The language of an inanimate narrator

Abstract We show by means of a corpus study that the language used by the inanimate first person narrator in the novel Specht en zoon deviates from what we would expect on the basis of the fact that the narrator is inanimate, but at the same time also differs from the language of a human narrator in the novel De wijde blik on several linguistic dimensions. Whereas the human narrator is associated strongly with action verbs, preferring the Agent role, the inanimate narrator is much more limited to the Experiencer role, predominantly associated with cognition and sensory verbs. Our results show that animacy as a linguistic concept may be refined by taking into account the myriad ways in which an entity’s conceptual animacy may be expressed: we accept the conceptual animacy of the inanimate narrator despite its inability to act on its environment, showing this need not be a requirement for animacy.


Introduction
Animacy is a major distinction in human cognition (Dahl 2008), for obvious reasons: simply put, it matters if that thing you saw from the corner of your eye is a bear, a rock, or another human being. By dividing the world around us into animate and inanimate entities we can attribute mental states and biological processes to them that allow us to predict the behaviour of other entities (Szewczyk & Schriefers 2011), which is crucial to survival. Naturally, as language should allow us to adequately refer to entities in the world, the animacy distinction is also one of the basic principles behind language, "so pervasive in the grammars of human languages that it tends to be taken for granted and become invisible" (Dahl & Fraurud 1996:47). The influence of animacy has been attested cross-linguistically at virtually every level of language, from word order, grammatical function, case and voice, to the choice of referential expressions (e.g., Yamomoto 1999;Dahl 2008;Malchukov 2008;Prat-Sala & Branigan 2000;van Bergen 2011;Vogels et al. 2013).
However, whilst the vast majority of psycholinguistic and typological research has equated linguistic animacy with biological animacy, looking for example at the distinction between rocks and hikers (e.g. Mak, Vonk & Schriefers 2006) or men versus mountains (cf. Hale 1973), animacy in language is rather more flexible. Consider examples (1) and (2)1: (3) Had ik de gave van het woord, dan zou ik nu beschrijven hoe het voelt om eindelijk een doek te zijn, een doek met maten, een afgemeten, met het vlijmendste stanleymes afgesneden, onherroepelijk strak stuk linnen gespannen om een stevig raam met latten van zes en maar liefst drie punt zes dik, met spieën en met in zijn rug een kruis. (p.9) "If I had the gift of speech, I would now describe what it feels like to finally be a canvas, a canvas with dimensions, a piece of linen that has been measured out, cut with the most razorish Stanley knife and irrevocably stretched tight around a sturdy frame with six-centimetre stretchers no less than threepoint-six thick, with wedges and a cross at the back." (p.7) As can be concluded from the utterance in (3), non-human narrators are conscious, they perceive, interpret, and experience things, e.g., they can remember, have feelings, get emotional etc. This means that they possess a mind and they have a life, which makes them partly human in the eyes of their readers. This is expected to influence readers' perception of these non-human narrators and their degree of identification or empathy with them. However, despite the fact that non-human narrators have human traits in varying degrees, they often deviate from human narrators in how they perceive and interpret events (Bernaerts et al. 2014: 74). Also, the way they participate in the events may differ. This holds for inanimate narrators even more than for narrating animals, as the former deviate the most from human beings. Bernaerts et al. (2014) argue that non-human narration always gives rise to a mix of distancing and identification effects in readers, or defamiliarization and empathy. Given that animacy plays a major part in grammar and discourse (cf. Dahl 2008;de Swart et al. 2008), we expect that having an inanimate narrator will affect the language that is used. More specifically, the question addressed here is whether the observed mix between human and non-human features of an inanimate narrator is reflected in the language attributed to them. Bernaerts et al. (2014) note that nonhuman narrators vary greatly in their physical and psychological features and thus can be more or less humanlike. For example, animals are clearly more closely related to human beings than cars. Cars, in turn, may be perceived as more closely related to human beings than objects that do not move, such as paintings. For this study we have chosen to investigate the language of an inanimate narrator, namely a canvas that in the course of the story becomes a portrait, and to compare it to the language of a common human narrator. The painting that is the first person narrator in Specht en zoon cannot move by itself and does not communicate with the human characters nor with other inanimate entities in its environment. It only communicates with its readers by telling them the story of its owner, a portraitist who is asked by a rich man to make a portrait of his dead son. We compared this novel to another novel written by the same author, De wijde blik (1992), which features a human first person narrator.
Before turning to the methodology and results of the corpus study in Section 3, we will discuss a number of grammatical features that are known to relate to animacy, viz., the use of pronouns, grammatical function, voice, thematic (or semantic) roles, and types of verbs. On the basis of this, we will formulate our hypotheses on the language used by the inanimate narrator in comparison to the language used by the human one.

