Nominalization in English

One prevalent assumption in the literature on nominalization is that the interpretation of external arguments in the prenominal position is governed by encyclopedic knowledge (see Marantz 1997, Harley and Noyer 2000). Thus, in the enemy’s destruction of the city, “the possessor can be interpreted as an agent/causer, based on our encyclopaedic knowledge about destroy” (Alexiadou, Anagnostopoulou, and Schäfer 2009:46). Since the possessive position (Spec,DP) is compatible with a range of semantic roles, it also supports the PATIENT interpretation of the internal argument, when it appears prenominally, as in the city’sdestruction.If the interpretation of the prenominal possessive is restricted by our knowledge about the world, then it is a puzzle why the prenominal argument of destruction in (1) must be a patient, while the prenominal argument of invasion in (2) can be either an agent or a patient. The same contrast holds for imprisonment in (3) and examination in (4).In what follows, I show that this question has not received a satisfactory explanation. I then propose that the mapping of arguments in nominals is akin to argument alternations in the verbal domain (Levin 1993) and that the observed differences between (1) and (2) should be explained by the lexical semantic properties of the head nouns.Anderson (1977), focusing on the distribution of internal arguments, proposes that they can be mapped into the prenominal position if they are affected: “changed, moved, altered in status or created” (1977: 15). This analysis predicts that the internal arguments of destruction and invasion can appear prenominally ((1b) and (2b)), but it does not explain why invasion allows its prenominal argument to be interpreted agentively, while destruction does not (e.g., (1a) vs. (2a)). The same criticism applies to the analysis by Doron and Rappaport Hovav (1991) that recasts the affectedness constraint in terms of event structure.In the literature on nominalization from the 1980s, it was assumed that the deverbal nominal inherits the argument structure of the verb and thus must realize its internal argument (see Chomsky 1981, Zubizarreta 1982, Lebeaux 1986, Roeper 1987). If the internal argument is unexpressed, as in (1a), the construction is ungrammatical under the agentive interpretation. This analysis implies that nominals such as invasion in (2a) are somehow different. This idea became a foundational assumption in Grimshaw 1990, where it is proposed that nominals that realize the internal argument (see (1b) and (2b)) and nominals that do not (see (2a)) belong to two different types, argument structure (AS) nominals and result (R) nominals, respectively. The two classes are associated with different clusters of properties. For example, ASnominals have an eventive interpretation and are mass nouns that cannot pluralize, while R-nominals refer to a result state of an action and are count nouns (see Grimshaw 1990 for the full list of diagnostics).Subsequent research has shown that the relation between argument realization and the properties identified by Grimshaw (1990) is not as straightforward as was previously thought. Roeper (1993), Borer (2003), Newmeyer (2009), and Alexiadou (2011) have challenged the assumption that eventive interpretation predicts the argument realization pattern. For example, Roeper (1993:204) points out that the objects of undergo and need in (5) and (6) refer to events, a hallmark of ASnominals, yet they show different patterns of argument realization: the nominals in (6) are grammatical without expressed internal arguments.While Roeper (1993:204) treats the argument realization pattern in (6) as a lexical idiosyncrasy, I would like to propose that the contrast between (5) and (6) is systematic and is due to the lexical semantics of the nominals.The analysis I propose is based on expert judgment, corpus study, and information extracted from Nomlex, a computational database for 1,000 nominals (Macleod et al. 1998).1 Focusing on deverbal nominals that surface with a single argument in the prenominal position, I identified three lexical classes: (a) nouns whose single prenominal argument must be a patient (PATIENT-dominant), as in (7); (b) nouns that require their single prenominal argument to be an agent (AGENT-dominant), as in (8); and (c) nouns that allow their single prenominal argument to be interpreted as agent or patient (SHARED-dominant), as in (9). A sample of each class is presented in the appendix.In what follows, I show that the choice of which event participant is mapped to syntax in single-argument constructions—patient in (7) and agent in (8)—correlates with Levin’s (1993) verb classes and argument alternations in the verbal domain, as summarized in table 1 in the appendix.The common property of the nominals whose single prenominal argument must be a patient is that they denote a change of state or a change of location. The change can be physical (destruction), psychological (humiliation), or political (democratization); it can be absolute (consumption and creation nouns) or gradual (escalation). The event can be bounded (destruction) or unbounded (education). What unifies this apparently diverse group is that the patient argument undergoes the change. Another relevant property is that the patient is constrained by the semantics of the head—it must be animate for humiliation (the victim’s humiliation) and inanimate for absorption (the chemical’s absorption)—but no such restrictions apply to the external argument.Verbal counterparts of PATIENT-dominant nominals belong to the following classes from Levin 1993: Change of State, Murder, Destroy, Amuse, and others (see table 1). Members of these classes participate in a number of alternations. Crucially, what these alternations have in common is that the patient argument is invariably preserved, while the agent either is not realized (causative and middle alternations) or is replaced with an oblique argument (instrumental subject and possessor subject alternations).