1 00:00:02,260 --> 00:00:07,040 About Mary as it comes together to wish her well as she heads into retirement? 2 00:00:07,040 --> 00:00:11,500 Well, there's a lot to say. I promise to keep this brief. I'm not the only person who's going to be speaking. 3 00:00:11,500 --> 00:00:18,340 And so I'm going to offer a few personal reflections. I don't if she realises this, but I have to admit to being rather starstruck. 4 00:00:18,340 --> 00:00:27,450 The first time that I met Mary LFG conference because that 2001 book kind of left, you'd been a constant reference during my masters in these studies. 5 00:00:27,450 --> 00:00:32,470 So I was a little bit tongue tied, but I think I got over it. 6 00:00:32,470 --> 00:00:41,620 But my abiding memory of interactions with Mary at that meeting and subsequent meetings at conferences over the years while I was the post student, 7 00:00:41,620 --> 00:00:47,800 was of how genuinely curious she was about what I was doing, my research. 8 00:00:47,800 --> 00:00:58,030 I remember her particularly very early on after I'd just met her spending an extended meal time with me at a conference talking through the analysis. 9 00:00:58,030 --> 00:01:02,590 I was preparing for my PhD thesis and how what mattered most to Mary was that 10 00:01:02,590 --> 00:01:06,880 she understood the data as she understood the analysis and how it worked, 11 00:01:06,880 --> 00:01:14,890 understood its significance and understood what I was trying to achieve. And to me, that really typifies Mary's approach and outlook. 12 00:01:14,890 --> 00:01:23,200 So for every inquisitive about language and the significance of the data, but also deeply interested in what another researcher is trying to achieve, 13 00:01:23,200 --> 00:01:28,570 no matter how much experience they may or may not have had at that point in academia, 14 00:01:28,570 --> 00:01:35,350 and truly generous in terms of how much time she's willing to spend not only learning about your research, 15 00:01:35,350 --> 00:01:45,910 but also very gently challenging your assumptions and analysis in such a way that you'd always be the better for having had that chat with Mary. 16 00:01:45,910 --> 00:01:49,780 It's an experience that I feel very fortunate to have had on many, 17 00:01:49,780 --> 00:01:58,180 many occasions from being a Ph.D. student to having her as a mentor while I was a postdoc through to being her colleague in the faculty to date. 18 00:01:58,180 --> 00:02:03,130 And of course, I've seen a mentor, many, many others in the same way. 19 00:02:03,130 --> 00:02:07,930 I'm not sure how Mary would feel about being regarded as a role model, 20 00:02:07,930 --> 00:02:11,710 but it's one of those things that you don't necessarily have much of a say in yourself. 21 00:02:11,710 --> 00:02:16,510 So as we take this opportunity to say goodbye and wish her well, 22 00:02:16,510 --> 00:02:22,720 I'd like to say a personal and a professional thank you to Mary for her generosity of time and spirit, 23 00:02:22,720 --> 00:02:32,580 for her inquisitiveness, for her rigour, and for showing the way to me and many, many others throughout her career. 24 00:02:32,580 --> 00:02:36,600 Oh, thank you very much, Louise. That's her. 25 00:02:36,600 --> 00:02:40,890 Well, it's you know, even if you clap, you're going to hear you just hear it. 26 00:02:40,890 --> 00:02:44,970 And, um, but yes, I'm going to pass. 27 00:02:44,970 --> 00:02:50,310 That is Jamie. Ah, there you are. Yeah, I'm here. Yeah. Hello. 28 00:02:50,310 --> 00:02:55,530 Well, what I have to say has a remarkable parallelism with the Louise that it's clearly a common thread going here. 29 00:02:55,530 --> 00:03:02,130 But um. Yeah, I mean, it's a very, very big Goffer term, but it's a pleasure to be invited back, 30 00:03:02,130 --> 00:03:06,090 especially because Mary has had such an important role in my academic career. 31 00:03:06,090 --> 00:03:13,890 I realised, looking back the other day that I was in her prelims grammatical analysis class that I researched for FHA syntax, 32 00:03:13,890 --> 00:03:20,760 that she supervised my master's thesis, she supervised my thesis, and then she was a mentor and a colleague when I started teaching as well. 33 00:03:20,760 --> 00:03:30,240 So really, she's been there the whole time. So, um, yeah, I'm very happy, very privileged to be able to see, uh, to do say some words now about her. 34 00:03:30,240 --> 00:03:36,180 Um, and really I want to because I think my role he really is as the representative of former students. 35 00:03:36,180 --> 00:03:42,750 So I wanted to say a few things, um, three things about Mary that I think have helped us as students. 36 00:03:42,750 --> 00:03:48,060 And I know that many of us sort of strive to emulate in that role model thing, 37 00:03:48,060 --> 00:03:55,200 emulate in our own professional work for the first as much as the with a boundless enthusiasm for for other peoples, 38 00:03:55,200 --> 00:03:57,690 for their of the students were great for ideas. 39 00:03:57,690 --> 00:04:04,650 I think it's very hard when you start out as an undergrad or a master's student trying to propose your own questions and your own thoughts, 40 00:04:04,650 --> 00:04:12,420 thinking that anything you have to say is original. Um, and Mary is very good at helping you find where it can be and show you how it can be. 41 00:04:12,420 --> 00:04:16,960 And, you know, that's not to say that it's everything is wonderful and there's no criticism. 42 00:04:16,960 --> 00:04:24,930 There's sometimes quite a crushing criticism when you realise you've missed something fundamental but that you 43 00:04:24,930 --> 00:04:29,460 are excited to improve because you can tell that Mary is excited to see the projects develop and to improve. 44 00:04:29,460 --> 00:04:36,090 And I think that's a huge bonus for students, really. Um, the second thing is, is generosity, generosity of spirit. 45 00:04:36,090 --> 00:04:40,680 I was going to be generous of your time also works, um, because I think, you know, 46 00:04:40,680 --> 00:04:47,820 it's clear to all of us who've been supervised by Mary that she's she cares very deeply about our well-being as well as our academic progress. 47 00:04:47,820 --> 00:04:51,900 And that can be as simple as kind of day to day work life balance stuff. 48 00:04:51,900 --> 00:04:57,630 But also I think it goes, you know, to sort of professional development and helping point out job opportunities, 49 00:04:57,630 --> 00:05:03,090 career development stuff and being very generous with her contacts and with the network and putting you in touch with people. 50 00:05:03,090 --> 00:05:09,720 And again, that makes a huge difference when you're starting out. So that's a really important thing that, you know, I think we should do so. 51 00:05:09,720 --> 00:05:15,120 But the third thing really is what I think of the two stem from, which is a remarkable humility, 52 00:05:15,120 --> 00:05:20,820 um, because I think Mary is an incredibly accomplished academic with a very successful career. 53 00:05:20,820 --> 00:05:27,900 And yet in a world and academic world that is so full of very big egos, she has managed to remain very humble. 54 00:05:27,900 --> 00:05:35,100 And it's I think that's what makes a good supervisor in any field, really is the, uh, 55 00:05:35,100 --> 00:05:40,560 not just in academia, but measuring your success by the success of those under you. 56 00:05:40,560 --> 00:05:43,320 And I think that's something that Mary does exceptionally well. 57 00:05:43,320 --> 00:05:51,300 And so it's heartening and encouraging to see that, you know, you can achieve so much and still remain so, you know, kind of open and genuine. 58 00:05:51,300 --> 00:05:57,030 And that's been a real inspiration. And I think Louise is right to say that that's been a role model for many of us. 59 00:05:57,030 --> 00:06:00,600 So I will, uh, stop talking briefly. 60 00:06:00,600 --> 00:06:09,210 But I just want to say on behalf of myself and as a self-appointed or at least appointed representative of former students, 61 00:06:09,210 --> 00:06:14,220 um, I wanted to say a really very hearty thank you, Mary, for everything you've done and everything you continue to do. 62 00:06:14,220 --> 00:06:18,660 Let's be honest, um, for all of us, even when we are long gone. Um, so. 63 00:06:18,660 --> 00:06:23,580 Yeah. And I hope you will enjoy a very well earned and prosperous retirement. 64 00:06:23,580 --> 00:06:28,440 Thank you. Thank you. Thank you so much, Jamie. 65 00:06:28,440 --> 00:06:38,520 Uh, I'm now going to move from students to to colleagues as Steve well, 66 00:06:38,520 --> 00:06:45,030 Jeremy and Louise have said quite a lot of the things that I wanted to say with which of course, 67 00:06:45,030 --> 00:06:50,730 I agree and I have one advantage, which is that I've known Mary for a lot longer than they have. 68 00:06:50,730 --> 00:07:00,270 And in fact, I embarrassed both by saying that I think Mary and I first met in the eighties when we were both working at Nasseri International, 69 00:07:00,270 --> 00:07:05,880 which was a research unit spun out of Stanford University. 70 00:07:05,880 --> 00:07:12,750 And we had a shared interest then and in the application of this strange algorithm called 71 00:07:12,750 --> 00:07:21,750 higher order unification to in Mary's case with Fernando Stuart and a Lipsitz and in my case, 72 00:07:21,750 --> 00:07:27,810 applications to to focus. So we had we had that in common. 73 00:07:27,810 --> 00:07:39,220 Mary subtlely left Asara International. But she went to a park where she worked with several people, one of whom is present today, 74 00:07:39,220 --> 00:07:50,410 Ron and Martin Kay, who were sort of old friends of mine by that time on the island, but not them. 75 00:07:50,410 --> 00:07:53,410 And we came across each other a number of times then, 76 00:07:53,410 --> 00:08:06,550 particularly when one of my students to Crouch went to work at a park and worked with Ron and Mary on a variety of of semantic and logical efforts. 77 00:08:06,550 --> 00:08:13,990 We also came across each other quite a number of times. And Ron to on an EU project called Fracas, which I was involved in, 78 00:08:13,990 --> 00:08:23,050 along with Robin Cooper and Stanley Peters and a young guy like I remember a number of meetings since they were in Germany. 79 00:08:23,050 --> 00:08:33,240 They probably involved lots of food and quite a lot of beer and the Black Forest and various other kind of nice watering holes. 80 00:08:33,240 --> 00:08:41,020 We eventually I was in linguistics and in Oxford from about 2000 to 2006. 