Pronouns
Animacy has been shown to influence the choice between pronominal or nominal realization of a referent. Human or animate entities are more often realized as personal pronouns than inanimate ones (Dahl 2008). Dahl (2008) uses the term egophoric reference for first and second person pronouns as well as for (second or third) person generic reference, as in You only live once (second person generic reference) or One never knows (third person generic reference), which typically also includes the speech act participants. Dahl (2008: 143) claims that reference to animate noun phrases in spoken discourse is egophoric in the majority (about 60%) of cases. Indeed, as Dahl (2008: 143) puts it: "[w]e speak more about ourselves than about others." Vogels et al. (2013) found it is not only lexical animacy that triggers the use of personal pronouns but perceptual animacy as well, i.e., inanimate entities that move in an animate way. Because we investigate first person narrators in our study, which are always referred to by first person pronouns, we cannot compare differences between the two novels in type of reference to the narrators. The inanimate narrator in one novel uses first person pronouns to refer to itself, while the human narrator in the other novel also uses first person pronouns to refer to himself. We may expect that the inanimate narrator does not talk about itself as much as the human narrator, but this is not necessarily due to its animacy. The first person human narrator in De wijde blik is also the main character in the story, but this is not the case for the first person inanimate narrator in Specht en zoon. Hence, we cannot make any predictions about the difference in the use of first person reference per se between the two novels, but we can make predictions about differences that derive from the use of first person pronouns for either an inanimate or a human referent. Comrie (1989) states that subjects of transitive clauses are mostly animate and definite while objects are lower in animacy and definiteness (see also Aissen 2003). Several corpus studies have shown that in the vast majority of cases in Swedish, Norwegian, and Dutch the subject outranks the object in animacy in transitive sentences (Dahl & Fraurud 1996;Øvrelid 2004;Bouma 2008). For example, Bouma (2008) in a corpus study of spoken Dutch finds that subjects are overwhelmingly animate (96%), whereas direct objects are for the greater part inanimate (90%). As for transitive sentences, subjects outrank the object in animacy in about 86%, whereas they rank equally in about 13%. Thus, sentences in which the object outranks the subject in animacy constitute only 1% of the transitive sentences (Bouma 2008: 257). Fauconnier (2011) claims on the basis of a 200-language sample that inanimate subjects are unexpected to such a degree that in a considerable number of languages inanimates cannot be used as subjects of transitive clauses at all. The subject-object asymmetry in animacy has also been shown to play a role in language processing and production (e.g., de Hoop & Lamers 2006;Branigan et al. 2008;Bornkessel-Schlesewsky & Schlesewsky 2009;Lamers & de Hoop 2014), and also in children's interpretation and production of transitive clauses (e.g., Hendriks et al. 2005;Hogeweg & de Hoop 2010;Cannizzaro 2012).

