Lip̄nē ‘in the face of’: A Locative Preposition with a Threatening Connotation

Article abstract The expression lip̄nē, literally "to the face of," is commonly translated as "before." In combination with the root ngp ("inflict/defeat"), this leads to awkward English translations; e


INTRODUCTION
The expression lip̄nē, formed by lə 'to' with the construct state of pānim 'face', usually functions as a preposition with a spatial or temporal meaning ('in front of'; 'before'). 2 It has a figurative sense when it marks an argument of the verb ngp 'inflict, defeat', and is then commonly translated either as 'before' ([N]KJV, A/R/ESV, NASB95) or as the Agent marker 'by' (NLT,NIV,NASB20). 3  ' . . ., and Abner was beaten, and the men of Israel, before the servants of David.' (KJV) 'And Abner and the men of Israel were beaten before David's servants.' (ESV) ' . . ., and Abner and the Israelites were defeated by David's men.' (NIV) For modern readers, the translation 'before' is hard to understand in this context, because in present-day English before has a 1 I am grateful to Johan Rooryck, Ellen van Wolde, and the anonymous reviewers of the Journal of Hebrew Scriptures for their thoughtful comments on earlier versions of this article.
2 This type of expression is commonly called a 'complex preposition'. Joüon and Muraoka (2006, §103o) describe them in Biblical Hebrew as 'pseudo-prepositions', i.e., 'a combination of one of the prepositions . . . and a substantive, often lexemes denoting parts of the body . . . in the status constructus'. According to Waltke and O'Connor (1990, §11.3a), their meaning is often not predictable from the constituent components and these expressions 'function syntactically as prepositions ' (ibid., §11.3.1a). In the theoretical linguistic literature there is some debate as to how these expressions should be analyzed purely spatial or temporal meaning (the bill is presently before Congress; the day before yesterday). Recent, freer translations have gone a step further in their interpretation and use the Agent marker 'by', an analysis that finds support from Sollamo (2003), Rodriguez (2017, 180), and Jones (2018). Although English translation is not the focus of this article, different choices reveal different interpretations of the underlying Hebrew construction, and thus show that this construction requires explanation. This will be the focus of this paper. In contrast to the recent tendency to interpret lip̄nē as an Agent marker, I will show here that the meaning of lip̄nē and before in cases like (1) is actually Locative. To explain its meaning in this context, I draw attention to a pragmatic connotation of the body part 'face' which expresses that the complement (e.g., the Philistines in [1a]) has some threatening influ- (Seppänen et al. 1994;Huddleston and Pullum 2002, 620-23;Hoffmann 2005;Pullum 2006). These issues have repercussions for the analysis of the Hebrew expressions as well, but this must be left for another occasion. For the present article we can assume that these expressions are functionally, if not syntactically, prepositions. However, to avoid any confusion I will refer to them as 'expressions', not prepositions.
3 I will use the capitalized terms Agent, Patient, Intermediary Agent, Instrument, Cause, Locative, and Source for thematic roles in the tradition of Gruber (1965). There is no universally agreed upon list of thematic roles and their definitions, but the ones I use here are all relatively standard in theoretical linguistics. It is important, however, that they are understood as prototypes in the sense of Dowty (1991, esp. 571-75), so that an argument can fill an Agent slot even if it is not in all aspects like the Proto-Agent. This is somewhat similar to the Actor and Undergoer macroroles in Role and Reference Grammar (Foley and Van Valin 1984), and the argument advanced in the present paper can be reformulated in that framework. For Dowty, the Proto-Agent is a sentient argument that volitionally brings about an event or change of state in another participant and moves relative to the position of another participant; finally it exists independently of the event (1991,572). The Proto-Patient undergoes a change of state and is causally affected by another participant; it is stationary relative to movement of another participant and does not exist independently of the event or not at all (ibid., 572-Dowty's additional property incremental theme is not relevant to us here). The traditional roles can be defined using these properties (ibid., 577): the Agent has volition, causation, and sentience (and movement, but this is not relevant for our purposes). The Patient has change-of-state and causally-affectedness (stationarity and dependent existence are again not relevant here). An Instrument has causation without volition or sentience. I define the Intermediary Agent as an Agent with low volitionality. For Cause I rely on Palancar (2008), who defines this role as "the role played by a given entity-normally conceived of as either an abstract or natural force-construed as the causal force which has brought about a certain state of affairs" (Palancar 2008, 27). A Locative picks out a position in a region, which can be spatial (The ball is in the box) or temporal (He was elected in 2008); a Source refers to the point of origin of an event of motion. For modern discussion on thematic roles, see the reference works (e.g., Davis 2011;Harley 2011;Primus 2016). 4 For an exhaustive list, see the table on p.14.