81 00:08:41,020 --> 00:08:47,460 And during that time I think we had a couple of girls are getting married there and finally got it. 82 00:08:47,460 --> 00:08:56,320 And unfortunately, I jumped ship of shortly afterwards to go back to computer science while we were colleagues. 83 00:08:56,320 --> 00:09:05,320 I was enormously impressed by her kind of Catholic tastes and linguistics, not necessarily in linguistics areas, 84 00:09:05,320 --> 00:09:13,240 but in the kinds of linguistic phenomena that she was interested in and the range of languages and the way that she was able to 85 00:09:13,240 --> 00:09:22,930 connect with students from far flung parts of the world and get them to download their knowledge of their exotic languages so that, 86 00:09:22,930 --> 00:09:25,480 you know, all of these things could be coded. 87 00:09:25,480 --> 00:09:33,320 And as a colleague, she was just incredibly supportive and also makes me very even tempered, which is a great virtue. 88 00:09:33,320 --> 00:09:41,920 As Jamie said, you know, many there are many egos and linguistics and we've had a few of them in Oxford, funnily enough. 89 00:09:41,920 --> 00:09:48,370 But anyway, Mary, it's been great working with you, and I'm sure you won't be really retiring, 90 00:09:48,370 --> 00:09:57,700 but I hope you'll manage to take it easy and enjoy not being in the Clarens and walking down little Clarendon Street in the morning. 91 00:09:57,700 --> 00:10:01,960 I think she's going to be retiring entirely. 92 00:10:01,960 --> 00:10:05,860 Is that grounds to look after her? And therefore, we are keeping her for a while. 93 00:10:05,860 --> 00:10:12,010 And I'm going to hand this over. And we decided, you know, she has to be some outside Oxford as well. 94 00:10:12,010 --> 00:10:19,000 This would be just runners here, too. But we have nice Vincent with us as well. 95 00:10:19,000 --> 00:10:22,570 OK, this was what I predicted. 96 00:10:22,570 --> 00:10:26,110 Basically, I get used to it by being at the end of the alphabet anyway. 97 00:10:26,110 --> 00:10:29,860 But when I was at the bottom of the list, I thought, I know what's going to happen. 98 00:10:29,860 --> 00:10:33,580 Everybody will have said everything that I would want to say. 99 00:10:33,580 --> 00:10:44,590 And all I even made a little list of of key words that included included things like generosity of spirit, breath, openness, a range of things. 100 00:10:44,590 --> 00:10:47,350 And then everybody goes unmentioned. 101 00:10:47,350 --> 00:10:58,000 I guess the sort of to context in which I know Mary, that are sort of between the the the Oxford based Student one and Steeves, 102 00:10:58,000 --> 00:11:06,190 which goes back to before the time when I first knew Mary, which was when the LFG meeting started in the in the 1990s. 103 00:11:06,190 --> 00:11:18,160 And so it's been at those. Annual meetings that I've met her and learnt to admire exactly her range of interest from morphology, 104 00:11:18,160 --> 00:11:28,750 formal semantic syntactic structure, pragmatic structure, and also at the same time exactly as other people have said, 105 00:11:28,750 --> 00:11:33,580 the range of languages and particularly the work that's been done on on on Austronesian 106 00:11:33,580 --> 00:11:42,160 languages and the big project on differential object marking with with Irina Nikolayevich. 107 00:11:42,160 --> 00:11:50,780 So all those skills, abilities and exactly her one word that people haven't used that I have on my list here is the niceness. 108 00:11:50,780 --> 00:11:56,290 She's just playing really nice person that if you want to go and ask her a question, however dumb, 109 00:11:56,290 --> 00:12:01,340 however ignorant, you still get a tolerant, understanding, coherent kind of answer. 110 00:12:01,340 --> 00:12:10,630 So exactly. And that is, as others have indicated, this isn't a quality that one always finds in our professional world. 111 00:12:10,630 --> 00:12:16,960 I guess the other context in which I've worked with her and known her is in work at the British Academy. 112 00:12:16,960 --> 00:12:26,020 And there she's an absolute example. We have a lot of work to do there in terms of evaluating project assessment and committee work. 113 00:12:26,020 --> 00:12:34,900 And you can guarantee if somebody hands out a bunch of post-doctoral projects to be evaluated or small grants to be evaluated, 114 00:12:34,900 --> 00:12:42,500 Mary's evaluations come in almost before you even started reading the email about what you ought to do. 115 00:12:42,500 --> 00:12:49,870 So in every respect, just a wonderful human being, a very, very fine scholar, 116 00:12:49,870 --> 00:12:57,280 and certainly somebody that shouldn't go off the intellectual horizon just because of this technical thing called retirement. 117 00:12:57,280 --> 00:13:05,050 So I wish you well, but I will agree with you entirely. 118 00:13:05,050 --> 00:13:11,350 And if I said if you were in the gardens of some college, I would have said, you know, who else would like to speak? 119 00:13:11,350 --> 00:13:17,290 But if this is not the medium to do that. So I'm going to have a we have a small other plan over here. 120 00:13:17,290 --> 00:13:24,220 Good to ask Ken if he could sort of barge in if possible. 121 00:13:24,220 --> 00:13:32,170 Let's see if this barging in is going to happen. Is he coming in and. 122 00:13:32,170 --> 00:13:37,300 Oh, yes. Yes. Good, excellent. Look, there's one more. 123 00:13:37,300 --> 00:13:40,960 I got it. OK, I got a reaction from. 124 00:13:40,960 --> 00:13:46,060 Very good. That is a cod, believe it or not. 125 00:13:46,060 --> 00:13:53,200 This is a card. That is a card. But the box, it's open the card. 126 00:13:53,200 --> 00:14:03,100 I believe this is happening. This is unbelievable. Wow. 127 00:14:03,100 --> 00:14:08,290 It's very hard. It's just it is it was the largest I could find. 128 00:14:08,290 --> 00:14:12,380 It took some effort. Oh. Wow. 129 00:14:12,380 --> 00:14:19,930 Wow, that's a lot of boxes, just perhaps can can help you open one to see what. 130 00:14:19,930 --> 00:14:46,700 So I'm sure it's a contribution from the entire faculty and a small craft, but. 131 00:14:46,700 --> 00:15:03,480 Wow. Well, it's very well wrapped. 132 00:15:03,480 --> 00:15:12,790 Well, yes, I suppose they were afraid of it just so I. 133 00:15:12,790 --> 00:15:27,720 The door. Yes, indeed. In Russia at least, you can't believe you organised it, required a bit of them. 134 00:15:27,720 --> 00:15:52,800 Have plenty of time. Nothing to do, really. So one of the things I enjoy doing is because I am afraid we can't quite see the. 135 00:15:52,800 --> 00:16:00,580 Oh yes. Now we can. Yeah, that is beautiful. Well, we were told that you like Japanese poetry. 136 00:16:00,580 --> 00:16:05,430 Wow. And that suspension of two giant boxes of Japanese. 137 00:16:05,430 --> 00:16:11,940 Yes. I have never seen that. 138 00:16:11,940 --> 00:16:20,810 A loss for words. But here we are it. Wow. 139 00:16:20,810 --> 00:16:25,560 Now I have to invite everybody over, you can come and eat. Exactly. 140 00:16:25,560 --> 00:16:47,440 Yes. You know, the reason for this, of course, is looking the other way. 141 00:16:47,440 --> 00:16:52,010 You don't have to open everything there. That's it's going to take some time, I think. 142 00:16:52,010 --> 00:16:58,070 Yeah, I think so. Plates as well. 143 00:16:58,070 --> 00:17:02,550 Yes. It has an entire set for you. Wow. 144 00:17:02,550 --> 00:17:06,380 Wow. Thank you. That's so nice. It's so unexpected. 145 00:17:06,380 --> 00:17:10,430 I have no idea. I thought we were going to have a nice talk from Rob. 146 00:17:10,430 --> 00:17:14,720 I wish you wouldn't show up. I'm desperate. 147 00:17:14,720 --> 00:17:24,260 Best to make sure that you could come. I even got married to try to persuade her to introduce run, but try to make sure that she was there. 148 00:17:24,260 --> 00:17:29,240 But then she said she had a cough and I got really nervous and so. 149 00:17:29,240 --> 00:17:33,260 OK, so I think you should move to run. Stop, which is also planned. 150 00:17:33,260 --> 00:17:38,910 And time for you to say thank you, though, once it's completely unexpected. 151 00:17:38,910 --> 00:17:48,550 Thank you. This is amazing. And it's so nice to see people that I haven't seen in such a long time to thank you. 152 00:17:48,550 --> 00:17:58,870 And teams that allow us to see more people, but they have sent messages and I'm sure they will send you more messages after this. 153 00:17:58,870 --> 00:18:07,300 I'm going to I'm not going to introduce Ron. I'm going to pass this on to was done who's in charge of run and then I'll take over at the end. 154 00:18:07,300 --> 00:18:14,530 Thank you. I'll do it. As someone who always struggles to do at most one thing semi competently, 155 00:18:14,530 --> 00:18:22,510 I'm in awe when I see Ron Kaplansky, which has so many points on it that it's quite intimidating. 156 00:18:22,510 --> 00:18:33,790 So of course we know him as a distinguished LFG person, in fact, as one of the people who designed it and implemented the computational side of it. 157 00:18:33,790 --> 00:18:35,110 But when you look at your CV, 158 00:18:35,110 --> 00:18:44,830 you can see he actually has a B.A. in mathematics and language behaviour from Berkeley and he did his Ph.D. in Harvard in social psychology. 159 00:18:44,830 --> 00:18:51,610 He was vice president at Amazon and he is a distinguished scientist at Nuance Communications. 160 00:18:51,610 --> 00:18:55,360 The list goes on and on and on. And I'm not going to bore you with it, 161 00:18:55,360 --> 00:19:03,910 but I just wanted to say how deeply impressive it is to see so many achievements besides the outstanding linguistic achievements. 162 00:19:03,910 --> 00:19:08,650 And so we are extremely pleased to have you here on to give a talk to us. 163 00:19:08,650 --> 00:19:11,440 Thank you very much for agreeing. 164 00:19:11,440 --> 00:19:23,800 Well, thank you for inviting me and inviting you to participate in this celebration of Mary's career and achievements. 165 00:19:23,800 --> 00:19:28,990 I think I I knew Mary from the other side. 166 00:19:28,990 --> 00:19:37,380 She was a student and then became a friend, close friend and colleague and collaborator. 167 00:19:37,380 --> 00:19:48,860 Over maybe thirty five years. Maybe Steve and I can have an arm wrestle about who first started working with her first. 168 00:19:48,860 --> 00:20:01,230 But we go back a long time and it's been incredibly productive interaction and collaboration for me, certainly, and I hope for Mary as well. 169 00:20:01,230 --> 00:20:11,220 So I'm very pleased to be able to participate in this over the years, I think in some circles, 170 00:20:11,220 --> 00:20:16,030 but I think, again, Nigel mentioned the alphabetical sequencing problem. 