Subjects and objects
The person and animacy hierarchies are often conflated (e.g. Comrie 1989;Yamamoto 1999), with first person representing one extreme and inanimate objects representing the other. First person pronouns, in everyday speech, can only refer to human beings. The apparent paradox makes Specht en zoon's use of a first person pronoun that refers to an inanimate entity quite an interesting case.
Cross-linguistically, pronouns typically fulfill the grammatical function of subject (Siewierska 2004;Bouma 2008), and it is 'marked' for a first person pronoun to fulfill the function of object (Aissen 1999). However, when a first person pronoun refers to an inanimate entity, this may alter our expectation, because inanimate subjects are very infrequent, as pointed out above. The following examples illustrate the use of the first person inanimate narrator in the novel Specht en zoon as the object of a transitive verb: (4) Hij komt me halen, het kan niet anders of ik word in het vuur geworpen. (p. 5) "He's coming to get me. There's no doubt anymore. He's going to throw me on the fire." (p. The fact that the first person inanimate narrator is used as a direct object in the above sentences is in accordance with the fact that it only plays a 'passive' part in the story, i.e., as an observer rather than as an actor. Other characters may act upon it, come and get it, carry it off, put it somewhere, but not the other way around. Therefore, our hypothesis is that the proportion of subjects among the first person pronouns referring to the inanimate narrator in the novel Specht en zoon is lower than the proportion of subjects among the first person pronouns referring to the human narrator in De wijde blik.

Active and passive sentences
Note that the transitive verbs halen 'get', wegdragen 'carry off', and neerzetten 'put' in (4)-(6) require a moving and acting subject. Since the inanimate narrator cannot do the things that are expressed by these verbs, it does not come as a surprise that it does not take the function of subject. However, an alternative way of expressing an event in which the first person pronoun has a Patient role is to use a passive sentence. Clearly, the advantage of using a passive construction in these cases is that the first person pronoun ends up as the subject. In (4) above, the second part of the sentence (repeated in 7) contains a passive construction in which the first person narrator has become the subject (although the English translation uses an active sentence in which it is the object).

(7) [I]k word in het vuur geworpen.
I get in the fire thrown "He's going to throw me on the fire." Two more examples of passives are given in (8) and (9). Whereas in (8) the passive construction is translated as active in English, the English translation of (9) also uses a passive.
(8) Twee weken later ben ik opgehaald. (p.9) two weeks later am I picked.up "Two weeks later he came back to pick me up." (p.7) (9) Hij schildert inderdaad mensen, mijn schepper, alleen maar he paints indeed people my creator only ptcl mensen, dat ontdekte ik al snel, ofschoon ik people that discovered I already quickly although I om te beginnen in een hoek van zijn atelier ben comp to start in a corner of his workshop am neergezet, met mijn voorkant tegen een tamelijk koude muur. put with my front against a rather cold wall (p.11) "He does paint people, my creator -people only. I soon discovered that, even though I was plonked down at first in a corner of his studio with my front against a fairly cold wall." (p.9) While first person passive Agents are quite uncommon across languages, first person passive Patients are to be expected (Aissen 1999). Aissen (1999: 689) argues that when the Patient is first person and 'prominent' in the discourse, the use of a passive construction is preferred in a language such as English. This also seems to hold for Dutch. Cornelis (1997) shows that people do not identify with (or even distance themselves from) the Agent in a Dutch passive construction. Therefore, we hypothesize that due to its first person inanimate narrator there will be more passive sentences in the novel Specht en zoon than in the novel De wijde blik.