ence over the subject (Israel). This connotation can be approximated with 'in the face of' or 'in confrontation with' in presentday English. I show that before also had this connotation in older stages of English, which shows that the traditional translations in (1) were perfectly intelligible at the time of writing. The problem is therefore not that the meaning of the Hebrew expression lip̄nē is obscure and requires explanation (the path taken by Sollamo [2003], Rodriguez [2017, 180], and Jones [2018]), but that older translations have become misunderstood due to changes in English. After elucidating these, there is no need for an Agent marker interpretation any more. Below I also show why this interpretation is in itself unlikely regardless of any alternatives. Before turning to the evidence for a connotation of threatening influence for Biblical Hebrew lip̄nē, let me clarify what I mean with this connotation by examining English in the face of and in confrontation with. English has phrases like in the face of adversity/death/ . . ., where the complement has an unpleasant connotation by itself. By contrast, combinations with a complement with a beneficial or pleasant connotation are infelicitous or receive an ad hoc negative connotation: #in the face of profit/the weekend/ . . .. Phrases that are of themselves neutral receive a negative connotation when used as the complement of these expressions: in the face of change/ . . .; in confrontation with the system/other communities/ . . .. Therefore, the prepositional expression provides a pejorative connotation if its complement does not have one already. The complement has a threatening influence in the sense that it has the ability to negatively influence another entity.
What might cause this connotation? It is striking that in the face of and in confrontation with both incorporate a word for 'face'. 5 In many languages, words for 'face' are related to concepts of authority and dignity. We find evidence for this in English (lose/save/retain face) and many languages around the world, as the examples in (2) demonstrate. 6 (2)a. Chinese (Yu 2001, 16): diu-lian 'lose face, be disgraced' (lit. 'lose-face') b.
Thai (ibid., 120): mâi hâi/wái nâ 'be too direct, be inconsiderate' (lit. 'not giving/keeping face') 5 Observe that confrontation comes from Latin frōns 'forehead'. 6 However, one must be careful not to generalize here (see, e.g., Littlemore 2019, 192-201 on cultural variation in embodied metaphor; cf. Strecker 1993 on the body part face in particular). All these languages show that the body part face is in some way related to concepts of authority and dignity, but the exact range in which these metaphors can be used will vary.  Kraska-Szlenk (2014, 30) recognizes the same relationship in human behaviour, in the fact that we hold our face up when we are proud and want to seem respectable while we bow our heads low when we are ashamed. This observation finds support in the Hebrew Bible itself, when Cain's face 'falls' (npl) when his offer is rejected by Yahweh (Gen 4:5-6). 7 The combination of cross-linguistic, psychological, and biblical evidence shows that these words are not arbitrarily related, but that this phenomenon reflects a psychological reality.
The 'face' is not just a metaphor for concepts like authority and dignity, it also comes to denote the space in which a person can exercise authority or enforce their dignity. 8 This is clear from the many idioms where a word for 'face' is combined with a spatial preposition (3). These expressions stem from the unwanted intrusion upon someone's sphere of authority or dignity.
(3)a. Dutch: iemand in zijn gezicht uitlachen 'laugh in someone's face' (Dikke van Dale 2015-2017, s.v. gezicht, 7) All in all, the cognitive underpinning for a connotation of threatening influence for the body part 'face' seems clear: the face is related to concepts of authority and dignity, and comes to denote the abstract region of a person's authority and dignity. When you enter this realm of authority, two things can happen: either you challenge their authority, as in (3), or you relinquish some of your own authority and permit that person's influence over you. This is seen as a threat, which gives rise to the connotation of threatening influence for in the face of and in confrontation with. 7 Compare also Job 29:24, where Job speaks of his former glory: '(the men around me) would not let the light of my face fall (hiphil npl)', i.e., they would pay Job respect in order to preserve his face. 8 Such an extension is common in spatial expressions. For example, front normally denotes a vertically oriented bounded region on an object (e.g., the front of a house), but in in front of this region is projected onto the surrounding (horizontally oriented) environment (Jackendoff 1996, 15). Similarly, words for 'face' can denote someone's authority and dignity, but also the projected region in which they can exercise their authority or enforce their dignity.
Dan 1:5: And at the end (of the years in training) they (the men in training) would serve (lit. stand before) the king. c.
And he (Ishmael) will be a wild ass of a man; his hand will be against everyone and everyone's hand will be against him; and in the face of all his brothers he will live. g.
