LIN329 Markedness or faithfulnes

LIN329: Assignment 1

Due in hard copy at the beginning of lecture on Sept. 23

Part 0:

Make sure to include your name and student number!

You must write up the assignment on your own, but you can talk to other students about it. If you talk to other students, you MUST write their names on your assignment.

Part 1: Markedness or faithfulness? (12 marks)

In this part, I will define a constraint for you, and you need to:

(a) State whether it is a markedness or faithfulness constraint (1 mark each)

(b) Briefly (1 sentence) explain how you know whether it is a faithfulness or markedness constraint (1 mark each)

(c) Keeping in mind that constraints are meant to be universal and reflect universal tendencies in typology and naturalness, briefly (1 sentence) explain whether you think this is a plausible universal constraint (2 marks each)

1. *VoicelessSonorant: Assign a violation for any candidate with a voiceless sonorant in the output.

2. Contiguity: Assign a violation for every pair of adjacent segments in the input that is not adjacent in the output.

3. *VoicedNasal: Assign a violation for any candidate with a voiced nasal in the output.

Part 2: Finish the tableaux (18 marks)

In this part, you need to fill in the following in the provided tableaux:

(a) any missing violation marks, based on the given candidates (2 marks per tableau)

(b) any missing candidates, based on the given violations (1 mark per tableau)

· Assume all violations are already provided for any missing candidates

(c) fatal violation exclamation points (2 marks per tableau)

(d) the winner (1 mark per tableau)

Here are the definitions of the constraints used in this part:

· *Complex: Assign a violation for any complex onset/coda in the output (an onset or coda consisting of 2 or more consonants)

· Dep-IO: Assign a violation for every segment in the output that is not present in the input

· Max-IO: Assign a violation for every segment in the input that is not present in the output

· Onset: Assign a violation for every syllable in the output that lacks an onset

· *VoicedObstruent: Assign a violation for any voiced obstruents (stops/fricatives/affricates) in the output.

· Ident-IO[voice]: Assign a violation for any segment that changes in voicing between the input and output.

Question 1:

/bank/

*Complex

Dep-IO

Max-IO

bank

banak

*

Question 2:

/abanak/

Dep-IO

Onset

Max-IO

*

tabanak

banak

Question 3:

/bala/

Max-IO

*VoicedObstruent

Ident-IO[voice]

bala

*

pala

Part 3: Rank the constraints (15 marks)

In this part, you are given:

· a pair of constraints (all of which have been defined in Part 2)

· an input/UR

· an output/SF

You need to provide the relevant competing candidate, the ranking of the pair of constraints and a tableau illustrating. Your answer to each question should:

(a) State the relevant competing candidate, i.e. a candidate that solves the violation of the correct output by violating the other provided constraint (1 mark each)

(b) State the correct order of constraints in form. Constraint X >> Constraint Y (1 mark each)

(c) Complete the provided tableau, including the competing candidate, constraints (ranked properly), violation marks, and fatal violation marks (3 marks each)

HINT 1: It may help you to create tableaux for both orders of the constraints, to confirm which tableau gives the correct answer. (But please only provide your final correct tableau.)

HINT 2: If your candidates aren’t such that each candidate violates one of the two constraints and not the other, then your competing candidate is likely wrong.

Question 1:

· Constraints: Dep-IO and *Complex

· Input: /brip/

· Output: [brip]

Competing candidate: _________

Constraint ranking: __________ >> ____________

Tableau:

/brip/

Fbrip

Question 2:

· Constraints: Max-IO and *VoicedObstruent

· Input: /dimu/

· Output: [imu]

Competing candidate: _________

Constraint ranking: __________ >> ____________

Tableau:

/dimu/

Fimu

Question 3:

· Constraints: Onset and Dep-IO

· Input: /upo/

· Output: [tupo]

Competing candidate: _________

Constraint ranking: __________ >> ____________

Tableau:

/upo/

Ftupo

深度学习作为人工智能的关键分支,依托多层神经网络架构对高维数据进行模式识别与函数逼近,广泛应用于连续变量预测任务。在Python编程环境中,得益于TensorFlow、PyTorch等框架的成熟生态,研究者能够高效构建面向回归分析的神经网络模型。本资源库聚焦于通过循环神经网络及其优化变体解决时序预测问题,特别针对传统RNN在长程依赖建模中的梯度异常现象,引入具有门控机制的长短期记忆网络(LSTM)以增强序列建模能力。 实践案例涵盖从数据预处理到模型评估的全流程:首先对原始时序数据进行标准化处理与滑动窗口分割,随后构建包含嵌入层、双向LSTM层及全连接层的网络结构。在模型训练阶段,采用自适应矩估计优化器配合早停策略,通过损失函数曲线监测过拟合现象。性能评估不仅关注均方根误差等量化指标,还通过预测值与真实值的轨迹可视化进行定性分析。 资源包内部分为三个核心模块:其一是经过清洗的金融时序数据集,包含标准化后的股价波动记录;其二是模块化编程实现的模型构建、训练与验证流程;其三是基于Matplotlib实现的动态结果展示系统。所有代码均遵循面向对象设计原则,提供完整的类型注解与异常处理机制。 该实践项目揭示了深度神经网络在非线性回归任务中的优势:通过多层非线性变换,模型能够捕获数据中的高阶相互作用,而Dropout层与正则化技术的运用则保障了泛化能力。值得注意的是,当处理高频时序数据时,需特别注意序列平稳性检验与季节性分解等预处理步骤,这对预测精度具有决定性影响。 资源来源于网络分享,仅用于学习交流使用,请勿用于商业,如有侵权请联系我删除!
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值