Thematic roles
In passive constructions the first person narrator will have the grammatical function of subject. This conflicts with our previous hypothesis that the first person narrator will take the role of subject relatively less often because it refers to an inanimate entity. Thus, we predict a lower frequency of subjects that refer to the inanimate narrator on the one hand, but a higher frequency of passive sentences in which the inanimate narrator is the subject on the other. If the first person inanimate narrator ends up as the subject of an active sentence less often than the first person human narrator, but at the same time more often as the subject of a passive sentence, then in the end the proportion of subjects may be the same for the two types of narrators. In order to tease apart the effects of the two hypotheses, we decided to also investigate the thematic roles assigned to the first person pronouns. Clearly, since the painting cannot act upon others but can be acted upon, we expect it to have the role of Patient more often than the human narrator, independently of its grammatical function. This is illustrated by sentences (4)-(9) above, which featured the painting as a Patient. Dowty's (1991) argument selection principle crucially distinguishes Agents from Patients in terms of their typical properties. The argument with the highest number of proto-Agent properties will end up as the subject of a transitive clause. Proto-Agents are volitionally involved in (or in control of) the event or state expressed by the verb, they are sentient, they are the causers of events or of changes in another participant, they move (relative to another participant), and they exist independently of the event. Proto-Patients undergo a change of state, they are incremental themes, they are causally affected by another participant, they are stationary relative to the movement of another participant, and/or they do not necessarily exist independently of the event. Primus (2012) argues that nearly all proto-Agent properties entail an animate Agent. By contrast, there is no animacy entailment for proto-Patients, even though individual transitive verbs, such as kill, may select an animate object. In John broke the cup, the object has the proto-Patient properties of undergoing a change of state, and being causally affected by the Agent. These proto-Patient properties are less clear in John touched the cup, except that the cup does not move, and for the objects of search, follow and await (cf. Malchukov 2005). Because of differences in proto-Patient properties, we have decided to use the more general label Theme instead of Patient, subsuming the Patient role under Theme. Theme is also used to label the role of the co-participant of an Experiencer in a transitive sentence, as in John likes the cup or John saw the cup. Subjects of intransitive verbs that are not Agents or Experiencers are also labeled Theme in our annotation, such as the subject of the posture verb staan 'stand' in (10): (10) Ik sta op de ezel en heb alleen het ergste te verwachten. (p.5) "I am on the easel and can only expect the worst." (p.1) Similarly, subjects of existential, locational, as well as nominal and adjectival predicates are labeled Themes, as in (11) ;Verhoeven 2014;2015). The co-participant of an Experiencer subject or object can have the role of a Theme or a Stimulus, dependent on the type of verb and the animacy of the argument (Lamers & de Hoop 2014;Verhoeven 2014). For instance, in the sentence John frightens me, John can actively or deliberately frighten me, whereas in the sentence The weather frightens me, the subject the weather can never play such an active role. We have decided to ignore such differences between Themes and Stimuli and we have labeled all co-arguments of Experiencers Themes. Contrary to the general pattern, (animate) Experiencer objects are more often pronominalized than (inanimate) subject Themes (Verhoeven 2015). The inanimate narrator in Specht en zoon can have the role of an Experiencer, as illustrated in (12) and (13), where it is the Experiencer subject of horen 'hear ' and herinneren 'remember', and in (14) where it is the Experiencer object of verontrusten 'unsettle'.
(12) Zulke dingen heb ik ze soms horen zeggen en ze begrepen elkaar. (p.6) "That's the kind of thing I heard them say, and they seemed to understand each other." (p.2) Because the first person pronoun in Specht en zoon refers to an inanimate entity, we hypothesize that it will be more suited to be a Theme than to be an Experiencer. Therefore we predict that the inanimate narrator in Specht en zoon will more often have the role of a Theme and less often the role of an Experiencer than its human counterpart in De wijde blik. The difference between the narrators in having the Experiencer role is expected to be less than the difference in taking up the Agent role, however, because whilst the painting cannot move by itself, it is conscious and thus able to experience things.
To summarize, based on established animacy patterns, we do not expect the inanimate narrator to have the role of Agent or Experiencer as often as the human narrator, despite the animate endowments it might share. Instead, we expect it to have the role of Theme more often.

