar with language and say that, just as mass can be used as energy and energy as mass, in the computation, atoms can be used as sets (as in {X}, {Y, Z} or {X}, {Y}) and sets can be used as atoms (as in {X} {Y, Z} or {X, Y} {Z, W}). If we relate atoms and sets in this way, we simply get recursion. Hence, we are, in fact, left with recursion and the workspace as the remaining ingredients we minimally need for human language, as shown in Table 6.2. Of course, we are linguists and not physicists, but playing around with ideas like these stresses the main point, namely, that scientists are always searching for the simplest possible solution no matter what field of interest they are working on. Line three of Table 6.2 belongs to the third factor. Recall that minimalist reasoning also implies derivation of aspects of language from language-independent laws. The three factors coming into play are summarized in Table 6.3. Deriving as much as possible from third-factor considerations is a minimalist strategy too. Simplest Merge is symmetric and labeling/Transfer is asymmetric. We create sets before we label them. Interestingly, the laws of nature also start with symmetry, and after a phase transition, symmetry breaks. Liquid water, for instance, is symmetrical. However, when water is cooling, a direction emerges and yields crystals. Hence, we get the following suggestive parallel in Table 6.4. Consequently, one might argue that the transition from symmetry (syntax) to asymmetry (interpretation) as we find it in natural language is actually third-factor driven because a law of nature is applied to language. What about Pair-Merge and the workspace now? Should it still be in the game? Chomsky mentions the possibility in his talks, but, if we take Table 6.2 A  Comparison Infinite (Expanding) Universe

Infinite Human Language

Mass = Energy Space General laws: gravity, principle of uncertainty, etc.

Atoms = Sets Workspace Efficient computation: binarity, NTC/ LCR, etc.

Table 6.3 The Interacting Three Factors First Factor

Second Factor

Third Factor

UG

external data

Uniform Innateness Minimize UG! (SMT)

diverse acquisition particular languages

general cognition & computational efficiency language-independent laws of nature Maximize the third factor!

170  Conclusions and Outlook Table 6.4 The Laws of Nature & the Transition from Symmetry to Asymmetry* Symmetry

Phase Transition

Broken Symmetry

symmetric, liquid water symmetric sets

Cooling labeling (=Transfer)

Crystals labeled sets

* In physics, the asymmetric effect is actually to be defined in group-theoretic terms. This means that the initial symmetry group is broken to a sub-group of the phenomenon’s symmetry group.

the new concept of MERGE and the idea of the workspace seriously, there should neither be an operation besides MERGE nor a separate dimension. Adjuncts are (complex) sets as any other phrase, and they are subject to Merge operations. A PP-adjunct can be topicalized by Internal Merge just as its host can be. Hence, either set is accessible in the workspace and thus in the same dimension, as the data in (18) show. (18) a. In his garden John planted a chestnut tree . b. (and) plant a chestnut tree John did in his garden. Both sentences can easily be derived under the proposal suggested in Chapter Five by referring to Simplest Merge only. A second operation Pair-Merge can be dispensed with. How can we derive the dichotomy repeatedly observed in the literature? Does the island behavior of adjuncts follow in a principled way? Interestingly, Johnson (2003) looks into the etiology of adjunct islands by referring to the procedural algorithm taking place in the numeration. He draws the conclusion that adjuncts need to be re-numerated to be successfully integrated into the structure. Merge operating on syntactic objects in the (re-)numeration is quite similar to the idea of operations applying to a workspace. Nevertheless, Johnson needs to resort to projection/labeling in syntax. Asymmetry in syntax and at the interface would implement an architectural redundancy that is neither minimal nor in accordance with computational efficiency nor with the laws of nature that suggest a clear distinction between the state of symmetry and the state of asymmetry (see Table 6.4 again). Hence, the redundancy should be abandoned. Recall that our proposal makes the correct prediction about adjuncts as islands, and this applies to subjects too. Adjuncts and subject copies need to be labeled first in order to resolve the {XP, YP} conflict. Being labeled is being transferred, and it follows automatically that neither adjuncts nor subjects may contain copies. If they do, they cannot be related to the higher elements, which therefore would only be interpretable as