171 00:20:16,030 --> 00:20:24,060 I think D comes before K I think in some circles, which is something known as Dalrymple at all. 172 00:20:24,060 --> 00:20:35,280 So it's been a long and really productive collaboration, so let's see, we did a dry run. 173 00:20:35,280 --> 00:20:44,960 Yeah, we did. So let's hope it works out. Let's let's see if if I click on the Magic Square. 174 00:20:44,960 --> 00:20:51,230 And then I see things on my desktop someplace, it's quick. 175 00:20:51,230 --> 00:20:58,270 Here. You see? 176 00:20:58,270 --> 00:21:04,410 Yes, I mean, the thing we don't want to see to people's. 177 00:21:04,410 --> 00:21:18,210 Incidents where there are. So this is our yes, you're seeing it now, thanks to make it the yeah. 178 00:21:18,210 --> 00:21:23,370 And now can everybody. See my screen? 179 00:21:23,370 --> 00:21:32,470 Yes, we can. So one of the things that, again, going back to. 180 00:21:32,470 --> 00:21:44,540 Mary and her qualities and her attitude towards things is she has this really rare, I think, combination of. 181 00:21:44,540 --> 00:21:55,070 Very, very careful attention to formal detail and the significance of formal properties and formal implications and at the same time a very, 182 00:21:55,070 --> 00:22:05,990 very deep and abiding interest in linguistic data and how linguistic data interacts with the formalism. 183 00:22:05,990 --> 00:22:16,580 And that's been very important for me, because as many of you know, and as in the introduction you saw, I'm actually not a data person. 184 00:22:16,580 --> 00:22:21,570 I'm actually as a social psychologist, not really a linguist. 185 00:22:21,570 --> 00:22:27,260 I'm sort of like passing as a linguist. 186 00:22:27,260 --> 00:22:34,910 But for me, it's just very, very important to work with them, to have worked with Mary and to work with John and other people who really, 187 00:22:34,910 --> 00:22:42,620 really are very, very good at the linguistic side of it and its relation to the formal issues. 188 00:22:42,620 --> 00:22:50,390 And so for me, that's been the source of a lot of productivity for me all through the years. 189 00:22:50,390 --> 00:23:02,060 So in this talk, formal aspects of underspecified features, I tried to find a talk that because I know that. 190 00:23:02,060 --> 00:23:08,500 A certain proportion of the audience is probably familiar with LFG and the technical issues and so forth, 191 00:23:08,500 --> 00:23:15,910 but I know this is more general audience and that not everybody has had the 192 00:23:15,910 --> 00:23:23,830 exposure but of the theory and the formalism of the issues that that brings up. 193 00:23:23,830 --> 00:23:31,840 So I thought that underspecified features was something that's common to a lot of issues, a lot of different approaches to language. 194 00:23:31,840 --> 00:23:41,390 And so I thought I would talk about that with but in the context of LFG and LFG formal. 195 00:23:41,390 --> 00:23:49,460 Notions that make it possible to tease out the different aspects of underspecified features. 196 00:23:49,460 --> 00:23:58,340 So in that context, we talk about different kinds of underspecified specification. 197 00:23:58,340 --> 00:24:01,640 It's not just a monolithic, single kind of notion. 198 00:24:01,640 --> 00:24:08,990 And I've kind of broken it out into what I'm calling just simply unspecified values, constrained values, 199 00:24:08,990 --> 00:24:20,480 unmarked values, default values and indeterminate values, different ways of underspecified, linguistic features. 200 00:24:20,480 --> 00:24:28,880 And the formal considerations in each case are what is the linguistic notion or intuition that that is being expressed? 201 00:24:28,880 --> 00:24:37,580 How is that written down in a grammar or in rules of some sort and some sort of specification? 202 00:24:37,580 --> 00:24:46,370 What are the linguistic representations that then encode of that notion of under specification? 203 00:24:46,370 --> 00:24:52,760 And then the final question I'm not going to say too much about is if it's underspecified. 204 00:24:52,760 --> 00:25:02,090 How does that get interpreted? How does the underlying significance of the specification propagate through syntax? 205 00:25:02,090 --> 00:25:11,450 So it relates to other dependencies and also into other modules of linguistics, semantics, for example, where discourse, whatever, 206 00:25:11,450 --> 00:25:17,610 and even to other computational applications, translation, search and others conversation, 207 00:25:17,610 --> 00:25:23,120 something that I've been actually thinking about the last couple of years. 208 00:25:23,120 --> 00:25:26,600 So that's the background or the setup for the talk. 209 00:25:26,600 --> 00:25:36,300 I think given that some part of the audience may not be too familiar with LFG, I'm going to go through a crash course, 210 00:25:36,300 --> 00:25:43,670 I hope for about ten minutes to explain what the fundamental formal concepts are and 211 00:25:43,670 --> 00:25:50,240 then just kind of appeal to those to explain these different kinds of specification. 212 00:25:50,240 --> 00:26:02,060 So in LFG, the starting point is structures that represent syntactic properties, and so we have what we call the constituent structure, 213 00:26:02,060 --> 00:26:12,710 the C structure, which is just the formal encoding of surface water and grouping and basically a traditional frame structure tree. 214 00:26:12,710 --> 00:26:21,660 But we also have. A separate structure, a structure, a very different formal type we call the functional structure, 215 00:26:21,660 --> 00:26:33,750 and that's a formal encoding of syntactic functions like subject and features like no singular or case, nominative, tense, present and so forth. 216 00:26:33,750 --> 00:26:45,960 So that's the the the basic fundamental linguistic representations that we assume in this syntactic theory. 217 00:26:45,960 --> 00:26:49,590 So where do these structures come from? 218 00:26:49,590 --> 00:26:56,670 How are they assigned to utterances? Well, the constituent structure is pretty traditional. 219 00:26:56,670 --> 00:27:07,020 We basically have noted disability conditions that are provided by context, free rewriting rules of traditional sort. 220 00:27:07,020 --> 00:27:16,590 And so chemistry can dominate empty because there's a rule that says that US can dominate AMPTP. 221 00:27:16,590 --> 00:27:21,990 And in fact, this little tree satisfies all the nodes, satisfy the mothers and daughters, 222 00:27:21,990 --> 00:27:28,880 satisfy this description, including the lexical categories and so forth. 223 00:27:28,880 --> 00:27:33,650 Functional structure is a little bit different. Again, it's an attribute value structure, 224 00:27:33,650 --> 00:27:43,100 but it's described by a set of equations and actually the quantified free theory of equality in the simple case. 225 00:27:43,100 --> 00:27:51,620 And so we got a set of equations that say, for example, that there's some F structure called F one, 226 00:27:51,620 --> 00:27:55,700 whose subject is an F structure that we're going to call F two. 227 00:27:55,700 --> 00:28:06,290 And if we look over here, we see that for this structure, if we call this part F one and we call this part of two, 228 00:28:06,290 --> 00:28:12,920 and in fact, F one has a subject which is labelled F to the F structure. 229 00:28:12,920 --> 00:28:20,060 Here is basically a model for this set of equations. 230 00:28:20,060 --> 00:28:23,930 Which we call the functional description or the description. 231 00:28:23,930 --> 00:28:32,550 So, again, an LFG, there's this clear separation between the structures and the descriptions of those structures. 232 00:28:32,550 --> 00:28:46,680 So the question is, how do you get that set of equations and LFG we have this notion of code description where we add annotations 233 00:28:46,680 --> 00:28:54,900 to the constituent structure rules and we use those annotations to produce the description of the structure, 234 00:28:54,900 --> 00:29:04,380 and then we check to see what a given candidate that structure, whether it satisfies all of those parts of the description. 235 00:29:04,380 --> 00:29:09,300 So the way we make that happen is we just simply assume that there's a correspondence and 236 00:29:09,300 --> 00:29:18,570 we call it to between the individual nodes of of a given C structure and the structure, 237 00:29:18,570 --> 00:29:25,000 the units of the structure that we're trying to describe and assign to this utterance. 238 00:29:25,000 --> 00:29:32,770 And the structure rule annotations then described the structure corresponding to the nodes that that rule, that myth. 239 00:29:32,770 --> 00:29:44,770 So this is the notation that we use. So we take this rule is to PvP and we have these little annotations with the up arrows and down arrows. 240 00:29:44,770 --> 00:29:58,600 And the idea is that this rule says that the F structure for the mother of the ENPI is the house as a subject, the structure of the MP. 241 00:29:58,600 --> 00:30:10,900 And in fact, when we built the correspondence function t onto this structure and we label the nodes and one and two and three, 242 00:30:10,900 --> 00:30:22,950 we're getting at the sanitation over there that says three of them to this little piece of structure is the image of this note over here. 243 00:30:22,950 --> 00:30:33,350 Under this particular Mappin. So that's kind of the setting and basically make one other move, we say, well, 244 00:30:33,350 --> 00:30:41,270 we're going to replace we're going to assume that the up arrow stands for the fee of the mother node that matches a given group pattern. 245 00:30:41,270 --> 00:30:47,180 Down Arrow stands for fee of the daughter node that it's attached to. 246 00:30:47,180 --> 00:30:57,070 And in this little configuration here with that rule laid out for the NAACP, can we label the nodes so we can talk about them? 247 00:30:57,070 --> 00:31:02,700 And under that correspondence, they're attached to the structure in that way. 248 00:31:02,700 --> 00:31:10,400 And we simply replaced the up and down arrows by what they instantiate to. 249 00:31:10,400 --> 00:31:13,760 And then because we don't need to keep track of this relationship anymore, 250 00:31:13,760 --> 00:31:22,290 we simplify by just naming them F1, F2 and so forth, and we get a set of equations in that way. 251 00:31:22,290 --> 00:31:30,710 And then the interpretation is that the rule admits an S over an MP VPI. 252 00:31:30,710 --> 00:31:38,290 And the F unit F two that corresponds to the MP note is the nominative subject of. 253 00:31:38,290 --> 00:31:51,010 The unit corresponding to the S.A. and the DP and the S.A. map to the same structure and that kind of 254 00:31:51,010 --> 00:31:59,620 map in that equation up close down is basically the way that heads and cokeheads are annotated and. 255 00:31:59,620 --> 00:32:09,430 So then we get a grammar fragment that just basically has these kinds of annotations propagated around and into the lexicon as well. 256 00:32:09,430 --> 00:32:16,890 But they're always uniformly assigned an interpretation with respect to a particular tree. 257 00:32:16,890 --> 00:32:21,480 So, again, the structure is a model of its equations. 