The man (Joseph) sternly warned us: "You will not see my face (i.e., enjoy my powerful presence) unless your brother is with you." The extensions of Biblical Hebrew pānim relevant here are schematized in the partial semantic-pragmatic network in (5). In this diagram, solid arrows represent semantic extensions (through grammaticalization), and dashed arrows represent pragmatic connotations. On the left we see the relationship with concepts of authority and dignity, as seen in (4ab). At the same time, the body part undergoes semantic bleaching and becomes a generic noun for 'surface' or 'front'. A similar process can be seen in words like facade and surface, as well as in the use of face for 'surface' in darkness was over the face of the deep (Gen 1:2 ESV). The combination of these two developments allows the interpretation of pānim as the 'realm' of authority and dignity, which, as explained above, leads to the connotation of threatening influence. The combination of pānim with prepositions (lip̄nē 'to the face of'; mippənē 'from the face of'; etc.), illustrated on the far right, is quite independent from this connotation and only depends on the generic meaning 'surface', 'front'. For instance, lip̄nē describes a location in a region described by pānim; mippənē describes the movement away from a location in that region; etc. By virtue of compositionality the connotation is still present in these derived expressions, but the connotation does not depend on the combination with a preposition.
I purposefully combine the noun pānim and prepositions built on this noun in one diagram in (5). The reason for this is that metaphorical extensions like that of threatening influence are found with more than one preposition (4a-f), as well as without any preposition (4g). Therefore, the extension is part of the meaning of pānim and did not occur as a result of the grammaticalization of these prepositional expressions. The fact that the extension is also present in these derived expressions indicates that it must have already been part of the meaning of pānim at the moment that derived expressions became lexicalized. The alternative explanation, that the connotation developed independently in each grammaticalized preposition, is less economical. 18 This is not to say that the expressions did not grammaticalize: I still allow for semantic change in each prepositional expression (following common grammaticalization clines). I simply do not see them as entirely separate lexemes, so that their meaning is composed of 'old' meaning common to most expressions built on pānim and 'new' meaning particular to specific expressions. Let us now look at English before again. The claim I will put forward here is that this preposition had the same connotation of threatening influence in earlier stages of English, but that this connotation has been lost. The word therefore instantiates some parts of the semantic network given for Hebrew pānim in (5). In present-day English the meaning of before is spatial and temporal, although the spatial sense is already somewhat archaic when compared to in front of (OED 2021, s.v. before). Etymologically, before is composed from the prefix be-'about', inherited from Germanic, and the adverb fore 'in or of the front' (ibid.). This latter component is ultimately cognate with Latin prō and Ancient Greek πρό, both meaning 'for'; a root, therefore, with a highly grammaticalized meaning for considerable time already. It is therefore hard to see whether before is ultimately related to the body part 'face'. Nevertheless, it is striking that in its spatial sense before refers particularly often to people (6), which might point to a relationship with a body part.
(6)a. Before a crowded Dallas press conference, a pleased Perot unveiled new versions of the old props.
An Italian proverb runs thus, 'Who flatters me before, spatters me behind.' (Marlburian 31 Jan. 1883, 3/1 in citation by OED 2021, s.v. before, A.I.3b) At any rate, we do know that the element fore had, at some point in time, a generic meaning 'front' similar to surface and facade. This can be seen in fossilized expressions like bring something to the fore and compounds like foreground. Furthermore, if we go back to the seventeenth century, we find quite clear evidence for a connotation of threatening influence (7ab). Both of these examples are concerned with fleeing from someone who exercises a kind of threatening influence (cf. [4c]). In (7a), the threatening connotation is also nicely paralleled by 'darest not . . . look me in the face' in the next line. In present-day English, this connotation can still be recognized in some idioms, like bow before (one bows, after all, only before entities with authority), but it is not productive any more; in the OED, the last compelling example similar to these dates from 1931. 19 (7)a.  (5) we saw that the connotation of threatening influence depends on the association with authority or dignity on the one hand and the generic spatial meaning 'front' on the other. The former could perhaps be provided from context in cases like (7), but we have seen that the 19 A 2001 example mentions retreating before a deity, but here the deity can hardly be seen as a causal Agent; this may be a simple spatial usage in the sense 'in front of'. 20 I.e., changing your place continuously. latter meaning is by now only retained in fossilized expressions. Even the spatial meaning 'in front of' is becoming archaic, while the derived temporal one becomes more and more prevalent (OED 2021, s.v. before). The prerequisites for the connotation of threatening influence, still attested in the seventeenth century, are not available any more in present-day English. As a result, before has lost this connotation, and its use in translations of (1) has become misunderstood. Finally, let us return to the root ngp 'defeat, inflict' seen in (1). Bearing in mind that English before used to have a connotation of threatening influence, the translations that use 'before' make perfect sense: the defeated party is in the vicinity of the prepositional object but also threatened by it. The expression X is defeated before Y describes that Y has authority over X. Now that we know that before's connotation of threatening influence was lost in present-day English, we can understand why more recent translations and scholars went looking for an alternative. However, to be defeated before someone in the sense of (7) is very different from being beaten or defeated by someone: the first describes the state of being beaten when confronted with some other party that is more powerful, whereas the second describes the event of being beaten by that other party. Crucially, the former does not imply any actual involvement of the second party: in examples like those in (7) it is the mere presence of the complement of before that is relevant rather than any physical action on their part. The expression 'in the face of' therefore seems a better translation for the combination of ngp and lip̄nē. In the sections below I will also address the question who the actual Agent of ngp is, if lip̄nē does not mark the Agent. I will argue that the Agent is Yahweh, who determines the outcome of battles.