Conclusions and Outlook 171 repetitions. Alternatively, the conflict remains, which also results in an infelicity, as we discussed in the previous chapter. Crucially, copies need to remain copies throughout the derivation and must be labeled/transferred as a single object simultaneously. If they are not, this might involve a violation of one of the conditions Chomsky puts forth in his talks, namely, coherence (or stability of semantic interpretation (15) v.). Given the new definition of MERGE, the operation either has a decreasing effect on the number of elements in the WS (“external” MERGE) or it keeps the cardinal number intact (“internal” MERGE). The derivation is thus a transition of workspace sequences up to the point where just a single object is left—or, alternatively, as we suggested, until a single object eventually completely labeled and containing labeled sets is available at the interfaces ready for interpretation. If the workspace must not be expanded by operations applying to it, then it makes good sense to assume that Transfer (= labeling) reduces the WS too by turning the sets into labeled objects that are no longer accessible to syntax. To sum up, it would be important to further clarify what is going on in the workspace and find out more about MERGE on the road ahead in minimalist theorizing. Seen from a minimalist standpoint, what seems to be obvious in any case is that analyzing adjuncts without resorting to extra-mechanisms or additional syntactic operations, but adhering to Simplest MERGE only and keeping to an efficient computation driven by the third factor, should be our ultimate goal. It goes without saying that the implications of our proposal need to be investigated closely. We have examined, in Chapter Five, how islands fit into the picture. There are ways to capture the phenomenon in terms of labeling with reference to the interfaces. We also looked into the puzzle that head movement poses for a minimalist account and worked out an analysis that is based on Simplest Merge. In Chapter Five, the labelingis-Transfer approach is argued to be able to cope with head movement without “adjunction” if coupled with the language-particular PHONside of the phenomenon. Hopefully, the discussion of the preceding chapters contributes to a better understanding of adjuncts in minimalism and invites other scholars to enter the debate by keeping to the most minimal assumptions, according to which simplifying and minimizing generative grammar in general—and UG in particular—necessarily goes hand in hand with maximizing the explanatory effects.

Notes 1. Of course, there are ways to capture the semantics of adjuncts. They might be taken as parts of the predicates (by means of predicate composition) or the rules of interpretation may analyze adjuncts as predicates themselves requiring

172  Conclusions and Outlook a “subject.” For instance, the adjunct in i. might semantically function as a predicate, which makes the meaning similar to the meaning assigned to ii.

i. He left the room [very quickly]. ii. [His leaving of the room] was [very quick].

2. In some more recent lectures in 2019, Chomsky also figured out the role of MERGE and the workspace.

References Chomsky, N. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, N. 2013. Problems of Projection. Lingua 130: 33–49. Chomsky, N. 2017a. Arizona Lecture. March 2017 at the University of Arizona. Chomsky, N. 2017b. Generative Linguistics in the 21st Century: The Evidence and the Rhetoric. A Lecture at the University of Reading May 11, 2017 at the University of Reading. Chomsky, N., Gallego, A. and Ott, D. 2019. Generative Grammar and the Faculty of Language: Insights, Questions and Challenges. To appear in Gallego, A. and Ott, D. (eds.), Generative Syntax: Questions, Crossroads, and Challenges. Special Issue of Catalan Journal of Linguistics, 2019. Gallego, A. 2010. An L-Syntax for Adjuncts. In Duguine, M., Huidobro, S. and Madariaga, N. (eds.), Argument Structure and Syntactic Relations—A CrossLinguistic Perspective, 183–202. Amsterdam: John Benjamins. Hornstein, N. 2009. A Theory of Syntax—Minimal Operations and Universal Grammar. Cambridge: Cambridge University Press. Hornstein, N. and Nunes, J. 2008. Adjunction, Labeling, and Bare Phrase Structure. Biolinguistics 2: 57–86. Hunter, T. 2015. Deconstructing Merge and Move to Make Room for Adjunction. Syntax 18: 266–319. Jakielaszek, J. 2017. A Minimalist View on the Syntax-Semantics Relationship— Turning the Mind into a Snowflake. Frankfurt am Main: Peter Lang GmbH. Johnson, K. 2003. Towards an Etiology of Adjunct Islands. Nordlyd 31: 187–215. Nakashima, T. 2017. On Strict Cyclicity and Label: Toward Elimination of Late Merge. Explorations in English Linguistics (EEL) 31. Narita, H. 2014. Endocentric Structuring of Projection-Free Syntax. Amsterdam and Philadelphia: John Benjamins. Oseki, Y. 2015. Eliminating Pair-Merge. In Steindle, U., Borer, T., Fang, H., Pardo, A. G., Guekguezian, P., Hsu, B., O’Hara, C. and Ouyang, I. C. (eds.), Proceedings of the 32nd West Coast Conference on Formal Linguistics, 303– 312. Somerville, MA: Cascadilla Proceedings Project. Sportiche, D. 2019a. Neglect (or Doing Away with Late Merger and Countercyclicity). Available at Lingbuzz/002775. Sportiche, D. 2019b. Somber Prospects for Late Merger. Linguistic Inquiry 50: 416–424. Zeijlstra, H. 2019. Labeling. Selection and Feature Checking. To appear in Smith, P., Mursell, J. and Hartmann, K. (eds.), Agree to Agree: Agreement in the Minimalist Programme, 137–174. Berlin: Language Science Press.