Verb classes
Thematic roles are generalizations over arguments of particular verbs (Lestrade 2010). For our annotation we have decided to not only annotate for thematic role, but also for verb class, since there is not necessarily a one to one mapping between the two. For example, subjects of perception verbs have the role of Experiencer, while subjects of cognition verbs can also have the role of Agent. A transitive action verb takes as one of its arguments a Theme, while a transitive cognition or psych verb takes as one of its arguments a Theme as well.
A difference can be made between cognition verbs such as think or mean (where the subject is an Agent like the Agent subject of tell or say), as illustrated in (15) and (16), and cognition verbs such as know in which the subject is an Experiencer, also illustrated in (16) Another distinction can be made between perception verbs such as see or hear of which the subject is an Experiencer, and perception verbs such as look or listen of which the subject is more actively paying attention and thus has the role of an Agent (cf. Viberg 1983;Malchukov 2005). Whitt (2009) calls verbs such as look and listen 'subject-oriented agentive perception verbs', verbs such as see and hear 'subject-oriented experience perception verbs', and verbs such as sound 'object-oriented perception verbs'. Note that the English verbs smell and taste can have all three different readings. We have decided to call see and hear 'sensory verbs', and look and listen 'action verbs' in our annotation. Examples (17) and (18) illustrate the uses of these verbs with the inanimate narrator as an Experiencer subject in (17) and as an Agent subject in (18).
(17) Ik zag eigenlijk alleen maar hoe klein hij was en onopvallend. (p.24) "All I saw was how small and nondescript he was." (p.28) (18) Ze zijn een wandeling gaan maken door het bos om Nimmerdor en voor het eerst kon ik op mijn gemak de wereld in kijken, wat zeggen wil de glazen schuifpui door de tuin in. (p.24) "They went for a walk through the woods around Withernot, and for the first time I was able to look out into the world at my leisure, which is to say through the sliding doors and into the garden." (p.28) The use of a verb such as klinken 'sound' was annotated as 'property'. An example is given in (19). Note that this type of verb is called 'object-oriented' by Whitt (2009)  in De asielzoeker than in the other two novels, although the number of action verbs still exceeded the number of cognition and state verbs. They concluded that a more fine-grained analysis of types of verbs and the context in which the verbs occur would be necessary to confirm and explain the impression that De asielzoeker is more 'static' than the other two novels, due to its passive third person main character. Because the inanimate first person narrator in Specht en zoon is also passive and observing rather than dynamic, our hypothesis is that there will be relatively fewer action verbs and more other types of verbs in Specht en zoon than in De wijde blik.

Hypotheses for the corpus study
In the previous subsections, we have come to the following hypotheses concerning the language of the first person inanimate narrator in the novel Specht en zoon, compared to the language of its human counterpart in De wijde blik: (i) The inanimate first person narrator of Specht en zoon will have the grammatical function of object more often than the human first person narrator of De wijde blik; (ii) There will be relatively more passive sentences in Specht en zoon than in De wijde blik; (iii) The inanimate first person narrator of Specht en zoon will have the thematic role of Theme more often and the thematic roles of Agent or Experiencer less often than the human first person narrator of De wijde blik; (iv) There will be proportionally fewer action verbs and more other types of verbs in Specht en zoon than in De wijde blik.

The corpus
We used digitized versions of the two Dutch novels by author Willem Jan Otten: De wijde blik (1992, 177 pages) and Specht en zoon (2004, 142 pages). From these novels we extracted all sentences containing a first person pronoun (i.e. ik 'I', me/mij 'me' and mijn 'my'). For Specht en zoon this resulted in 1312 sentences; for De wijde blik in 2395.

Annotation
As discussed in Section 2, we are interested in the relative distribution of grammatical function, voice, thematic role and verb type as they are used with first person pronouns referring to the narrator of the respective books. We annotated the corpus on these features for every sentence containing a first person pronoun. An overview of the features used and their possible values can be found in Table 1. Speaker was annotated only when the first person pronoun did not refer to the narrator, e.g., in the case of reported direct speech. These cases were excluded from further analysis since we were interested not in the distribution of grammatical functions, verb types and thematic roles in the novels per se, but in these distributions in relation to the first person narrator. Grammatical function was annotated only if the pronoun was an argument of the verb (i.e. a subject, (in)direct object or prepositional object). We excluded adjuncts because we were interested specifically in the role the inanimate narrator takes up in the argument structure of verbs, in relation to their grammatical function and thematic role. Elliptical sentences were also excluded: whilst in the majority of cases unambiguous reconstruction of the argument structure was possible, we felt it best to look only at fully-formed sentences as this allows for a fully bottom-up reconstruction of our findings. Voice is self-explanatory, taking the values of 'active' or 'passive'. We annotated for the most common thematic roles, subsuming Patient, Stimulus, and Theme under 'Theme', as discussed in Section 2.4. We distinguished frequent verb types along the lines discussed in Section 2.5. The feature took the value [Verb], a repetition of the lexical verb itself, when the verb proved hard to classify.
Each novel was annotated by two of the authors. The first 400 sentences were annotated by the authors jointly, after which disagreements were resolved and the authors continued individually.