Having made a positive case for a Locative interpretation of lip̄nē (i.e., 'in the face of' rather than the Agent marker interpretation 'by'), we should now also show that an Agent marker interpretation is unlikely regardless of any alternatives. Indeed, there are many reasons on various levels to show that it is very unlikely that lip̄nē would be an Agent marker. In the remainder of this article I discuss five reasons: one from narrative structure, two language-internal, one theoretical, and one typological.

CONTEXT
To properly appreciate the function of lip̄nē in contexts like (1) we must first understand the meaning of the verb ngp 'inflict, defeat'. We can do this by looking at the structure of the larger narrative in which it usually appears. Previously I have shown that the description of battles in the Hebrew Bible follows a rather rigid pattern (Staps 2018, 169-71), which can be described as a prototypical scenario in the sense of Van Wolde (2009, 59-60). One can distinguish seven stages, each of which is characterized by the use of specific verbal roots. Not every stage is necessarily explicitly mentioned in each episode, but the order in which they appear is fixed. The stages can be described as:  (6) it with the edge of the sword, and all the people that were in it: he left no survivors (7). The events described in war contexts of ngp fit the same sevenstage pattern. Because the data set underlying Staps (2018) only included episodes where the verb lḥm 'to fight' is used, not all instances of ngp are covered-in fact, only two cases were included (Deut 1:42; 1 Sam 4:10). Because in these two examples not all stages are explicit, they do not provide enough evidence to determine whether ngp belongs in stage 5 or stage 6. I then preliminarily placed them in stage 5 (ibid., 177-81), based on the involvement of God with this root in the active voice (see the next section) and the meaning of the verb in general. However, an Agent marker interpretation places ngp in stage 6, so I will make a better case for ngp as a stage 5 verb here.
First of all, it is clear that stage 5 and 6 are distinct stages because they frequently co-occur and are always in the same order (ibid., 177-81). The physical conclusion is always performed by a human Agent, whereas the settlement can be determined by God (8b), since war in the Hebrew Bible is a religious event (cf. inter alia Walzer 1992). It is therefore logical that Yahweh plays some role in this prototypical scenario, indeed a pivotal role. It is Yahweh who decides who wins; without him on one's side, one better does not go to battle at all (Deut 1:42). However, while there is evidence for God's involvement in stage 5, the Agent in stage 6 is always human. God is seen as the divine orchestrator of the battle, determining the outcome, but he does not physically participate in it. He may play a different role in battles that are narrated differently, but not in the ones that adhere to this scenario.
The 24 or 25 occurrences of ngp in war contexts that were not yet considered in Staps (2018) also fit the pattern described above. 22 These examples allow us to be more specific as to the place of this root in the prototypical scenario. Concretely, there are instances where a stage 6 verb is present as well, confirming the classification of ngp as a stage 5 verb: (9)a. Then the Philistines arranged (3) themselves to meet Israel. As the battle (4) spread out, Israel was defeated (5) in the face of the Philistines. They struck (6) around 4,000 men on the battlefield. As can be seen, the root ngp is a stage 5 verb in both the qal (active voice, 9a) and the niphal (non-active voice, 9b  (2019), to which may be added the response by Jones (2020). The point of discussion is to what extent the niphal expresses the passive and/or middle voice. Although we are discussing ngp in the niphal, this discussion is only tangentially related to the matter at hand: my analysis that with the observation that God plays a role in stage 5 but not stage 6. The fact that the typical order of events and the way the battle is narrated in general is the same regardless of the stem suggests that the semantic content, and hence argument structure, of the verb is the same as well. In other words, in both the active and the passive voice we must understand this root to have three arguments. In the active voice these can all be made overt, as in (9a): the Agent as the subject, the Patient as the object, and a third argument marked by lip̄nē. In the passive voice we expect the same three arguments in the underlying argument structure. It would be odd if lip̄nē marked the Agent in this situation, because this would (i) remove God from the argument structure when compared to the active voice and (ii) make it impossible to express the third argument (which is expressed by lip̄nē in the active voice). Instead, we should understand an unexpressed Agent in the underlying argument structure. 24 When ngp is used in the niphal, God's involvement is frequently made clear with other verbs in the direct context. 25 The simplest explanation is then that God is still the Agent in the events narrated with a niphal, but that the Agent is, as usual in Biblical Hebrew, not specified. Therefore, lip̄nē does not mark the Agent.