258 00:32:21,480 --> 00:32:22,260 What does that mean? 259 00:32:22,260 --> 00:32:32,700 It means that we have structure to satisfy all the equations in this F description that we read out of the tree, given the limitations. 260 00:32:32,700 --> 00:32:38,270 And so if we had an ungrammatical sentence like John likes, they. 261 00:32:38,270 --> 00:32:52,370 Which is a pronoun is a nominative pronoun, and we add it to the description of the state of equations that we got from the structure, 262 00:32:52,370 --> 00:33:00,560 the fact that they is in the object position and with all these efforts and so forth instantiated, 263 00:33:00,560 --> 00:33:10,460 we can infer that F Seven's case is equal to accusative and F Seven's case is nominative. 264 00:33:10,460 --> 00:33:17,060 And that is a conflict because nominative is not equal to Cusato by and the conflict 265 00:33:17,060 --> 00:33:22,340 comes from the transitivity in the substitution of the axioms of equality. 266 00:33:22,340 --> 00:33:27,890 So we have inconsistent equations and this description, therefore has no model. 267 00:33:27,890 --> 00:33:33,380 And that's why we judge the sentence to be ungrammatical. 268 00:33:33,380 --> 00:33:39,090 By the way, can people see the mouse when I move it? Does that show up? 269 00:33:39,090 --> 00:33:44,500 And. No, no, I don't think it does. 270 00:33:44,500 --> 00:33:50,050 OK, well, then then I'll be careful. Thank you. 271 00:33:50,050 --> 00:33:57,520 OK, so we want to make one other assertion, namely that if Stretch is not only a model, but it's a minimal model. 272 00:33:57,520 --> 00:34:07,660 So the structure has to satisfy all the equations. And we have a description, but it also has to be the case that no smaller structure satisfies them. 273 00:34:07,660 --> 00:34:11,980 So if we took enough structure that we had before and we simply say, well, 274 00:34:11,980 --> 00:34:22,630 that structure could also why not have tents past the certainly compatible with all of the equations that we have? 275 00:34:22,630 --> 00:34:32,820 It's just extra information. Well, we can remove that piece of information and get a smaller structure, and that would still satisfy. 276 00:34:32,820 --> 00:34:39,720 The equations, the description, so this structure with tense past is not minimal, 277 00:34:39,720 --> 00:34:50,470 and therefore it is not the structure that the phrase structure grammar with the instantiated annotations. 278 00:34:50,470 --> 00:34:59,430 Well. Produce a description of and so this is not the structure of this sentence. 279 00:34:59,430 --> 00:35:12,120 Now, I just mentioned that this notion of a minimal model analogy is equivalent to the notion of fully licence that you see in construction grammar. 280 00:35:12,120 --> 00:35:17,790 And it's just whether you're thinking of it from the point of view of the description or the point of view of the structure. 281 00:35:17,790 --> 00:35:25,470 But basically unrequited extensions in LFG would be unlicenced in construction grammar. 282 00:35:25,470 --> 00:35:37,440 So these linguistic intuitions that we're talking about, I think are not just local to LFG, but in fact are shared across different theories. 283 00:35:37,440 --> 00:35:44,070 So you can say one other thing, which about the description, which we allow destruction. 284 00:35:44,070 --> 00:35:47,250 And when you have destruction in the equations, 285 00:35:47,250 --> 00:35:56,660 you get a set of alternative description descriptions because we basically convert this to a disjunctive, normal form. 286 00:35:56,660 --> 00:36:06,110 And each of those descriptions then must be a model where we test it to see if it is a model, if it is consistent and if it has a minimal model, 287 00:36:06,110 --> 00:36:15,380 and sometimes we get more than one satisfied, viable alternative description, in which case we get multiple models. 288 00:36:15,380 --> 00:36:20,480 And in that case, we recognise that the sentence is ambiguous. 289 00:36:20,480 --> 00:36:28,970 And so this is a classic ambiguity that chicken is ready to eat, whether or not the chicken is hungry or the chicken's cooked and ready. 290 00:36:28,970 --> 00:36:38,630 And its lexical entry would have a disjunction, saying that its subject controls the subject of the compliment, 291 00:36:38,630 --> 00:36:42,920 where the subject controls the object of the compliment. 292 00:36:42,920 --> 00:36:55,010 And that would give us the two different descriptions, one of which we have this five XCOM subject, the other we have had five XCOM object. 293 00:36:55,010 --> 00:37:05,750 And from that we get the two different models, one in which chicken is both the subject of the main clause and the subject. 294 00:37:05,750 --> 00:37:11,660 Both a compliment and the other is the subject to the main clause and the object of the compliment, 295 00:37:11,660 --> 00:37:18,590 and with the different constructions, pronouns are thrown in from the construction itself. 296 00:37:18,590 --> 00:37:23,180 So the can we have models with structures? 297 00:37:23,180 --> 00:37:27,110 We have models. We have structures. 298 00:37:27,110 --> 00:37:30,830 We have descriptions. We have models. We have minimal models. 299 00:37:30,830 --> 00:37:36,950 And we have destruction in this kind of architecture. 300 00:37:36,950 --> 00:37:43,490 And this has been around actually since the late 70s. 301 00:37:43,490 --> 00:37:50,030 So with some embellishments and a little bit more explication over the years 302 00:37:50,030 --> 00:37:57,800 from what I've defined in and around nineteen seventy eight and one remark, 303 00:37:57,800 --> 00:38:10,760 I think because that's a long time and linguistic theory years for something to be basically stable and basically still informative and useful and, 304 00:38:10,760 --> 00:38:19,490 and actually the subject of a lot of exploration and a lot of linguistic description and analysis. 305 00:38:19,490 --> 00:38:25,700 So that way I think it's been quite successful. So to summarise, we have the different types, 306 00:38:25,700 --> 00:38:32,640 the different types of structures that have different types of descriptions that are related through this code, 307 00:38:32,640 --> 00:38:34,400 description, idea, 308 00:38:34,400 --> 00:38:45,980 this notion of a structural correspondence like the and we have structures that are models of the descriptions and in particular the minimal models. 309 00:38:45,980 --> 00:38:56,970 Now, in doing this, we want to we worry about complexity and trade-offs and complexity as we're doing linguistic analysis. 310 00:38:56,970 --> 00:39:09,300 So we worry about the trade offs between cluttering up the structures, complicating the descriptions and enriching the formalism. 311 00:39:09,300 --> 00:39:16,060 And under specification itself is interesting because the goal of it is to reduce the clutter. 312 00:39:16,060 --> 00:39:23,200 However, it's distributed so the explanations become more transparent, 313 00:39:23,200 --> 00:39:30,310 and that's why this is from a conceptual point of view, an interesting thing to think about. 314 00:39:30,310 --> 00:39:35,980 So, again, these are the varieties that I'm going to kind of go through and not in great detail, 315 00:39:35,980 --> 00:39:44,710 but try to give you the intuitions and a little bit of a formal apparatus that goes along with those intuitions. 316 00:39:44,710 --> 00:39:52,320 So unspecified values and this is the most uncluttered. 317 00:39:52,320 --> 00:39:53,940 Kind of under specification, 318 00:39:53,940 --> 00:40:04,800 and the notion is that any value is possible for a future and in the description no value is specified and in the F structure, 319 00:40:04,800 --> 00:40:16,300 no attribute or value appears. And the syncretic fish and sheep in English are. 320 00:40:16,300 --> 00:40:23,260 The trivial examples of this, where neither fish nor sheep are marked for no, 321 00:40:23,260 --> 00:40:34,200 they're unspecified and the so the description would not say that number has any particular value. 322 00:40:34,200 --> 00:40:40,410 And the structure that goes along with that description. 323 00:40:40,410 --> 00:40:49,020 Knows that it's tough and knows that one is if one is accused of from the construction that they appear in in this sentence, 324 00:40:49,020 --> 00:40:58,650 but there is no marking of no. So the fact that these two credit forms of. 325 00:40:58,650 --> 00:41:04,920 Don't don't specify number is simply just. 326 00:41:04,920 --> 00:41:14,400 Omitted the future just doesn't appear at all now that the game here is that we don't particularly want 327 00:41:14,400 --> 00:41:21,600 to include either in the grammar rules or lexical entries or in the structures or in the descriptions, 328 00:41:21,600 --> 00:41:32,760 values that are simply uninformative. Now, there's a distinctive alternative, obviously, and that's more cluttered. 329 00:41:32,760 --> 00:41:39,570 So suppose we knew from some separate declaration a universal of grammar. 330 00:41:39,570 --> 00:41:48,270 For example, the number could take on the value, singular and plural and voice, could take on the values, active and passive and so forth. 331 00:41:48,270 --> 00:41:55,620 And to represent that this is these are Socratic forms, then syncretic forms, we would say, OK, 332 00:41:55,620 --> 00:42:06,570 this is just a function that either the number of fish is singular or the number is plural and the same thing for cheap. 333 00:42:06,570 --> 00:42:16,710 And when we do that and we multiply things out to these alternative descriptions and destructive normal form, 334 00:42:16,710 --> 00:42:25,100 we basically get four different structures that correspond to those four different just descriptions. 335 00:42:25,100 --> 00:42:39,770 So we've replaced vagueness where the feature is not specified at all anywhere to a four way ambiguity, but in fact there's no more information. 336 00:42:39,770 --> 00:42:48,980 It's just multiply it out in a more explicit way and again, across through a comparison. 337 00:42:48,980 --> 00:42:56,990 And there's a requirement that these feature structures are complete and what they 338 00:42:56,990 --> 00:43:03,500 call sort of and therefore it would push towards ambiguity instead of vagueness, 339 00:43:03,500 --> 00:43:11,100 because every proper representation has to have every feature instantiate. 