Results
Application of the exclusion criteria above resulted in 863 sentences for Specht en zoon and 1791 sentences for De wijde blik. Chi-square tests of independence were carried out on all annotated features between novels. A significant interaction effect of grammatical function by novel was found: the proportion of firstperson objects in De wijde blik (16.3%) was lower compared to that in Specht en zoon (19.8%), χ 2 (1) = 4.85, p < 0.05. We also found a significant effect of voice in the expected direction, but in general the occurrence of passive voice was rare in both novels (0.7% versus 2.3% respectively, χ 2 (1) = 10.63, p < 0.01). The distribution of thematic roles between novels can be found in Table 2. There was a significant interaction effect of Thematic role by novel: χ 2 (5) = 250.3, p < 0.01. The distribution found is markedly different between the two novels: the Agent role dominates for De wijde blik whilst the contribution of the Agent role in Specht en zoon concedes to the predominance of the Experiencer and Theme roles. This is more clearly illustrated in Figure 1, based on Table 2. For the analysis of verb type, we distinguished between action, cognition, sensory and other verbs. 'Other' thus included existential, copular, posture, property, unaccusative, and unclassifiable verbs, as these categories were not very frequent separately, nor very informative. The distribution of verb types between the two novels is given in Table 3.
Brought to you by | Radboud University Nijmegen Authenticated Download Date | 6/24/19 2:50 PM  There was a significant interaction effect of verb type by novel: χ 2 (3) = 190.1, p < 0.01, with Specht en zoon containing proportionally more cognition and sensory verbs and De wijde blik containing proportionally more action verbs.