The data of a different root, knʕ 'subdue', supports this. Like the qal of ngp, the hiphil of knʕ is trivalent: God is the Agent, an army is the Patient, and a second army is marked by lip̄nē (10a). The niphal of the two roots is used identically: God is not mentioned, the Patient is the subject, and the second army is still marked by lip̄nē (10b). This root also shows that the argument marked by lip̄nē cannot be an Intermediary Agent or Instrument. For such a function the hiphil would use a ditransitive construction (Joüon and Muraoka 2006, §125u). 26 lip̄nē is not an Agent marker is compatible with both the position that the niphal expresses only the middle voice and the position that it expresses only the passive voice. For this reason I will agnostically refer to the niphal as a 'non-active voice'. 24 Joüon and Muraoka (2006, §132c): 'As a rule a proper passive form can be used only if the author of the action (the agent) is not named' (emphasis original). The normal way to express the Agent in Patientoriented syntax is to use a relative clause. Thus for 'the innocent blood shed by Joab' we get 'the innocent blood which Joab shed' (1 Kgs 2:31). For edge cases where prepositions might mark agency, see Bicknell (1984, 43-51), but this is definitely not the default. 25 Lev 26:17 ('I will set my face against you'); Num 14:42; Deut 1:42 ('Yahweh is not among you'); Deut 28:7, 5 ('Yahweh will allow . . . to happen'); 1 Sam 7:10 ('Yahweh thundered loudly . . . against the Philistines and confused them'); 1 Kgs 8:33; 2 Chr 6:24 ('your people, Israel, sinned against you'); 2 Chr 20:22 ('God wanted to give them in Joash' hand' in v. 20). 26 Of the studies I consulted, Sollamo (2003, 622-25) was the only one to take lip̄nē as an Agent marker with other verbs than ngp 'inflict, defeat' and knʕ 'subdue'. She understands the passive Agent as 'an adjunct which expresses the originator or instrument of the action expressed by the passive predicate or participle and which in most cases can be converted into the subject of a corresponding active construc-(10)a. In the meaning 'stumble', the root is intransitive and of little relevance to us here. However, the context of illness and injury is highly relevant, even though lip̄nē is not found in this environment. The exact semantic boundaries of this context are not exactly clear; it is best understood with a prototype model. The prototype is rather concrete and involves Yahweh 'inflicting' a disease upon humans. There is ample evidence for illness in a number of examples, such as the root ʔnš 'become ill' in (11a). 36 When disease is mentioned explicitly the verb is always in the qal and the Agent is always Yahweh. Somewhat removed from this prototype, but still clearly related, are cases where God inflicts misfortune, but not necessarily disease (11b). We also have two cases where humans and animals injure each other (Exod 21:22,35 Yahweh will pass through to strike / inflict something upon the Egyptians. Observe that in almost every example God is the Agent of ngp. This is also the case for the war contexts in which ngp is in the active voice (12). Therefore, as already seen above, ngp is primarily a verb of divine intervention. If lip̄nē were an Agent marker, we would expect its complement to be God. This is however never the case: lip̄nē always marks an army, while armies are never the Agent in the active voice. And Yahweh defeated the Cushites in the face of Asa and in the face of Judah. The second language-internal reason why we should not take lip̄nē as an Agent marker is even clearer. In table 1 we see that lip̄nē is also used when ngp is in the active voice, as in (12). In the active voice, the Agent slot is taken by the subject. Therefore, with an Agent marker interpretation, lip̄nē must have another function in these four instances. A Locative interpretation does not suffer from this problem. 37

MARKING WOULD NOT BE SEMANTICALLY RESTRICTED
The fourth reason why I argue that lip̄nē is not an Agent marker stems from grammaticalization theory. Grammaticalization is the process 'whereby particular items become more grammatical through time' (Hopper and Traugott 2003, 2). In our case, the particular item is the expression lip̄nē: literally this means 'to the face of', but over time it acquires the more grammatical senses of 'in front of' (spatial Locative) and 'before' (temporal Locative). 38 According to Rodriguez (2017, 180) this meaning has 37 Jones (2018, 225-26) solves this by reading lip̄nē as an Instrument marker: 'by/through/using'. But like an Agent marker reading, this interpretation as an Instrument marker can be ruled out on the basis of theoretical and typological arguments similar to the ones shown in the next sections. In short, it is unclear how lip̄nē would have developed from a spatial preposition to an Instrument marker, and we cannot explain why it would only be an Instrument marker in the context of ngp 'defeat'. 38 For our present purposes we can loosely define 'more grammatical' using the distinction between content words (example, accept, green) on the one hand and function words (of, and, or, it, this) on the other: the latter group is more grammatical than the former. then further grammaticalized into Agent marking, and such a development is implicitly understood by Sollamo (2003) and Jones (2018) as well.