(10) and (11) show that causative alternation verbs such as escalate and their nominal counterparts behave alike with respect to argument realization: if a single argument is mapped into the structure, it must be the patient argument.Additionally, verbal counterparts of many PATIENT-dominant nominals participate in the middle construction.While destroy does not participate in either the causative or the middle alternation, it shows the instrumental subject alternation in (14), from Levin 1993:239. In (14b), the agent is replaced with an instrument in the subject position, but the patient is unaffected. It is also the patient argument that is realized in the nominal construction in (15b).The possessor subject alternation in (16) achieves the same effect as the instrumental subject alternation, and the argument realization pattern in (17) parallels that in (15b).Ideally, these parallels in the verbal and the nominal domains should be explained from a lexical semantics perspective. A good candidate analysis is proposed by Rappaport Hovav and Levin (2012) for causative alternation predicates. The authors argue that these predicates have monadic roots; that is, they encode only the patient argument (see also Guerssel et al. 1985:54). This analysis is motivated by the fact that the internal argument of causative alternation verbs is thematically constrained, while the external argument is not—it can be an agent, an instrument, or a cause. Since the root lexically encodes the internal argument, in anticausative constructions such as (10b) it is the patient argument that is syntactically realized. This analysis can be straightforwardly extended to nominal counterparts of causative alternation verbs, such as escalation in (11), and, possibly, to other members of the PATIENT-dominant class.In AGENT-dominant nominals, the agent argument has a more prominent syntactic and semantic status with respect to the event denoted by the nominal. If a nominal denotes a change, it is usually the agent argument that undergoes change (e.g., the climber’s ascent vs. *the trail’s ascent). Moreover, the properties of the external argument are restricted by the semantics of the root. For example, love and other subject experiencer nominals require an animate subject, but their internal argument can be either animate or inanimate. These semantic properties have syntactic correlates. In the verbal domain, the corresponding predicates are represented by Admire, Correspond, Meet, and Inherently Directed Motion classes from Levin 1993 (see table 1). Members of these classes participate in alternations that have the following pattern: the agent argument is preserved, while the patient argument either is unexpressed or is replaced with an oblique argument. This is just the opposite of what we observed for the PATIENTdominant group.A good number of the corresponding verbal constructions for the AGENT-dominant class can surface without the theme argument, as in (18a). In the nominal domain, the internal argument, if realized, can often be expressed with an oblique preposition (e.g., John’s attackonthe hospital). If a single argument is expressed, it must be the agent argument (18b) (see also the unspecified object and preposition drop alternations for Inherently Directed Motion and Meet verbs).Moreover, Admire verbs participate in the possessor object alternation, shown in (19), adapted from Levin 1993:73. In this alternation, the agent is unaffected but the original patient argument is absorbed into the argument-expressing attribute. The single argument in the nominal construction in (20) must have an agentive interpretation.Finally, a sizable subset of AGENT-dominant nominals (Meet verbs and Correspond verbs) participate in the simple reciprocal alternation, shown for the symmetrical predicate collide in (21) and its nominal counterpart collision in (22).Even though both event participants of symmetrical predicates appear to have equal status (see Levin 1993), the alternative mapping options reflect the privileged status of the agent.2 It is the agent that is mapped into a more prominent syntactic position in (21b) and (22b), while the patient argument is realized with an oblique preposition or dropped, as in (22c). The latter option is available in nominal constructions only.The observation that the agent arguments of some predicates are structurally and semantically more prominent than their patients is by no means new, and has been reflected in a number of theoretical proposals (see Hopper and Thompson 1980, Van Valin 1990, and Nunes 1993 on nominalizations). Levin (1999) proposes that patient arguments of simple noncausative verbs are not encoded in the event template and that their mapping to syntax is optional (see also DiDesidero 1999 on the extension of this analysis to psych verbs such as love and Rappaport 1983 for a similar analysis). In a similar spirit, Alexiadou (2011) proposes that the theme argument of Admire-type nominals is introduced vP-internally as a PP rather than as a DP. The nominal data presented above are consistent with these proposals, and suggest that the similarities between the nominal and the verbal domains have a lexical semantic basis.SHARED-dominant nominals are more flexible in that the possessive argument can be either agent or patient. I would like to suggest that this flexibility is due to the fact that the nominals belonging to this class lack a most salient event participant; that