Discussion
We started our corpus study with the following four hypotheses: (i) The inanimate first person narrator of Specht en zoon will more often have the grammatical function of object than the human first person narrator of De wijde blik; (ii) There will be relatively more passive sentences in Specht en zoon than in De wijde blik; (iii) The inanimate first person narrator of Specht en zoon will more often have the thematic role of Theme and less often the thematic roles of Agent or Experiencer than the human first person narrator of De wijde blik; (iv) There will be proportionally fewer action verbs and more other types of verbs in Specht en zoon than in De wijde blik.
Brought to you by | Radboud University Nijmegen Authenticated Download Date | 6/24/19 2:50 PM With regard to grammatical function, our results reflect the cross-linguistic preference for pronouns to fulfill the grammatical role of subject. The contribution of pronouns in the object role is significantly higher for Specht en zoon compared to De wijde blik, confirming our first hypothesis. As discussed in 2.2, the animacy distribution in transitive sentences is quite clear, i.e. inanimate subjects are highly infrequent. The relative distribution of subject and object roles between the two novels shows only relatively minor differences, however (19.8% versus 16.3% object roles, respectively). Specht en zoon's first person inanimate narrator presents an interesting case since it highlights the tension between avoiding inanimate subjects on the one hand and the subject preference of pronouns on the other. Our results show that this tension is resolved firmly in favour of the pronoun, indicating that whilst the inanimate narrator is not a typical animate, it is endowed with sufficient animate properties to warrant reference by means of a first person subject pronoun.
Turning to voice, our second hypothesis was based on the observation that passivisation is an excellent linguistic strategy to allow the first person pronoun to fulfil the subject role, even when the referent is incapable of carrying out the action denoted by the verb. Although the hypothesis was indeed verified, the percentage of passive sentences remains very low in both novels and the difference is insufficient to explain the subject-pronoun preference of the inanimate narrator.
The third hypothesis was about the distribution of thematic roles. As predicted, the inanimate narrator in Specht en zoon is less often an Agent and more often a Theme than the human narrator in De wijde blik. This is what we expected on the basis of animacy. The first person inanimate narrator is never the subject of the verb lopen 'walk' for example, except once in the idiomatic gevaar lopen 'be (lit. walk) in danger', but here it is not an Agent. When the verb lopen 'walk' is used in the novel, it is always somebody else, i.e. a human being, who is walking. Two examples are given in (20) and (21) What these two sentences also illustrate is the fact that the inanimate narrator easily fulfills the role of Experiencer. It can notice and see the walking of somebody else in (20) and (21). This is also necessary for the narrator of a story. Clearly, no story would be possible in case of a narrator who could not observe and report on events going on in the story world. This explains the fact that the part of our third hypothesis concerning the Experiencer role is falsified. It is not the case that the narrator in Specht en zoon has the role of Experiencer less often than the one in De wijde blik. In fact, it is the other way around. The inanimate narrator takes up the thematic role of Experiencer relatively more often than its human counterpart. In terms of verb types, the inanimate narrator in Specht en zoon combines with cognition verbs (such as merkte 'noticed' in (20) and sensory verbs (such as zag 'saw' in (21)) relatively more often than the human narrator in De wijde blik. We believe that the verification of our hypothesis that the inanimate narrator is an Agent less often than the human narrator goes hand in hand with the falsification of our hypothesis that the inanimate narrator is an Experiencer less often. Precisely the fact that the inanimate narrator in this story does not move by itself and therefore cannot easily be a true Agent, makes it a passive narrator that merely observes and experiences things (and comments upon them), hence a true Experiencer (cf. Fagel et al. 2012). In other words, the fact that the inanimate narrator is a painting that cannot move but still has to narrate the story, triggers its Experiencerhood which is then reflected in the language it uses. As to the degree of defamiliarization or 'unnaturalness' experienced by readers (Alber 2016;Bernaerts et al. 2014), we assume that part of the explanation lies in the fact that readers identify themselves less with, or empathize less with, 'passive' narrators such as the painting in Specht en zoon, but this may also vary among readers (cf. Nijhof & Willems 2015).
Brought to you by | Radboud University Nijmegen Authenticated Download Date | 6/24/19 2:50 PM De Swart and de Hoop (2018) argue explicitly for a distinction between conceptual and grammatical animacy. They claim that conceptual animacy is a gradient notion which may be informed by a myriad of relevant features to greater or lesser extent, but that this is ultimately resolved to a binary distribution of grammatical animacy. This was clearly the case for our inanimate narrator: conceptually, the painting is not a prototypical animate entity, lacking agentive properties, but the distribution of pronoun use, grammatical function and voice indicates it is treated as such linguistically. In this sense the inanimate narrator presents an empirical basis similar to that of inanimate referents endowed exclusively with agency (i.e. the distinction between natural forces and tools, cf. Lowder & Gordon 2015), and thus provides a complementary opportunity to separate a thematic property from animacy in e.g. processing studies.
Our results confirm that the biological or ontological status of entities is ultimately not very informative in explaining grammatical patterns associated with animacy. The ontologically inanimate painting was in fact closer to a traditional human narrator than to an inanimate object in its grammatical distribution. Clearly, conceptual animacy is the more informative notion here. We are certainly not the first to argue that conceptual animacy drives linguistic expression (cf. Fraurud 1996;Yamamoto 1999), but the extreme case of the inanimate narrator presented here allows for a more nuanced view as to which conceptual features contribute and to what extent. Specifically, the inanimate narrator presented clear differences with a traditional human narrator at this conceptual level: we noted that the agency of the painting was severely limited, resulting in a conceptually animate entity based almost exclusively on Experiencerhood.

Conclusion
Fundamental to human cognition, the distinction between living and non-living entities is attested crosslinguistically in virtually all levels of language, and is relatively well understood. Expressions referring to animate entities may be realised with first person pronouns, since only animate entities can refer to themselves using language. Expressions referring to animate entities show an overwhelming preference to be the subject of a transitive clause. Agents are predominantly animate, Experiencers are necessarily animate.
First person inanimate narrators in literary fiction turn these notions on their head. We presented results from a corpus study comparing the inanimate narrator in Specht en zoon to an animate, human narrator in De wijde blik by the same author. Our study showed remarkable similarities in the grammatical distributions of both narrators rather than a clear animate-inanimate split. This indicates we should consider animacy in language not as reflecting a binary, biological property of the referent but rather as reflecting conceptual animacy, which is more fluid and may be expressed along a continuum of degrees of animacy. This is most noticeable in the distribution of thematic roles and verb types: whereas the human narrator in De wijde blik has a clear preference for the Agent role, associated with action verbs, the inanimate narrator in Specht en Zoon is predominantly an Experiencer, observing and commenting on the world around it.