Grammaticalization, however, is not simply the acquiring of new, more grammatical, meanings. Grammaticalization theory provides a framework from which falsifiable predictions follow, against which we can check the likelihood of lip̄nē having developed an Agent marking function. 39 One of these predictions is that as a lexeme proceeds to acquire more grammatical meanings, it undergoes certain morphological changes that can be schematized by the cline in (13).
(13) content item > grammatical word > clitic > inflectional affix (Hopper and Traugott 2003, 7) Based on this cline it is clear that Agent marking is a highly grammatical function: in nominative-accusative languages with overt case marking, the Agent is marked with an inflectional affix in the active voice (the nominative case). Furthermore, many such languages mark the Agent in non-active voice with an inflectional affix as well, like the instrumental case in Russian. On the other hand, the spatial and temporal senses of lip̄nē ('before') are usually expressed with a preposition, i.e., a grammatical word (even if that preposition requires some case ending). These Locative senses are therefore less grammatical than Agent marking.
Furthermore, as words proceed down this cline, they lose semantic and pragmatic meanings. As Hopper and Traugott (2003, 94) put it, 'as grammaticalized forms become increasingly syntacticized or morphologized they unquestionably cease over time to carry significant semantic or pragmatic meaning.' Thus, while the syntactic and morphological environment in which they occur may become more restricted, the semantic-pragmatic environment will become less restricted. For example, casemarking nominative-accusative languages use the nominative case (which is typically reduced in syntactic freedom and phonetic substance) for the subject regardless of the semantic or pragmatic context; this function is heavily embedded in the grammar.
We thus see that in grammaticalization two processes occur simultaneously: on the one hand, grammatical functions such as Agent marking are acquired, but this goes hand-in-hand with a reduction in semantic-pragmatic content ('bleaching'), syntactic freedom, and phonetic substance. It is unfortunate that this second aspect is not always taken into account by Hebraists. In the specific case of lip̄nē, we see that there is no loss of pragmatic content (the connotation of threatening influence is still present), no semantic bleaching (there are only two verbs in combination with which it would be an Agent marker). There is furthermore no evidence for reduction in syntactic freedom or phonetic substance. All in all, this is unlikely for a highly grammatical function like Agent marking. On the other hand, the Locative sense with its connotation of threatening influence that I propose is not only clearly less grammatical but also found in more semantic environments, like the example of ʕāmaḏ lip̄nē 'stand before' in the sense of 'serve' (Dan 1:5).

TYPOLOGY: NO ARBITRARY SEMANTIC SHIFTS
Finally, when a word undergoes semantic shifts or grammaticalization, it does not acquire new meanings arbitrarily. These processes follow common paths fueled by cognitive processes. For example, lip̄nē came to mean 'in front of' because it describes the space close to or directed towards (lə) the face (pānim). Since the processes through which meaning changes are not particular to any language, we would expect similar expressions in other languages-indeed, English before and in front of are parallels to lip̄nē in this sense. This spatial sense then developed into a temporal one ('before') on the basis of the TIME IS STATIONARY AND WE MOVE THROUGH IT metaphor (Lakoff and Johnson 1980, 43), in which events are seen as locations. Again, English before has here undergone the same development as Biblical Hebrew lip̄nē. However, it is not at all clear how one of the senses of lip̄nē would have developed into an Agent marking function, and Sollamo (2003), Rodriguez (2017), and Jones (2018) do not attempt to explain this. In default of a cognitive explanation, we can consider the a priori likelihood that lip̄nē developed an Agent marking function from its Locative meaning, by searching for expressions in other languages that may have undergone the same shift. The typological study of Palancar (2002) examines 148 languages with Agent markers (176 distinct markers), of which 87 languages have nominative-accusative alignment like Hebrew (106 passive Agent markers in total: Palancar 2002, 16-17). Palancar cross-references these markers with their other functions. In nominative-accusative languages, Agent markers commonly share their Agent marking function with the Source (47%), Cause (33%), Locative (27%), and Instrument (23%) functions (ibid., 41-43). 40 At first sight, it may seem that the 27% of passive Agent markers with a Locative function suggests that lip̄nē may be an Agent marker as well. However, closer inspection shows that it is not common that a grammatical item has only Locative and passive Agent marking functions: in 14 out of the 18 times that Locative and passive Agent co-occur, other functions are involved as well, like Cause or Instrument. It seems that the Agent marker developed from such intermediary categories rather than directly from the Locative function (ibid., 206-7).