340 00:43:11,100 --> 00:43:21,760 And LFG, we tried to. Suppressed that again, so the important stuff shows up. 341 00:43:21,760 --> 00:43:28,990 So and I don't know if you those of you who've seen. 342 00:43:28,990 --> 00:43:40,840 Heg, future structures, they really are quite enormous and partly that's because of the sort solving that we're talking about here, 343 00:43:40,840 --> 00:43:51,220 but also because in that theory they take it as they have an advantage that all the linguistic information, 344 00:43:51,220 --> 00:43:55,300 syntactic information can be encoded in a single structure, 345 00:43:55,300 --> 00:44:04,730 that you can encode precedence and dominance, relationships that we could separately in the tree. 346 00:44:04,730 --> 00:44:13,250 With these future properties and it's kind of an Occam's Razor argument that says, well, 347 00:44:13,250 --> 00:44:19,250 if you can coated in one formal system, that's an advantage from the point of view. 348 00:44:19,250 --> 00:44:25,880 We have a different intuition about it, which is let's separate these things out. 349 00:44:25,880 --> 00:44:32,330 And hopefully the interactions between them will be relatively few compared to the interactions 350 00:44:32,330 --> 00:44:40,160 between the elements of the sea structure internally and the elements of the structure internally. 351 00:44:40,160 --> 00:44:50,420 So that's kind of a conceptual framing of our value statement, I guess, of what these theories tend to go for. 352 00:44:50,420 --> 00:45:02,660 So now when you do have these completely unspecified features, they can take on particular values determined by the context that they appear in. 353 00:45:02,660 --> 00:45:05,840 So in the context of when the fish eats the sheep, 354 00:45:05,840 --> 00:45:16,100 where the no agreement from eats imposes no singular on its subject, then we would expect to find in the structure. 355 00:45:16,100 --> 00:45:24,770 We do find that the fish now is marked as no singular, not because of its inherent marking, 356 00:45:24,770 --> 00:45:31,340 which was unspecified, but because the context imposed that marking. 357 00:45:31,340 --> 00:45:42,150 Whereas in the case of eight, where it's not marked at all, that feature would not appear because the minimal model would not contain it. 358 00:45:42,150 --> 00:45:49,270 Now we also get the case that these kind of inferred values. 359 00:45:49,270 --> 00:45:57,130 May clash by virtue of transitivity, so we have this ungrammatical sentence, 360 00:45:57,130 --> 00:46:09,940 the fish that eat swims where there seems to be an agreement between the verb of the clause, the relative clause and the main clause. 361 00:46:09,940 --> 00:46:19,000 But that's a consequence of the fact that they are both imposing information on fish, 362 00:46:19,000 --> 00:46:27,700 which itself is unspecified, but in but these two assertions basically assert conflicting values. 363 00:46:27,700 --> 00:46:34,120 And the sentence therefore doesn't have a model structure. The description doesn't have a model. 364 00:46:34,120 --> 00:46:44,290 So there have been also some ways of dealing with these kinds of syncretic situations by introducing a new class of values, 365 00:46:44,290 --> 00:46:47,960 pushing the destruction into the structure. 366 00:46:47,960 --> 00:46:59,130 And Larry Carter made one of the early proposals for this, so you might think of the value of number four fish is. 367 00:46:59,130 --> 00:47:09,630 A special value was singular or plural, and then you don't get the multiplication that's added low, 368 00:47:09,630 --> 00:47:17,040 but you still know what the value possibilities are in the easy case, 369 00:47:17,040 --> 00:47:25,830 this is limited to very simple values, doesn't obviously extend to where the alternatives have to do with different 370 00:47:25,830 --> 00:47:33,540 relations being assigned and the interaction of more intricate value dependencies. 371 00:47:33,540 --> 00:47:43,920 And it also has a feeling of mixing up the true theory of the computation that you do to verify things with the model theory, 372 00:47:43,920 --> 00:47:49,500 which is just to look at the results and see what you have. 373 00:47:49,500 --> 00:48:00,930 This kind of representation is very good for computation. I don't think it's so good for linguistics and there is a computational interpretation of 374 00:48:00,930 --> 00:48:06,060 all this that is sound incomplete in the work that John Max when I did over the years. 375 00:48:06,060 --> 00:48:10,980 And it's the kind of thing that's actually needed for high speed passing your generation. 376 00:48:10,980 --> 00:48:18,690 It's basically a way of converting the descriptions from destructive normal form to conjunctive normal form. 377 00:48:18,690 --> 00:48:25,890 But again, I don't think it's good for linguistic explanation in particular. 378 00:48:25,890 --> 00:48:34,140 So now I'm going to turn to a next kind of understood figure, but I'm calling constrained under specification. 379 00:48:34,140 --> 00:48:43,200 So the notion here is that some, but not all possible values are allowed. 380 00:48:43,200 --> 00:48:47,340 So it's a proper subset of the declared value. 381 00:48:47,340 --> 00:48:54,710 So case, for example, that has accusative, nominative and genitive. 382 00:48:54,710 --> 00:49:05,410 Well. And you have a word like that, I like it, that is either unmarked or ambiguous with respect to accusative. 383 00:49:05,410 --> 00:49:15,130 And nominative appeared an object on such a position, but it cannot appear as a possessive, so it's not possible for him to be. 384 00:49:15,130 --> 00:49:21,740 For it to be. Genitive. 385 00:49:21,740 --> 00:49:31,100 So how can we describe that? Well, we can have a disjunction of possible values, so in its lexical entry, 386 00:49:31,100 --> 00:49:39,290 it says that its case is accusative or its case is nominative and an equivalent logically 387 00:49:39,290 --> 00:49:45,680 equivalent to a little bit more succinct formulation of that is to say that its case is a member 388 00:49:45,680 --> 00:49:54,110 of the accused it of the nominative and then the axiom of extension and so forth would basically 389 00:49:54,110 --> 00:50:01,280 say that the destruction is an immediate consequence of that more compact representation. 390 00:50:01,280 --> 00:50:10,700 And that's just saying that we can add step membership and sets to the description language in a really easy and obvious way. 391 00:50:10,700 --> 00:50:19,940 And it lets us to express things a little bit more succinctly with a little bit less repetition and clutter. 392 00:50:19,940 --> 00:50:28,100 So give him that description, the f structure will have specific values in the minimal models. 393 00:50:28,100 --> 00:50:34,830 And again, depending on the context, there may be several models, but maybe not all possible models. 394 00:50:34,830 --> 00:50:39,120 So we have a notion of restricted ambiguity. 395 00:50:39,120 --> 00:50:51,360 Now, an alternative to that is to use negative constraints to simply say some values are not possible, like genitive is not possible. 396 00:50:51,360 --> 00:50:59,700 So instead of saying what values are possible in enumerating them, we can enumerate the values that are to be excluded. 397 00:50:59,700 --> 00:51:04,110 Now, that may be a smaller set, easier to describe and so forth. 398 00:51:04,110 --> 00:51:19,000 So we might say that. Simply by writing not equal and adding that to our description, language that the case is not genitive. 399 00:51:19,000 --> 00:51:26,410 Instead of saying that the case is a member of the most succinct form of the accusative and nominative and we the structure, 400 00:51:26,410 --> 00:51:32,670 then perhaps there would be no value in the minimal model unless it's required by other constraints. 401 00:51:32,670 --> 00:51:43,320 And the interpretation of this is in the minimal model itself, and the interpretation is that if the feature case is present, 402 00:51:43,320 --> 00:51:50,020 it may not be, but if it is present, it must not be genitive. 403 00:51:50,020 --> 00:51:56,170 So a negative specification here is interpreted in the mental model. 404 00:51:56,170 --> 00:52:07,140 As opposed to putting in all the ways in which. Case could not be genitive, which would revert to the explicit enumeration again, 405 00:52:07,140 --> 00:52:17,880 so this negative specification is brings brings into view the significance of 406 00:52:17,880 --> 00:52:27,000 the minimal model in the interaction between descriptions and satisfy ability. 407 00:52:27,000 --> 00:52:34,950 But what if you wanted to say that either the feature must be present? 408 00:52:34,950 --> 00:52:44,310 Or it must be absent. But in the constrained here, we don't where we're setting this constraint, we actually don't care what the values are. 409 00:52:44,310 --> 00:52:51,380 We don't want to enumerate them at all. Let somebody else someplace else in the grammar to the enumeration. 410 00:52:51,380 --> 00:53:04,100 So, again, we have John likes Mary. And in terms of the categories John liking Mary, what would have the same sea structure? 411 00:53:04,100 --> 00:53:09,380 But it's a participle and and not it. Not a text form. 412 00:53:09,380 --> 00:53:13,190 So how would we describe that? 413 00:53:13,190 --> 00:53:21,380 Well, one way is to clutter up the description and the structure structure by adding basically a complement of values, 414 00:53:21,380 --> 00:53:31,040 a new value that is inconsistent with any of the other values that are allowed for that feature. 415 00:53:31,040 --> 00:53:38,240 So for likes, we could say that it's Tensas Prison, but it is not a participle. 416 00:53:38,240 --> 00:53:45,730 And for liking, we could say that it's a participle. 417 00:53:45,730 --> 00:53:57,250 But it's not a tense form, and in the grammar, we could specify because we know that the VPs in essence, are supposed to be tensed. 418 00:53:57,250 --> 00:54:01,180 We can say that the tense is not mom, not none. 419 00:54:01,180 --> 00:54:11,830 Or we could say with the participle is none. And either way, we would exclude the John liking Mary and get a tense form. 420 00:54:11,830 --> 00:54:23,560 But we would also get it with this not unclutter description, but cluttered f structure where at the top level we're saying, yeah, it is tense. 421 00:54:23,560 --> 00:54:32,860 But also, by the way, we're recording the fact that it's not a participle, which is kind of a silly thing to see. 422 00:54:32,860 --> 00:54:36,520 So there's an alternative to that. 423 00:54:36,520 --> 00:54:45,240 That again, goes up this clustering idea. And here we're talking about what we call existential constraints. 