Index

Note: Page numbers in italics indicate a figure and page numbers in bold indicate a table on the corresponding page. A-bar-movement 15, 64, 109, 162 accessible (to syntactic operations) 29, 59, 67, 78, 100 – 2, 104, 106 – 7, 111 – 13, 120 – 3, 126 – 9, 131, 148 – 9, 165 – 8, 170 – 1; see also inaccessible (to syntactic operations) Adger, D. 1, 2, 153n7 adjunct-argument dichotomy 9, 12; see also asymmetry between arguments and adjuncts adjunction 26 – 7, 39 – 40, 43 – 4, 45, 46 – 7, 53 – 4, 57, 70, 80, 82, 130 – 1, 134, 138, 142 – 3, 147, 163, 171 affix-hopping 141, 144 agree 33 – 4, 69, 113, 114, 116 – 17, 136 – 8, 164; see also agreement agreement 23, 28, 32 – 3, 37, 41, 99, 106, 109, 113, 158n38 Alexiadou, A. 13, 14, 45, 71 A-movement 15, 17, 33, 51, 56, 106, 108 anti-reconstruction see reconstruction architectural paradox 126, 128, 129 Arregi, K. & Pietrazko, A. 137 asymmetry see symmetry asymmetry between arguments and adjuncts 10, 16, 17, 44, 52, 56, 78 bare phrase structure 28 – 9; see also BPS Biberauer T. & Roberts, I. 137 binarity 35, 38, 114, 115, 169 binding 14 – 15, 31, 36 – 40, 50, 56, 47, 76, 77, 80, 109, 162 Biskup, P. 52, 56

Bobaljik, J. 42, 137 Boeckx, C. & Stjepanovic, S. 137 Bowers, J. 61 – 2, 91n10, 137 BPS 29, 57, 91n7 Carstens, V., Hornstein, N. & Seely, T. D. 143, 157n34 cartographic approach 43, 45 case 27, 28, 30, 37, 53, 77, 91n5, 158n38, 158n41 Cecchetto, C. & Donati, C. 137 Chomsky, N. 21 – 2, 24, 26 – 7, 29, 30, 32 – 3, 35 – 6, 53 – 4, 70, 71, 83, 98, 119, 131, 134, 137, 140 – 1, 164, 168 – 9 Chomsky, N., Gallego, A. & Ott, D. 4, 101, 165 – 7 Cinque, G. 41 – 2, 46 Citko, B. 137 computational efficiency 22 – 3, 31, 35 – 8, 52, 83, 107, 119, 165, 167; see also SMT copies 31, 36, 55, 82 – 3, 86, 90, 98, 100 – 3, 105 – 13, 120 – 1, 124 – 5, 127 – 8, 131, 135 – 6, 139, 143, 146, 149 – 50, 170 – 1 copy-theory of movement 31 Cormack, A. & Smith, N. 62 cross-linguistic variation 22, 28, 43, 109, 113, 116, 145, 154n16, 158n38 deVos, M. 137 Distributed Morphology 115, 141, 156n27 do-support 133, 144, 157n37