Furthermore, those languages that do attest a direct development from Locative to passive Agent marker tend to be centred around Oceania, so this cline may depend on areal features. This pattern is also found with ergative Agent markers (ibid., 259-61): Agent markers tend to develop directly from Locative markers with high frequency in Oceania, and the only two cases outside Oceania are of the Chukotko-Kamchatkan family, spoken in the far east of Russia. Direct development of Agent marking from Locative functions therefore seems to be highly localized. In contrast, the use of body parts as spatial prepositions and the relationship between 'face' and authority and dignity is ubiquitous. Hence it seems a priori highly unlikely, though indeed not impossible, for lip̄nē to have developed an Agent marking function, especially since no examples of Cause or Instrument marking have been found.

RELATED WORK
In the previous sections I have first made a positive case for a connotation of threatening influence for the word pānim 'face' and the series of prepositional expressions derived from this noun. I have argued that this analysis is particularly useful to read occurrences of lip̄nē 'before' with the verb ngp 'defeat, inflict', and advanced five arguments why the recent interpretation of lip̄nē as an Agent marker 'by' in this context is incorrect. It is now time to discuss how this analysis relates to the work of the proponents of the Agent marker interpretation. First of all, Sollamo defines the Agent as 'an adjunct which expresses the originator or instrument of the action expressed by the passive predicate or participle and which in most cases can be converted into the subject of a corresponding active construction ' (2003, 623). She notes that lip̄nē 'usually means "in the presence of, in front of, before" ' and that this meaning is also found in passive clauses (2003,623). However, she goes on to mention 'a few cases [in which] it is better to interpret lip̄nē as the preposition of an agent ' (2003, 623). The argument is purely exegetical: it is based on the subjective question which interpretation is 'better'. But on the other hand Sollamo generally does not consider the wider context, such as the prototypical war scenario I described above, and thus relies too much on a translation like 'defeat' to decide whether a verb requires an Agent in the passive voice or not. For the combination with ngp 'defeat, inflict', she writes: 'Because ngp lip̄nē indicates the victor and not where the victory was won, it is simplest to regard the preposition lip̄nē as the preposition of the agent here ' (2003, 624). She thus takes thematic roles as discrete: the argument is either Locative or Agent; since it cannot be Locative, it must be the Agent. In my analysis, the argument has indeed some Agent properties (such as sentience and the power to bring about an event or change of state), but not others (like actually bringing about that event or change of state, because a comparison with the active voice shows that Yahweh is the actual Agent).
It is also worth noting that Sollamo's definition of the passive Agent combines very broad semantic properties ('originator or instrument of the action') as well as syntactic ones ('adjunct'; 'can be converted into the subject of a corresponding active construction'). This is broad to the point that it loses its explanatory power. Even if lip̄nē would mark the Agent with ngp according to this definition, it does not tell us whether the army would be the 'originator' of the action or the 'instrument' of an action by some other originator. At the same time, we cannot narrow the semantic aspects of the definition because they are tied to the syntactic properties. A syntactic subject need not be an Agent (in my stricter sense), as is evidenced by well-known examples like The hammer broke the window. Therefore, if the definition is to include syntactic aspects, it seems it necessarily loses explanatory power in semantics.
The next work to discuss the combination of lip̄nē with ngp is Rodriguez (2017). 41 He begins with the question whether lip̄nē is actually a 'word', and argues that we should ask this question for each usage instead of the construction in general: we can 'identify which utterances are composites of two words and which are utterances where the two words have become a frozen union ' (2017, 167-68). This is an enhancement of the position that lip̄nē is either always a 'composite' (and hence structurally transparent) or always a 'frozen union'. I want to take this even further, arguing that there is no need to categorize usages as either 'composite' or 'frozen union', because even a frozen union with a grammaticalized meaning can retain meaning of the composite parts. Thus the grammaticalized Locative meaning of lip̄nē 'before' (in which we do not understand an actual face) still carries meaning of the noun pānim (including the connotation of threatening influence). This allows me to distinguish between meaning that results from grammaticalization (Locative) and meaning that results from metaphorical extension (threatening influence). This way we do not need to assume a theoretically and typologically unlikely grammaticalization path. Furthermore, this approach allows us to attribute the metaphorical extension to the noun pānim rather than the prepositional expression lip̄nē. As a result, we have a more economical explanation for the same connotation with other prepositions and standalone pānim (see example 4).