424 00:54:45,240 --> 00:54:53,550 So, again, the description of the structure, we want absolute values to be absent. 425 00:54:53,550 --> 00:55:00,060 So we would like likes to just say that it's tenses present and liking just to say 426 00:55:00,060 --> 00:55:07,020 that it's a present participle and not get mixed up one feature with the other, 427 00:55:07,020 --> 00:55:15,540 but set up the description. So again, extending the description language to minimise the descriptions and the structures. 428 00:55:15,540 --> 00:55:28,980 So setting up the description language so that we can specify with notation the requirements that we want to impose on the minimal model. 429 00:55:28,980 --> 00:55:42,500 So if we write just how tense a designator for the future without a value, we interpret that that the minimal model must have some value for tense. 430 00:55:42,500 --> 00:55:47,260 If we write. And we wouldn't do this before. 431 00:55:47,260 --> 00:55:56,990 But with the negation sign, we say that the participle does not exist in the middle one minimal model. 432 00:55:56,990 --> 00:56:04,630 We can write that as well. But of course, our preference in this situation would be to just say that we want it to be tense, 433 00:56:04,630 --> 00:56:07,990 to not worry about all the other things that could have been. 434 00:56:07,990 --> 00:56:13,660 So the idea is that the description of language itself is extended with these existential constraints. 435 00:56:13,660 --> 00:56:18,400 They apply to the minimal model of all the other equations. 436 00:56:18,400 --> 00:56:25,170 And if the constraint is positive, it's satisfied if the minimal model contains an attribute. 437 00:56:25,170 --> 00:56:33,930 That attribute with any value and if it's negative, the model's OK, if that does not contain. 438 00:56:33,930 --> 00:56:47,350 An attribute at all. So this is part of the election strategy, the media strategy is to enrich descriptions and the description language. 439 00:56:47,350 --> 00:56:55,300 In service of expressing abstract notions that seem to have linguistic significance and be able to 440 00:56:55,300 --> 00:57:05,340 maintain the simple structures and even as the simple way in which the description is deployed. 441 00:57:05,340 --> 00:57:14,810 So. And turning to another kind of unmarked value that I call the conventionally unmarked values, 442 00:57:14,810 --> 00:57:18,890 and these are more like an analogy to what you see in phrenology, 443 00:57:18,890 --> 00:57:28,980 where you know that if Nassau's an unmarked feature, then there's a convention that says at some point, you know, that that means. 444 00:57:28,980 --> 00:57:39,390 That it's positive, so the notion here is that no future value means a particular one, presumably one that is. 445 00:57:39,390 --> 00:57:48,190 Unexceptional or common or typical, not worth specifying, but you know that it's their. 446 00:57:48,190 --> 00:57:54,760 And again, the idea is the destruction you have to scriptura not cluttered with these unexceptional values. 447 00:57:54,760 --> 00:58:00,230 An example here is the fact that in English, the first and second person. 448 00:58:00,230 --> 00:58:12,700 Are exceptional, the person barking only applies. To pronouns, but all the other nouns in the world are unexceptional in third person, 449 00:58:12,700 --> 00:58:20,100 so why should we have to always specify everywhere that nouns are third person? 450 00:58:20,100 --> 00:58:30,690 So for fish, we would have to write if we had to specify that the person is third. 451 00:58:30,690 --> 00:58:36,360 But we'd rather just say, look, we know the. That person is not there. 452 00:58:36,360 --> 00:58:40,800 It means that it's third if anybody ever cares. 453 00:58:40,800 --> 00:58:53,790 So basically in syntactic dependencies that might apply to missing values, we can use existential constraints to code these marking conventions. 454 00:58:53,790 --> 00:59:01,410 So can the unmarked value here. The convention is that in this particular case, if you don't see a person. 455 00:59:01,410 --> 00:59:10,200 It acts as if the person had been specified as third, but we don't need to say that it goes without saying so, 456 00:59:10,200 --> 00:59:14,220 we have a problem, obviously, with subject for agreement in English. 457 00:59:14,220 --> 00:59:22,490 So it says that either it's marked for person. 458 00:59:22,490 --> 00:59:33,020 Which would be true for the first and second persons and not true for the third or as plural eat, so eat and eat. 459 00:59:33,020 --> 00:59:40,580 I would say it's not marked for purpose, which means a third. 460 00:59:40,580 --> 00:59:56,950 And it's singular. So, again, the structure doesn't have to record third person everywhere, ordinary nouns, we're using this disjunction and. 461 00:59:56,950 --> 01:00:04,960 Existential, positive and negative in the description language itself to code the intuition that goes with a more explicit feature. 462 01:00:04,960 --> 01:00:12,910 So the structure of consumer semantics must be informed if it matters. 463 01:00:12,910 --> 01:00:18,700 And if we don't actually fill in the interface, 464 01:00:18,700 --> 01:00:26,080 that the interface doesn't actually fill in the value according to a marking convention that stated off to the side. 465 01:00:26,080 --> 01:00:34,270 But the grammar itself has embedded within it by the use of the way it refers to person. 466 01:00:34,270 --> 01:00:40,410 Assigning a particular meaning to the absence of a value. 467 01:00:40,410 --> 01:00:45,230 Now, that's a little bit different from but it has a feeling like default values. 468 01:00:45,230 --> 01:00:51,870 So here the notion is that an exception on unexceptional values always appear, 469 01:00:51,870 --> 01:00:58,620 even if it's hard to find a place in the grammar where you would specified. 470 01:00:58,620 --> 01:01:05,100 So it may be easier to find the place, but it may mean that you have to specify them everywhere. 471 01:01:05,100 --> 01:01:15,430 Whenever you're specifying another value, you have to have an alternative that says if that one isn't taken, then stick in the default. 472 01:01:15,430 --> 01:01:25,420 So the idea is here we're getting an uncluttered description with a single statement that assigns the default if no value was assigned elsewhere, 473 01:01:25,420 --> 01:01:34,090 and there isn't a need to add disjunctive alternatives at all the places where other values are assigned. 474 01:01:34,090 --> 01:01:43,210 So the description is uncluttered, but here we're not interested in uncovering the structure. 475 01:01:43,210 --> 01:01:50,480 In this case, we we have the minimal model. Always will contain a future value. 476 01:01:50,480 --> 01:01:54,800 It will always be filled in in the middle animal model. 477 01:01:54,800 --> 01:01:58,500 So how do we do that again? 478 01:01:58,500 --> 01:02:09,570 We're using existential constraints and destruction. So we want to assign the nominal nominative as the default case. 479 01:02:09,570 --> 01:02:15,710 That would be the assertion that either the case exists in the A model, that is, 480 01:02:15,710 --> 01:02:28,330 it was assigned as accusative or nominative or whatever it was, or dative or gender or whatever someplace else and some other construction or. 481 01:02:28,330 --> 01:02:37,970 We're going to be looking at a model where we can assert that the case itself is nominative and if it was filled in. 482 01:02:37,970 --> 01:02:43,520 Elsewhere to some other value, that model, that that description will fail. 483 01:02:43,520 --> 01:02:52,190 If it wasn't filled in elsewhere, that minimal model will now have the case nominative filled in. 484 01:02:52,190 --> 01:02:58,410 So we end up with two have descriptions to animal models possibly. 485 01:02:58,410 --> 01:03:07,350 Again, the description of the accident with the existential will fail if there is no separate assignment someplace else for case, 486 01:03:07,350 --> 01:03:12,260 the description with the quality will fail somewhere else. 487 01:03:12,260 --> 01:03:19,450 There's an incompatible assignment, and otherwise mom is just filled in. 488 01:03:19,450 --> 01:03:27,220 So one of these descriptions is always satisfied, there's never a failure that comes from assigning a default feature, 489 01:03:27,220 --> 01:03:29,710 and that's kind of the meaning of what it means to be a default. 490 01:03:29,710 --> 01:03:42,220 You can always throw it in and it kind of lfg throwing it in at the end is formalised as testing against the minimal model. 491 01:03:42,220 --> 01:03:54,840 So to compact the notation a little bit with over the years introduced a template that just wraps up. 492 01:03:54,840 --> 01:04:03,090 Under a name default, this kind of destruction in the way it operates so that we can simply write the default value 493 01:04:03,090 --> 01:04:13,920 from this case and that expands out to either there is a case or the case has that value. 494 01:04:13,920 --> 01:04:19,260 Now, the default kind of is like unmarked, but it's not. 495 01:04:19,260 --> 01:04:23,040 The values for default always appear. 496 01:04:23,040 --> 01:04:30,670 The values for unmarked never appear, and the absence of value has a meaning. 497 01:04:30,670 --> 01:04:35,950 When we're crossing it, module interfaces, it may be that these. 498 01:04:35,950 --> 01:04:42,600 Converge because the marketing conventions apply to. 499 01:04:42,600 --> 01:04:52,540 To fill out the unmarked case, even if it wasn't really necessary throughout the grammar. 500 01:04:52,540 --> 01:04:58,540 So there's one other aspect in the formalisation of defaults, 501 01:04:58,540 --> 01:05:08,280 which I describe as the separation of the terms with the notation scope and the value scope of the default. 502 01:05:08,280 --> 01:05:15,060 So the notation scope has to do with where you can write down these up arrows and 503 01:05:15,060 --> 01:05:24,400 down arrows and what they refer to to add the default specification to the grammar. 504 01:05:24,400 --> 01:05:31,080 And in all of you, you have to write them somewhere. So that's the notation scope. 505 01:05:31,080 --> 01:05:32,070 And again, 506 01:05:32,070 --> 01:05:40,620 some of the theories there are these metter principles that operate that are kind of universally quantified statements over the feature structure. 507 01:05:40,620 --> 01:05:49,620 So we might say, well, no matter how the feature structure rises for all units of our feature structure, 508 01:05:49,620 --> 01:05:59,920 if it has some property, it has some other property. We don't have that in our and theories that do have that kind of formulation. 