174 Index Epstein, S. D. 73 Epstein, S. D., Kitahara, H. & Seely, T. D. 4, 26, 33, 36, 56, 76, 83, 84, 137 Ernst, T. 42 – 5, 46, 71 extended projection 137, 145 Extension Condition 30, 36, 55, 134, 139, 142 externalization 22, 31, 41, 115 – 17, 138, 140 – 1, 143, 145, 146, 151, 157 – 8n37, 158n38 External Merge 31, 33, 97, 104, 110 – 11, 113, 114, 126, 128, 132, 137, 141, 149, 151, 168 feature-checking 30, 32 feature sharing 77, 98, 100, 106, 113 free labeling 110, 129 free syntax and projection-free syntax 28, 32, 38, 70, 72, 80 – 1, 85, 119, 136, 140, 151 – 2 Fukui, N. & Narita, N. 26, 86, 137 Full Interpretation 37 – 8, 72, 87, 111, 114 Gallego, A. 79 – 80, 83, 164 GB 26 – 8, 39, 47, 70, 163 German 44, 105, 109, 113, 130, 133, 135, 145 – 6 Government & Binding 1, 2, 3, 26, 28, 42, 46, 47, 70, 131; see also GB Harizanov, B. & Gribanova, V. 132, 137 Harley, H. 137 Hornstein, N. & Nunes, J. 57 – 9, 70, 71, 75 – 6, 86, 119, 163, 164 human language capacity 21 – 2, 29 Hunter, T. 75 – 6, 83, 164 inaccessible (to syntactic operations) 34, 36, 38, 52, 55, 60, 76 – 7, 79, 88 – 90, 97, 100 – 4, 106 – 8, 111 – 13, 122 – 3, 127, 148, 167 inclusiveness 29, 31, 36, 38, 44, 57, 114 Internal Merge 22, 31 – 3, 36, 37, 40, 56, 64, 69, 80, 83, 97 – 100, 102 – 5, 107, 109 – 12, 114, 120, 124, 127, 134, 136 – 41, 143, 146, 151, 165, 170 Irurtzun, A. & Gallego, A. 55 Ishii, T. 89 – 90

islands: adjunct island 5, 54, 56, 113, 120, 125 – 6, 129, 151, 170; island behavior 72, 79, 87, 90, 100, 110 – 11, 121, 170; islandhood 1, 16; subject island 120, 125, 151, 156n31; wh-island 123 – 6 Jakielaszek, J. 82 – 3, 90, 164 Johnson, K. 45, 170, 156n23 Koeneman, O. 137 Koeneman, O. & Zeijlstra, H. 1 labeling: LA 32 – 4, 79, 81 – 3, 86 – 7, 98 – 101, 104 – 5, 113; labeling algorithm 32, 75, 77, 78, 82 – 3, 85 – 6, 98, 100, 104; labeling conflict 33, 82, 98, 100, 102 – 3, 104, 110 – 11, 119, 122, 124 – 6, 128, 142 – 3, 147 – 9, 164 – 5; labeling of adjuncts 102, 110 LaCara, N. 137 language acquisition 20, 21, 28 Larson, R. K. 39 – 40, 46 Lasnik, H. 156n24 Last-Merge see Late-Merge Late-Merge 50 – 2, 69 – 70, 78 – 9, 115, 119 Law of Conservation of Relations see LCR LCR 36, 47n3, 83, 90, 101, 114, 115, 134, 139, 167, 169 linearization 31, 59, 137 – 8; see also externalization; linear order linear order 16, 23, 31, 34, 42, 43, 46, 59, 85, 79, 129, 138, 141 – 2, 145, 157n37 locality 36, 42, 125, 129, 138, 139, 143, 151 Matushansky, O. 137 McCloskey, J. 133 MERGE 165 – 8, 170 – 1 minimal computation 72, 84, 114, 126, 164 Minimal Search 33 – 4, 36 – 8, 97 – 104, 106, 110 – 12, 114, 119, 128, 130, 132, 142, 149 Move 30 – 2, 60, 114; see also Internal Merge Move-Alpha 26, 28, 144 Müller, G. 137 multiple adjuncts 8, 10, 11, 41, 58, 65, 74, 79, 80