The last author to discuss here is Jones (2018). He starts out with a methodology much like the one I followed in section 3, investigating the number of arguments ngp 'defeat, inflict' requires and semantic properties (2018, 220). However, he includes all usages of ngp, including the ones that occur in the context of illness and injury. As a result, he concludes that the Agent of ngp can be human (based on Exod 21:22 in the illness and injury context), whereas such cases do not occur in the war context. There is good reason to separate these contexts: both the niphal of ngp and a third argument marked by lip̄nē only occur in the war context. That lip̄nē also occurs in the active voice is ignored in Jones's discussion of the arguments of ngp. He only discusses this briefly in his discussion of 1 Sam 4:2 (2018, Then the Philistines arranged themselves to meet Israel. As the battle spread out, Israel was defeated (ngp niphal) in the face of the Philistines. They struck around 4,000 men on the battlefield. When the people came to the camp, the elders of Israel said: 'Why has Yahweh defeated (ngp qal) us today before the Philistines?' Jones admits the possibility that Yahweh is the 'deleted agent in 4:2' (2018, 225), but prefers to take the Philistines as the Agent because they are involved in 'striking around 4,000 men'. He argues that the speech of the elders in 4:3 cannot be used to define the description of the event by the narrator in 4:2 (2018, 227). This is an important observation, but by pulling the two verses apart entirely Jones ignores that 4:3 attests to the fact that ngp in the active voice often occurs with lip̄nē. For 4:3, Jones proposes to read lip̄nē as a marker of the Instrument ('Why did the Lord defeat us today by the Philistines?': 2018, 227, emphasis original). It is more economical, however, to attribute the same function to lip̄nē in all occurrences with ngp, both active and non-active. Jones does not accept Locative as a universal function, branding it as an 'unhelpful and inappropriate translation ' (2018, 217). However, he understands Locative in a very strict sense and does not seem to have considered metaphorical extensions. Furthermore, it is clear that he is working with the modern sense of English before, while I have shown that this translation is not as bad as one may think-only old-fashioned.
Finally, Jones observes that the niphal of ngp occurs almost exclusively with lip̄nē, implying that this suggests that it may mark the Agent: 'It should at the very least be curious that the passive of a transitive action "defeat" appears with a so-called locative lip̄nē in nearly every occurrence ' (2018, 227). However, as I have shown above, the same is true for nearly every occurrence in the active voice (if we limit ourselves to the war context). Therefore, that lip̄nē is also found in the niphal is not so surprising, and an explanation should not be sought in its co-occurrence with this stem. Jones is right to reject a strictly Locative reading, but the solution can be found in the connotation of threatening influence instead of an entirely different function for lip̄nē.

CONCLUDING REMARKS
In summary, translations and other scholars have proposed that lip̄nē is an Agent marker when used with ngp 'inflict, defeat' and perhaps a few other verbs. This is incorrect for five reasons: (1) the larger context of this verb shows that the Agent of these events must be God, but lip̄nē marks humans; (2) the Agent of the verb ngp in the active voice is God, but again lip̄nē marks humans; (3) lip̄nē is also found with ngp in the active voice, when the Agent slot is already filled by the subject; (4) Agent marking is a highly grammaticalized function that should not be restricted to a specific semantic context; (5) it is cross-linguistically unlikely for a Locative preposition to directly develop an Agent marking function. Instead, we have seen evidence from various languages that the body part 'face' is related to concepts of authority and dignity, which can give rise to a connotation of threatening influence. This leads to the use of lip̄nē 'to the face of' with ngp 'inflict, defeat' in war contexts, as in (1a): 'Israel was defeated before the Philistines' (ESV). Here, 'before' denotes the threatening influence of the Philistines, and a more suitable translation in presentday English would be 'in the face of'.
The exegetical implications of my proposal are clear. When ngp 'inflict, defeat' is used in the niphal stem, it is the state of being defeated that is described-rather than an event. 42 Based on parallels with the active voice, it is clear that the implicit Agent effecting this state is God. He is left unexpressed because the niphal focuses on the state itself, not the event that caused it. The argument marked by lip̄nē has the same function as in the active voice: it marks the dominant party in that state; the party that has a threatening influence over the subject (e.g., Israel in [1a]). The implication is that God has allowed Israel to be defeated in confrontation with the Philistines. 42 Testen (1998, 138) argues that the niphal ingressively describes the entering into the state described by the verb, but Van Wolde's resultative category (2019, 467; 2021, 438) focuses primarily on the state itself rather than the entering into that state. The exact aspect is not important here, as long as the state of being defeated is in view.