509 01:05:59,920 --> 01:06:00,630 HPLC, 510 01:06:00,630 --> 01:06:12,390 I think of as one of them that can introduce a lot of hard mathematical issues to side ability issues when you do these universal qualifications. 511 01:06:12,390 --> 01:06:20,040 So we avoid that kind of thing because we really do want to have concrete computational behaviour. 512 01:06:20,040 --> 01:06:30,150 So that's the notation scope, the value scope. Is the fact that explicit values can propagate? 513 01:06:30,150 --> 01:06:46,570 By transitivity or substitution of equality from one place in the structure to another, and that describes the scope of the default. 514 01:06:46,570 --> 01:06:58,420 So in this little diagram here, we might specify inside the noun phrase that the case defaults to nominative. 515 01:06:58,420 --> 01:07:09,210 But the value spoke and read the value nominative could be supplied from any construction anywhere. 516 01:07:09,210 --> 01:07:20,160 And it would possibly fill in that value and so. 517 01:07:20,160 --> 01:07:30,710 The default is kind of a global anywhere in the middle, a model that's going to be established, but we might want to lower the value scope. 518 01:07:30,710 --> 01:07:38,960 So that this default has to be resolved within the boundaries. 519 01:07:38,960 --> 01:07:41,470 You basically say. 520 01:07:41,470 --> 01:07:55,480 If the noun phrase internally and it's the description of the nodes in that structure with its up arrows and counters doesn't provide value for case, 521 01:07:55,480 --> 01:08:04,660 then at this point we're going to assert that case because there is effectively the idea almost 522 01:08:04,660 --> 01:08:12,750 as that there's a minimal model within a constituent where we can evaluate these constraints. 523 01:08:12,750 --> 01:08:22,130 These defaults and then add the values there and now if the scope above. 524 01:08:22,130 --> 01:08:33,870 The structure above conflicts, it's an incompatible value, there will be a failure as opposed to, oh, we simply won't use the default. 525 01:08:33,870 --> 01:08:46,770 So in the embeddedness, notion of default is the is the domain in which the default is activated, 526 01:08:46,770 --> 01:08:55,710 and we might want to say, for example, that some features are. 527 01:08:55,710 --> 01:08:59,160 Receive their default value, their lexical defaults, 528 01:08:59,160 --> 01:09:13,060 and the fact it's instantiated to the lexical default value when the lexicon is exited and therefore it's established and can be checked, 529 01:09:13,060 --> 01:09:22,560 for example, in the higher, higher scope. So there are variations on this. 530 01:09:22,560 --> 01:09:30,570 And the issue in the formalism is how do you specify something that you'd like to be able to control? 531 01:09:30,570 --> 01:09:40,080 And again, in the trade off between, you could cut up lots of features that propagate the scope and where you are and what you're doing. 532 01:09:40,080 --> 01:09:40,800 But again, 533 01:09:40,800 --> 01:09:55,390 we've realised that in a built in a predicate where you can simply write down the case must be complete in the noun phrase with interpretation. 534 01:09:55,390 --> 01:10:08,650 If if the case feature doesn't exist. Within the noun phrase, it will never be specified within the noun phrase and the default applies, 535 01:10:08,650 --> 01:10:15,810 it becomes normative and that's the way that feature is seen in any higher context. 536 01:10:15,810 --> 01:10:21,360 So the last thing I want to mention is what we call indeterminate values, 537 01:10:21,360 --> 01:10:33,370 and that's the really challenging case for equality and unification based theories, because that's the case where transitivity fails. 538 01:10:33,370 --> 01:10:46,460 And the problem is that the same form satisfies appears to satisfy conflicting feature requirements on the classic case against. 539 01:10:46,460 --> 01:10:54,290 The German treaty relative and I don't speak German, only vaguely going to pronounce this, but the. 540 01:10:54,290 --> 01:11:06,640 So here we have a free relative where the relative pronoun boss is the object of the Ferguson and the subject of Goodrich. 541 01:11:06,640 --> 01:11:20,860 And in the one case, there's the specifications accuse it of when it's the object and then the other, it's the specific specified as being nominative. 542 01:11:20,860 --> 01:11:31,120 And obviously that's going to fail and getting rid of the arrows and the F1 and all that, I'll just substitute the verbs to keep track here. 543 01:11:31,120 --> 01:11:44,330 So basically it all comes down to the Gessen's objects cases accused of on the other side, the subject cases nominative. 544 01:11:44,330 --> 01:11:50,270 The pronoun is both the object of one and the subject of the other. 545 01:11:50,270 --> 01:12:01,410 Now you do substitution of equality and transitivity, so you take each of those in middle equality, substitute them in the. 546 01:12:01,410 --> 01:12:08,700 Case features and you end up with the best cases of Cusato accusative and its sequel, 547 01:12:08,700 --> 01:12:16,140 The Nominative, which implies that nominative and sequel to Accusative, and that's not true. 548 01:12:16,140 --> 01:12:24,100 So. The equality here seems to be too strong because we'd like this to be grammatical, 549 01:12:24,100 --> 01:12:32,350 and yet the machinery rules it out now it's the equality and the transitivity that's doing the damage here. 550 01:12:32,350 --> 01:12:43,720 It's in the construction itself. That really doesn't matter if, how, if or how we specify or under specify the properties of the pronoun. 551 01:12:43,720 --> 01:12:49,960 So we could say that it's destructively OK for it to be nominative or accusative. 552 01:12:49,960 --> 01:12:54,610 We can not say anything. We could say that the case belongs to the set. 553 01:12:54,610 --> 01:12:59,140 We can say that the case is not genitive for data or whatever we want to say. 554 01:12:59,140 --> 01:13:01,320 All of that doesn't matter. 555 01:13:01,320 --> 01:13:13,360 Well, because it's not really the fact that the pronoun which is underspecified in a way because it cannot be data and so forth, 556 01:13:13,360 --> 01:13:19,660 but that's not the source of the problem. The problem is equality itself. 557 01:13:19,660 --> 01:13:31,090 And so Mary and I actually took on this challenge in the 90s and we came up with a solution. 558 01:13:31,090 --> 01:13:38,830 And the idea was to replace atomic values and the transitivity of equality that go with them with 559 01:13:38,830 --> 01:13:47,260 stats and set membership for the class of these features that can be indeterminate in this way. 560 01:13:47,260 --> 01:13:58,690 So the value specification is to say that the case is the nominative accusative, as opposed to saying that it's a member of the set. 561 01:13:58,690 --> 01:14:10,690 But the value checking is done, done by membership and then the structure you have a value which is in fact a set value in the structure itself. 562 01:14:10,690 --> 01:14:24,950 So if we looked at this before, well, now we're going to say that the case is equal to nominative accusative before the succinct form was membership. 563 01:14:24,950 --> 01:14:30,590 Before we tested by saying that the value was accused of. 564 01:14:30,590 --> 01:14:38,550 Now we're going to test by checking to see whether the value that we're interested in is a member of the. 565 01:14:38,550 --> 01:14:45,600 So now we look at that little deduction where we've gotten the membership instead of the quality, 566 01:14:45,600 --> 01:14:51,360 and we end up when we distribute the equality and do the transitivity, 567 01:14:51,360 --> 01:14:58,290 just that that set has two different members, which is perfectly fine because that's what it is. 568 01:14:58,290 --> 01:15:02,610 But it only has just those particular members. 569 01:15:02,610 --> 01:15:12,620 So if anywhere else. Was asserted that the case was genitive testing in that way. 570 01:15:12,620 --> 01:15:19,720 That would be unsatisfying as it should be. Now. 571 01:15:19,720 --> 01:15:29,560 And in this setup, grammatical functions and many other features, the determinate ones we can call them, still propagate with equality. 572 01:15:29,560 --> 01:15:38,500 There are a lot of other interactions with other kinds of constructions, with coordination, with modification and so forth, 573 01:15:38,500 --> 01:15:49,120 that just the basic idea has to interact with and in some context is in determining values. 574 01:15:49,120 --> 01:15:56,350 Canaro and that's work that Mary and Tracy Sadler worked on. 575 01:15:56,350 --> 01:16:02,260 And actually, I also looked at that problem in a later paper. 576 01:16:02,260 --> 01:16:03,650 I'm not going to say any more about that. 577 01:16:03,650 --> 01:16:13,650 I'm just going to stop here with the summary, just to point out that under specification is an important technique. 578 01:16:13,650 --> 01:16:21,500 With the goal of suppressing clutter so that generalisations are easier to see. 579 01:16:21,500 --> 01:16:23,960 The point here is that it's not a singular notion, 580 01:16:23,960 --> 01:16:32,540 the different types of specification or formalised in different ways and express different intuitions. 581 01:16:32,540 --> 01:16:40,130 The tree architecture and the formalism allows us to kind of articulate how all that works. 582 01:16:40,130 --> 01:16:46,280 And importantly in all of this is this notion of the minimal model that we can 583 01:16:46,280 --> 01:16:56,000 expect and examine and and compare some of the things that are not mentioned, 584 01:16:56,000 --> 01:17:06,560 we can actually conceptualise and examine later at a different point in the architecture. 585 01:17:06,560 --> 01:17:14,970 So with that, I will stop and I hope I haven't gone beyond the margins of. 586 01:17:14,970 --> 01:17:21,060 Of the time period, but thank you all. 587 01:17:21,060 --> 01:17:26,200 Thank you very much, Ron. Thank you, Ron. That was because a few minutes for questions. 588 01:17:26,200 --> 01:17:32,340 So who would like to start? Perhaps you want to stop sharing the screen run, 589 01:17:32,340 --> 01:17:38,460 so I have to stop sharing within an hour or two that you just click on the same box that we did last night. 590 01:17:38,460 --> 01:17:45,330 I just got to find the box now on the box. 591 01:17:45,330 --> 01:17:50,670 OK, so there's the box. I'm sharing. 592 01:17:50,670 --> 01:17:54,310 That do it. Yes. You're back. 593 01:17:54,310 --> 01:18:01,110 Yes, thank you so much. Susan Roesgen is in charge of leading the questions. 594 01:18:01,110 --> 01:18:12,360 Yes. There seems to be impressed silence in following a talk with great interest, 595 01:18:12,360 --> 01:18:19,830 and I've tried to compare it with how things are done in phonology that Janet has a question. 