is, unlike the PATIENT and AGENT-dominant nominals, they do not lexically identify one argument as more prominent. This is perhaps due to the fact that arguments of these nominals are neither prototypical agents (they cannot be characterized in terms of the transmission of force (Croft 1991); e.g., the enemy’s attack vs. the scientist’s classification) nor prototypical patients (there is no change in the “material integrity” of the object (Hale and Keyser 1987); e.g., the city’s destruction vs. the object’s representation). In fact, a considerable number of nominals belonging to this group refer to cognitive activities or denote events that can be characterized by low transitivity (see Hopper and Thompson 1980). Verbal counterparts of these nominals are most notably represented by Levin’s (1993) Characterize, Judgment, Sight, and Assessment classes (see table 1).Invasion, of course, is different from cognitive activity predicates in that it seems to be a rather prototypical AGENT-PATIENT nominal akin to destruction. However, since the external argument of invasion undergoes change of location in the direction of some goal, it can also be viewed as a THEME-GOAL nominal (see Jackendoff 1987). What we know about themes and goals is that in the verbal domain the arguments associated with these roles can swap syntactic position, if the corresponding verb participates in the dative or the locative alternation.Such a syntactic interchangeability is also available for arguments of invasion, even though in this case, as shown in (24), the alternation affects the external and the internal arguments rather than the two internal arguments as in (23).If this proposal is on the right track, I expect that the discourse and information structure principles would affect the mapping of arguments in SHARED-dominant nominals, similar to the choice of word order in the English dative alternation and locative alternation constructions (see Arnold et al. 2000, Bresnan et al. 2007).The results of the empirical study presented in this squib suggest that encyclopedic knowledge alone is not sufficient to explain the mapping of arguments in nominals (see also Sichel 2010). I propose that the observed pattern depends on the lexical semantics of nominals and correlates with Levin’s (1993) verb classes. As pointed out by a reviewer, these results are consistent with the assumptions of the Distributed Morphology framework: if the nature of the category-less root determines the mapping of arguments, then it is expected that the mapping pattern will be consistent across the nominal and the verbal domains. Yet how encyclopedic knowledge interacts with the lexical semantics of the root still awaits a systematic explanation.Members of the nominal classes identified in the squib and the corresponding verbal classes from Levin 1993 are listed in table 1.PATIENT-dominant class:(Other) Change of State: corrosion, democratization, diffusion, disintegration, evaporation, explosion, intensification, stabilization, unionization; Amuse: alienation, consolation, demoralization, entertainment, humiliation, intimidation, pacification, satisfaction, terrorization; Murder: assassination, elimination, execution, slaughter; Destroy: annihilation, demolition, destruction; Banish: extradition, removal; Remove: eradication, expulsion; Sight: redemption, retrieval; Appoint: adoption; Begin: termination; Characterize: establishment; Complete: initiation; Create: creation; Put: installmentAGENT-dominant class:Admire: admiration, appreciation, fear, hatred, love, resentment, respect, tolerance; Correspond: collision, competition, cooperation, quarrel, rendezvous; Meet: battle, consultation, fight, visit; Inherently Directed Motion: ascent, descent, entry; Advise: admonition, warning; Avoid: avoidance, evasion; Chase: chase, pursuit; Conjecture: assertion, suspicion; Mix: fusion, merger; Say: proclamation, suggestion; Chitchat: conversation; Contiguous Location: domination; Declare: declaration; Exchange: trade; Judgment: congratulation; Want: desire; Amuse: reassurance, temptation; Sight: extortion; Characterize: recollectionSHARED-dominant class:Characterize: certification, characterization, classification, description, identification, portrayal, recommendation, representation, treatment; Sight: examination, inspection, investigation, observation; Judgment: celebration, compensation, impeachment, punishment; Assessment: analysis, assessment, evaluation; Amuse: encouragement, provocation, stimulation; Advise: instruction; Leave: abandonment, desertion; Transfer of Message: demonstration, explanation; Amalgamate: introduction; Performance: performance; Separate: segregation

Argument (complex analysis) · Interpretation (philosophy) · Linguistics · Nominalization · Noun · Possessive · Semantic property · Computer Science · Language, Discourse, Communication Strategies · Language, Metaphor, and Cognition · Philosophy · Syntax, Semantics, Linguistic Variation

• Case Assignment and Argument Realization in Nominals

  Open Access•Anastasia Smirnova, Ray Jackendoff•Language•2017

  Add to list

• Case and the Event Structure of Nominalizations

  Luis López•Linguistic Inquiry•2018

  Add to list

• Connecting the typology and semantics of nominal possession

  Open Access•Albert Ortmann•Morphology•2018

  Add to list

• English verb classes and alternations

  Beth Levin•English verb classes and…•1993

  Add to list

• Lectures on Government and Binding

  Noam Chomsky•Lectures on Government and Binding•1993

  Add to list

• Transitivity in Grammar and Discourse

  Paul J Hopper, Scott A Thompson et al.•Language•1980

  Add to list

• Semantic Parameters of Split Intransitivity

  Robert D Van Valin•Language•1990

  Add to list