Index  175 Nakashima, T. 77 – 8, 83, 164 Narita, H. 80 – 1, 83, 164 Neglect 162 No Tampering Condition 31, 77; see also NTC NTC 31, 36, 38, 51, 83, 101, 114, 115, 134, 139, 167, 169 Nunes, J. & Uriagereka, J. 45, 90 Oseki, Y. 76 – 7, 83, 164 Pair-Merge 3, 52 – 7, 59, 61, 69 – 70, 76 – 7, 80 – 4, 89 – 90, 115, 119, 131, 134, 136 – 9, 142 – 3, 148, 163 – 4, 167, 169 – 70 Parallel Merge 36, 79, 80, 136, 168 percolation 65, 69 Pesetsky’s paradox 40 Phase Impenetrability Condition 34; see also phases; PIC phases 34, 36, 38, 86 – 7, 106 – 7, 109, 123, 143 PHON (interface) 16, 22, 24 – 5, 31, 34, 36 – 7, 46, 69, 86, 88 – 9, 97, 102, 106, 108 – 13, 114, 116 – 17, 128 – 9, 136 – 8, 140 – 5, 147, 150, 158n38 PIC 35, 36, 38, 78 Platzack, C. 137 principle of uniformity 22 probe-goal 113, 139 Radford, A. 1 reconstruction 14 – 15, 17, 38, 52, 55, 76 – 7 recursive 4, 11, 24, 26, 31, 34, 116, 127, 166 Relational Definition of Projection 30 repetition 101, 107 – 8, 111 – 12, 120 – 1, 129, 171 Richards, M. 56, 84, 86 – 8 root 116 – 17, 131 – 3, 135 – 6, 144 – 6, 149, 151, 153n10, 155n20 Rubin, E. 55 – 6 Saito, S. 52, 56, 78 Schoorlemmer, E. & Temmerman, T. 137 scopal approach 45 Seely, T. D. 84 – 6, 97, 152n1, 154n15 segments 26 – 7, 30, 38, 44, 46, 57 Self-Merge 82, 90, 164 – 5 Self-Pair-Merge 89 – 90

SEM (interface) 22, 24, 25, 31, 37, 72 – 3, 88 – 9, 100 – 2, 104, 106 – 10, 112 – 13, 114, 116 – 17, 120 – 3, 126, 128 – 9, 138, 147, 166 – 7; see also PHON (interface) set formation 4, 5, 24, 29, 32 – 4, 70, 84 – 5, 88, 99, 102, 110, 114 – 17, 130, 139 – 40, 142, 146 Set-Merge 53 – 6, 61, 84, 86, 89, 119, 134, 136, 148, 164 Sideward Merge 136, 137, 168 Simplest Merge 4, 5, 26, 32, 34 – 5, 38, 70, 72, 85, 88, 96, 102, 108, 110, 113 – 14, 116, 118 – 19, 124, 126, 128, 130 – 1, 137 – 42, 146, 147, 148, 152, 162, 165, 169 – 71; see also free syntax and projectionfree syntax; MERGE; set formation Simplest Transfer 128 simplicity 22, 29, 35, 69 – 70, 151, 164 SIMPL(ify) 55 – 6, 84, 89 – 90 Single Event Condition 122, 129 SMT 22, 24, 31, 33, 38, 52, 70, 75, 83, 85, 88, 90, 119, 126, 130, 147, 162, 164, 168, 169 spec-head-agreement 28, 32 Sportiche, D. 52, 56, 90n2, 162 Stepanov, A. 50 – 2, 70, 119 Stockwell, R. 85 strong minimalist thesis 4, 69, 113; see also SMT successive cyclic movement 33, 52, 86, 102, 106, 109, 123 Surányi, B. 130, 137 symmetry 38, 99, 102, 119, 148, 169 – 70 Takita, K., Goto, N. & Shibata, Y. 98 theta 1, 9, 27 – 8, 31, 37, 53, 79, 108, 162 third factor 23, 26, 31, 33 – 8, 83 – 4, 88, 107, 113 – 15, 119, 128, 169, 171 three factors 23, 169; see also third factor Transfer 34, 36, 38, 55 – 6, 61, 76, 79 – 80, 83, 86, 88 – 90, 97, 100, 102, 104, 110 – 15, 119, 126 – 8, 147 – 8, 158n38, 165 – 7, 169 – 71; see also Simplest Transfer transparent adjuncts 77, 121 – 3, 129 Truswell, R. 56, 122 T-to-C movement 133, 145 T-v-X-“alliance” 145

176 Index UG 1, 3, 4, 19 – 24, 26, 28, 29, 30, 33, 34, 35, 38, 69, 74, 84, 88, 96, 113 – 17, 137 – 8, 166, 169 unbounded number of adjuncts 4, 10, 11, 16, 17, 24, 45, 72 uniformity (and diversity) 21 – 2, 116 – 17, 138, 140, 147 Universal Grammar 3, 20, 22; see also UG

VP-ellipsis 12, 17, 27, 45, 58 VP-preposing 12, 27, 27, 40, 58, 105 wh-movement 17, 50, 56, 77 – 8, 99, 103, 109 X-bar 1, 26 – 8, 38 – 9, 43, 45, 163 Zeijlstra, H. 61 – 8, 70, 119, 163, 164 Zwart, C. J.-W. 137