596 01:18:19,830 --> 01:18:30,500 Let's start with Janet. I ca I should not be answering the asking the first question, 597 01:18:30,500 --> 01:18:40,940 and we need to have at least a couple of remarks from Mary before I jump in and say anything about this, 598 01:18:40,940 --> 01:18:50,370 about just fine talk to which she contributed so much to. 599 01:18:50,370 --> 01:18:55,530 There you go again, Larry. I knew I had. OK, so, Ron, it's great to see you. 600 01:18:55,530 --> 01:19:09,460 I haven't seen you in a very long time. And so in the phonology and if I understand correctly, in semantics and syntax, the, um, 601 01:19:09,460 --> 01:19:16,800 uh, the default, um, uh, actually, there's a lot of contextual influences on the default, 602 01:19:16,800 --> 01:19:25,500 uh, related to how people have built up expectations from, uh, the structural position or what came before in syntax and semantics. 603 01:19:25,500 --> 01:19:38,390 That might even be the previous sentence. Uh, do you have any, uh, clues about how to handle that in this type of an approach? 604 01:19:38,390 --> 01:19:45,170 So I think so the question is about contextual information that influences the defaults. 605 01:19:45,170 --> 01:19:51,020 Yes, that's right. And so you need to, like, push that around somehow get right. 606 01:19:51,020 --> 01:20:04,370 So there are a couple of issues. One is the way history is imposed on what's happening in a particular utterance. 607 01:20:04,370 --> 01:20:13,250 So if we think of the particular. So. The if you think about it differently, 608 01:20:13,250 --> 01:20:22,070 there's a certain kind of delay that's built in or a deferral that's built into this notion of default, the intuition of a default is. 609 01:20:22,070 --> 01:20:30,800 Wait until you know everything that you're going to know and then if you haven't been told. 610 01:20:30,800 --> 01:20:38,660 Stick this in, and if you need to pass that off to somebody who cares and was waiting for you to tell them, 611 01:20:38,660 --> 01:20:46,530 you know, whether it was person 3rd or whether it was, you know, whatever property you're looking at. 612 01:20:46,530 --> 01:20:56,250 Well, the idea of the default is deferral until now, the question is, is it deferred within the syntax or as I say, 613 01:20:56,250 --> 01:21:04,110 the value scope with a very limited kind of value scope or larger value scope or within the sentence as a whole? 614 01:21:04,110 --> 01:21:10,850 Or does that scope get enlarged in some way? So it includes history as well. 615 01:21:10,850 --> 01:21:24,130 And that's what that trade off is. So how long or over what structures do you want to do the deferral before you make the commitment? 616 01:21:24,130 --> 01:21:32,680 And people also talk about the physical values that, you know, sticking to value, 617 01:21:32,680 --> 01:21:42,580 but if something else comes along and overrides it, then I can replace that value without getting a clash. 618 01:21:42,580 --> 01:21:50,580 This kind of set up here is trying to formalise what it means to be in that situation. 619 01:21:50,580 --> 01:22:01,040 And again, it comes to what is the scope of which this default bigger on this real. 620 01:22:01,040 --> 01:22:05,940 No, that's no good. 621 01:22:05,940 --> 01:22:15,500 You you know, but that's the intuition here, it's an intuition of deferral in the absence of information that may come later. 622 01:22:15,500 --> 01:22:19,520 How much later? Not too much later. Yeah. 623 01:22:19,520 --> 01:22:31,700 Yeah, that's too much. Yeah. For example, I mean, you could take these these type of things and say, well, just keep them in the description. 624 01:22:31,700 --> 01:22:37,200 Keep both models open. And then. 625 01:22:37,200 --> 01:22:42,090 The context comes along and picks and helps to pick out one of those models. 626 01:22:42,090 --> 01:22:57,300 So it's a very good question. Well, it fits in this framework, so when you add other stuff, it would tend to imply that you're. 627 01:22:57,300 --> 01:23:02,610 Mental representation of the context is basically using the same type of representation 628 01:23:02,610 --> 01:23:07,620 and pushing a lot of fragmented formalism is that it's not a corrective action, 629 01:23:07,620 --> 01:23:20,680 right? Exactly. Or, you know, some sort of mental representation of the original book we call the mental representation of grammatical relations. 630 01:23:20,680 --> 01:23:26,970 And the idea was our hypothesis from my work as a psychologist. 631 01:23:26,970 --> 01:23:40,070 And the pre-election history of this was that these notions of grammatical relation were the primitive. 632 01:23:40,070 --> 01:23:44,340 Cognitive. Notions. 633 01:23:44,340 --> 01:23:51,890 That's what we needed to have a formal way of modelling. 634 01:23:51,890 --> 01:24:02,500 So. You know, I think you know what's going on, you had a question and I've taken my hand down, I think I've been answered. 635 01:24:02,500 --> 01:24:17,400 I would like to come next. I want to make just another comment about if you see my comment is now that you are retiring. 636 01:24:17,400 --> 01:24:22,990 Maybe we can do more of this. That would be ideal chance. 637 01:24:22,990 --> 01:24:27,120 Yes. Would you like to say something? 638 01:24:27,120 --> 01:24:36,570 Yeah, I actually I mean, along those lines, I'm actually kind of along the lines of what you were just talking about with with Janet's question. 639 01:24:36,570 --> 01:24:45,450 I was kind of thinking about this. Now I forget what you call it, the local one that says that, you know, if you if you have a feature value, 640 01:24:45,450 --> 01:24:51,500 it has to be you have to find it within a certain local domain, you know, you would never find it at. 641 01:24:51,500 --> 01:25:00,510 Yes, so it seems to me as kind of an. We could probably do that, not just a formal distinction, 642 01:25:00,510 --> 01:25:07,290 that's kind of an interesting type of logical difference between different kinds of features and trying to figure out what that difference is. 643 01:25:07,290 --> 01:25:14,220 And I just wonder, I mean, is it morphologically expressed teachers versus teachers that aren't morphologically expressed? 644 01:25:14,220 --> 01:25:20,700 Is it features that are expressed on the head always or in that particular language or I mean, 645 01:25:20,700 --> 01:25:23,110 there's all sorts of things that could correlate with that. 646 01:25:23,110 --> 01:25:28,470 I just wondered if you had any thoughts about things that tend to correlate with that kind of stuff. 647 01:25:28,470 --> 01:25:33,370 Like, I have an intuition and I think probably a lot of people. 648 01:25:33,370 --> 01:25:38,320 I have an intuition about this kind of lexical scope. 649 01:25:38,320 --> 01:25:46,630 If you look at morphology, so obviously there's a whole morphological component and we know from the syntax, 650 01:25:46,630 --> 01:25:54,930 we can think of these four form items, but of course, it's all decomposed. We have some lexical rules and all that kind of stuff. 651 01:25:54,930 --> 01:26:03,260 But the lexical there's a notion that when you leave the lexicon, 652 01:26:03,260 --> 01:26:12,560 certain features that may be more difficult to exactly pin down where they had to be specified that, you know. 653 01:26:12,560 --> 01:26:16,510 When you're leaving the lexicon, they get pinned down. 654 01:26:16,510 --> 01:26:27,310 And people have had this intuition and it shows up in the literature, and I've always been troubled by it because from the point of view, 655 01:26:27,310 --> 01:26:33,530 infrastructure doesn't make any sense at all, this notion of lexical integrity. 656 01:26:33,530 --> 01:26:40,210 And, you know, I've always seen well, that's what you say when you want to argue against minimalism. 657 01:26:40,210 --> 01:26:50,200 But otherwise, it has no significance, just says that you can't have infl separated off someplace and syntax. 658 01:26:50,200 --> 01:27:04,760 It has to be in a word. So but that notion of lexical integrity maybe has to do with this notion of the values code. 659 01:27:04,760 --> 01:27:11,820 And that's that's my intuition about what's going on here and now, whether it's inside a noun phrase or, 660 01:27:11,820 --> 01:27:25,180 you know, higher up things, that may be a different set of issues. But it's kind of a way of relating, of controlling the source of information. 661 01:27:25,180 --> 01:27:31,580 And giving this kind of non monotonic way of making decisions in the middle of derivations. 662 01:27:31,580 --> 01:27:39,660 Well, you know, if it's something that I know from a very prepared theoretical. 663 01:27:39,660 --> 01:27:46,740 Feels right. And the question is, how would you formalise it, and that's what this is, how would you formalise? 664 01:27:46,740 --> 01:27:52,130 That feels right, Knowshon? 665 01:27:52,130 --> 01:27:59,480 Now, had we been in a real situation, we could have continued this for a long time and unfortunately, I see that people are leaving us. 666 01:27:59,480 --> 01:28:03,520 So that's a wrap. But yes, a. 667 01:28:03,520 --> 01:28:14,840 Yes, I just Mary, as as the faculty board chair this time, but also as a friend was right next door to you on the corridor, 668 01:28:14,840 --> 01:28:22,130 I just would like to end by saying thank you so much for whatever you've done for the faculty all along, 669 01:28:22,130 --> 01:28:31,040 not just for your students and your colleagues, but for the entire faculty as diverse as the U.S. and everything else that you've done for us. 670 01:28:31,040 --> 01:28:33,410 But I'm just going to miss you tremendously. 671 01:28:33,410 --> 01:28:40,490 But fortunately, I'll be leaving soon after so that we can both be at retirement and work more and and specification. 672 01:28:40,490 --> 01:28:45,530 And I can ask you about technology, as I used to before. 673 01:28:45,530 --> 01:28:52,610 So thank you very much from from the faculty. And this is what's coming. 674 01:28:52,610 --> 01:28:57,620 And it's been overwhelming and really, really delightful. Excellent. 675 01:28:57,620 --> 01:29:04,490 Well, we hope that we can meet in person again very soon so that we extend this longer. 676 01:29:04,490 --> 01:29:11,750 But, Ron, can you thank you also for being so very kind to join us today in this way. 677 01:29:11,750 --> 01:29:18,800 So I'm not going to ask anyone else who wants to speak because otherwise people will either miss their dinner or everything else. 678 01:29:18,800 --> 01:29:22,790 And so, as John Coleman says, more uproarious applause. 679 01:29:22,790 --> 01:29:30,588 Right. Thank you very much indeed.